US20200366936A1

Movatterモバイル変換

Info

Publication number: US20200366936A1
Application number: US16/985,572
Authority: US
Inventors: Tung Nguyen; Heiner Kirchhoffer; Detlev Marpe
Original assignee: GE Video Compression LLC
Current assignee: Dolby Video Compression LLC
Priority date: 2012-01-20
Filing date: 2020-08-05
Publication date: 2020-11-19
Also published as: MX379029B; US20210120271A1; DK2805419T3; CN107302363A; CN107302369A; HK1246053A1; IL292629B2; CN107302705B; KR20170100048A; IL265447B; KR102293126B1; US20180227597A1; US20190349604A1; AU2016204082A1; JP7227283B2; KR20170100050A; US10462487B2; KR20150004930A; RU2017145307A; UA124087C2

Abstract

An idea used herein is to use the same function for the dependency of the context and the dependency of the symbolization parameter on previously coded/decoded transform coefficients. Using the same function—with varying function parameter—may even be used with respect to different transform block sizes and/or frequency portions of the transform blocks in case of the transform coefficients being spatially arranged in transform blocks. A further variant of this idea is to use the same function for the dependency of a symbolization parameter on previously coded/decoded transform coefficients for different sizes of the current transform coefficient's transform block, different information component types of the current transform coefficient's transform block and/or different frequency portions the current transform coefficient is located within the transform block.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

The present application is a continuation of U.S. patent application Ser. No. 16/745,696 filed Jan. 17, 2020, which is a continuation of U.S. patent application Ser. No. 16/522,884 filed Jul. 26, 2019, now U.S. Pat. No. 10,582,219, which is a continuation of U.S. patent application Ser. No. 16/285,761 filed Feb. 26, 2019, now U.S. Pat. No. 10,462,487, which is a continuation of U.S. patent application Ser. No. 15/948,085 filed Apr. 9, 2018, now U.S. Pat. No. 10,271,068, which is a continuation of U.S. patent application Ser. No. 15/621,702 tiled Jun. 13, 2017, now U.S. Pat. No. 10,045,049, which is a continuation of U.S. patent application Ser. No. 14/335,439 filed Jul. 18, 2014, now U.S. Pat. No. 9,712,844, which is a continuation of International Application No. PCT/EP2013/051053, filed Jan. 21, 2013, and additionally claims priority from U.S. Application No. 61/588,846, filed Jan. 20, 2012, all of which are incorporated herein by reference in their entireties.

BACKGROUND OF THE INVENTION

The present invention is concerned with transform coefficient coding such as transform coefficients of a transform coefficient block of a picture.

In block-based image and/or video codecs, a picture or frame is coded in units of blocks. Among same, transform-based codecs subject blocks of the picture or frame to a transformation so as to obtain transform coefficient blocks. For example, the picture or frame may be predictively coded with a prediction residual being transform coded in units of blocks and then coding the resulting transform coefficient levels of the transform coefficients of these transform blocks using entropy coding.

In order to increase the efficiency of entropy coding, contexts are used in order to precisely estimate the probability of the symbols of the transform coefficient levels to be coded. However, in the recent years, the demands imposed onto picture and/or image codecs has increased. In addition to the luma and chroma components, codecs sometimes have to convey depth maps, transparity values and so forth. Moreover, the transform block sizes are variable within an increasingly large interval. Due to these varieties, codecs have an increasing number of different contexts with different functions for determining the context from already coded transform coefficients.

A different possibility of achieving high compression rates at a more moderate complexity, is adjusting a symbolization scheme to the coefficients' statistics as precise as possible. However, in order to perform this adaptation closely to the actual statistics, it is also mandatory to take various factors into account thereby necessitating a huge amount of differing symbolization schemes.

Accordingly, there is a need for keeping the complexity of transform coefficient coding low while nevertheless maintaining the possibility of achieving a high coding efficiency.

SUMMARY

According to an embodiment, an apparatus for decoding a plurality of transform coefficients having transform coefficient levels from a data stream may have: a context adaptive entropy decoder configured to, for a current transform coefficient, entropy decode a first set of one or more symbols from the data stream; a desymbolizer configured to map the first set of one or more symbols onto a transform coefficient level within a first level interval in accordance with a first symbolization scheme; an extractor configured to, if the transform coefficient level onto which the first set of one or more symbols is mapped in accordance with the first symbolization scheme is a maximum level of the first level interval, extract a second set of symbols from the data stream, wherein the desymbolizer is configured to map the second set of symbols onto a position within a second level interval in accordance with a second symbolization scheme which is parameterizable in accordance with a symbolization parameter, wherein the context adaptive entropy decoder is configured to, in entropy decoding at least one predetermined symbol of the first set of one or more symbols from the data stream, use a context depending, via a function parameterizable via a function parameter, with the function parameter set to a first setting, on previously decoded transform coefficients, and wherein the apparatus further includes a symbolization parameter determinator configured to, if the transform coefficient level onto which the first set of one or more symbols is mapped in accordance with the first symbolization scheme is a maximum level of the first level interval, determine the symbolization parameter depending, via the function with the function parameter set to a second setting, on the previously decoded transform coefficients.

Another embodiment may have a picture decoder including an inventive apparatus, wherein the picture decoder is configured to, in decoding a picture, retransform blocks of the picture from transform coefficient blocks, wherein the apparatus is configured to sequentially decode a plurality of transform coefficients of the transform coefficient blocks, transform coefficient block by transform coefficient block, with using the function for transform coefficient blocks of different sizes, for transform coefficient blocks of different sizes, and/or for transform coefficient blocks of different information component type.

Another embodiment may have a picture encoder including an inventive apparatus, wherein the picture encoder is configured to, in encoding a picture, transform blocks of the picture into transform coefficient blocks, wherein the apparatus is configured to code a plurality of transform coefficients of the transform coefficient blocks, transform coefficient block by transform coefficient block, with using the function for blocks of different sizes.

According to another embodiment, a method for coding a plurality of transform coefficients having transform coefficient levels into a data stream may have the steps of: symbolization mapping a current transform coefficient onto a first set of one or more symbols in. accordance with a first symbolization scheme, if the current transform coefficient's transform coefficient level is within a first level interval, and if the current transform coefficient's transform coefficient level is within a second level interval, onto a combination of a second set of symbols onto which a maximum level of the first level interval is mapped in accordance with the first symbolization scheme, and a third set of symbols depending on a position of the current transform coefficient's transform coefficient level within the second level interval, in accordance with a second symbolization scheme which is parameterizable in accordance with a symbolization parameter; context adaptive entropy encoding including, if the current transform coefficient's transform coefficient level is within the first level interval, entropy encoding the first set of one or more symbols into the data stream, and, if the current transform coefficient's transform coefficient level is within the second level interval, entropy encoding the second set of one or more symbols into the data stream, wherein the context adaptive entropy encoding involves, in entropy encoding at least one predetermined symbol of the second set of one or more symbols into the data stream, using a context depending, via a function parameterizable via a function parameter, with the function parameter set to a first setting, on previously coded transform coefficients; and if the current transform coefficient's transform coefficient level is within the second level interval, determining the symbolization parameter for the mapping onto the third set of symbols depending, via the function with the function parameter set to a second setting, on the previously coded transform coefficients; and if the current transform coefficient's transform coefficient level is within the second level interval, inserting the third set of symbols into the data stream.

Another embodiment may have a computer program having a program code for performing, when running on a computer, an inventive method.

An idea of the present invention is to use the same function for the dependency of the context and the dependency of the symbolization parameter on previously coded/decoded transform coefficients. Using the same function—with varying function parameter—may even be used with respect to different transform block sizes and/or frequency portions of the transform blocks in case of the transform coefficients being spatially arranged in transform blocks. A further variant of this idea is to use the same function for the dependency of a symbolization parameter on previously coded/decoded transform coefficients for different sizes of the current transform coefficient's transform block, different information component types of the current transform coefficient's transform block and/or different frequency portions the current transform coefficient is located within the transform block.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the present invention will be detailed subsequently referring to the appended drawings, in which:

FIG. 1 shows a schematic diagram of a transform coefficient block comprising transform coefficients to be coded and illustrates the co-use of a parametrizable function for context selection and symbolization parameter determination in accordance with an embodiment of the preset invention;

FIG. 2 shows a schematic diagram of symbolization concept for transform coefficient levels using two different schemes within two level intervals;

FIG. 3 shows a schematic graph of two appearance probability curves defined over possible transform coefficient levels for two different contexts;

FIG. 4 shows a block diagram of an apparatus for coding a plurality of transform coefficients in accordance with an embodiment;

FIGS. 5aand 5bshow schematic diagrams of a structure for the data stream resulting in accordance with different embodiments;

FIG. 6 shows a block diagram of a picture encoder in accordance with an embodiment;

FIG. 7 shows a block diagram of an apparatus for decoding a plurality of transform coefficients in accordance with an embodiment;

FIG. 8 shows a block diagram of a picture decoder in accordance with an embodiment;

FIG. 9 shows a schematic diagram of a transform coefficient block so as to illustrate a scan and a template in accordance with an embodiment.

FIG. 10 shows a block diagram of an apparatus for decoding a plurality of transform coefficients in accordance with a further embodiment;

FIGS. 11aand 11bshows schematic diagrams of symbolization concepts for transform coefficient levels combining two or three different schemes within partial intervals of the whole interval range;

FIG. 12 shows a block diagram of an apparatus for coding a plurality of transform coefficients in accordance with a further embodiment; and

FIG. 13 shows a schematic diagram of a transform coefficient block so as to illustrate, in accordance with a further embodiment, a scan order among the transform coefficient blocks following a sub-block order defined among sub-blocks into which the transform coefficient block is subdivided in order to illustrate another embodiment for designing the parametrizable function for context selection and symbolization parameter determination.

DETAILED DESCRIPTION OF THE INVENTION

With respect to the description below, it is noted that the same reference sign is used in these figures for elements occurring in more than one of these figures. Accordingly, the description of such an element with respect to one figure shall equally apply to the description of another figure in which this element occurs.

Moreover, the description brought forward below preliminarily assumes the transform coefficients to be coded as being two-dimensionally arranged so as to form a transform block such as a transform block of a picture. However, the present application is not restricted to image and/or video coding. Rather, the transform coefficients to be coded could, alternatively, be transform coefficients of a one-dimensional transform such as used, for example, in audio coding or the like.

In order to explain the problems that the embodiments described further below face, and the way the embodiments further described below overcome these problems, reference is preliminarily made toFIGS. 1 to 3, which show an example of transform coefficients of a transform block and their general way of entropy coding, which is then improved by the subsequently explained embodiments.

FIG. 1 exemplarily shows ablock10 oftransform coefficients12. In the present embodiment, the transform coefficients are two-dimensionally arranged. In particular, same are exemplarily shown as being regularly arranged in columns and rows although another two-dimensional arrangement is also possible. The transform which led to thetransform coefficients12 or transformblock10 may be a DCT or some other transform which decomposes a (transform) block of a picture, for example, or some other block of spatially arranged values into components of different spatial frequency. In the present example ofFIG. 1, thetransform coefficients12 are two-dimensionally arranged in columns i and rows j so as to correspond to frequency pairs (f_x(i), f_y(j)) of frequencies f_x(i), f_y(j) measured along different spatial directions x,y such as directions perpendicular to each other, where f_x/y(i)<f_x/y(i+1) and (i,j) is the position of the respective coefficient intransform block10.

Often thetransform coefficients12 corresponding to lower frequencies have higher transform coefficient levels compared to transform coefficients corresponding to higher frequencies. Accordingly, often many of the transform coefficients near the highest frequency component of thetransform block10 are quantized to zero and may not have to be coded. Rather, ascan order14 may be defined among thetransform coefficients12 which one-dimensionally arranges the two-dimensionally arranged transform coefficients12 (i,j) into a sequence of coefficients at an order, i.e. (i,j)□k, so that it is likely that the transform coefficient levels have a tendency of monotonically decreasing along this order, i.e. it is likely that coefficient level of coefficient k is greater than coefficient level of coefficient k+1.

For example, a zigzag or a raster scan may he defined among thetransform coefficients12. According to the scan, theblock10 may be scanned in diagonals from, for example, the DC component transform coefficient (upper left-hand coefficient) to the highest frequency transform coefficient (lower right-hand coefficient) or vice versa. Alternatively, a row-wise or column-wise scan of the transform coefficients between the just mentioned extreme component transform coefficients may be used.

As described further below, in coding the transform block the position of the last non-zero transform coefficient L inscan order14 may be coded into the data stream first, with then merely coding the transform coefficients from the DC transform coefficient alongscan path14 to the last non-zero transform coefficient L—optionally in that direction or in counter direction.

In any case, in order to code the transform coefficient levels of thetransform coefficients12, different symbolization schemes are used in order to cover different portions or

intervals

16,18 of therange interval20. To be more precise, transform coefficient levels within afirst level interval16, except for the ones equal to a maximum level of thefirst level interval16, may simply be symbolized onto a set of one or more symbols in accordance with a first symbolization scheme. Transform coefficient levels, however, lying within thesecond level interval18, are mapped onto a combination of symbol sets of the first and second symbolization schemes. As will be noted later, third and further intervals may follow the second interval accordingly.

As shown inFIG. 2, thesecond level interval18 lies above thefirst level interval16 but overlaps with the latter at the maximum level of thefirst level interval16, which is 2 in the example ofFIG. 2. For transform coefficient levels lying within thesecond level interval18, the respective level is mapped onto a combination of the first symbol set corresponding to the first level interval's maximum level in accordance with the first symbolization scheme, and a second symbol set depending on a position of the transform coefficient level within thesecond level interval18 in accordance with the second symbolization scheme.

In other words, thefirst symbolization scheme16 maps the levels covered by thefirst level interval16 onto a set of first symbol sequences. Please note that the length of the symbol sequences within the set of symbol sequences of the first symbolization scheme may even be merely one binary symbol in case of a binary alphabet and in case of thefirst level interval16 merely covering two transform coefficient levels such as 0 and 1. In accordance with an embodiment of the present application, the first symbolization scheme is a truncated unary binarization of levels ininterval16. In case of a binary alphabet, the symbols may be called bins.

As will be described in more detail below, the second symbolization scheme maps the levels within thesecond level interval18 onto a set of second symbol sequences of varying length wherein the second symbolization scheme is parameterizable in accordance with a symbolization parameter. The second symbolization scheme may map the levels withininterval18, i.e. x—the maximum level of the first interval, onto a Rice code having a Rice parameter.

In particular, thesecond symbolization scheme18 may be configured such that the symbolization parameter varies a rate at which a length of the second scheme's symbol sequences increases from the lower bound of thesecond level interval18 to an upper bound thereof. Obviously, an increased length of the symbol sequences consumes more data rate within the data stream into which the transform coefficients are to be coded. Generally, it is advantageous if the length of the symbol sequence onto which a certain level is mapped correlates with the actual probability at which the transform coefficient level to be currently coded assumes the respective level. Naturally, the latter statement is also valid for the levels outside thesecond level interval18 within thefirst level interval16 or for the first symbolization scheme in general.

In accordance with the embodiments described below, the symbols of the symbol sequences of thefirst symbolization scheme16 are entropy coded in a context adaptive way. That is, a context is associated with the symbols, and the alphabet probability distribution associated with the selected context is used for entropy coding the respective symbol. The symbols of the symbol sequences of the second symbolization scheme are inserted into the data stream directly or using a fixed alphabet probability distribution such an equal probability distribution according to which all members of the alphabet are equally probable.

Contexts used in entropy coding the symbols of the first symbolization scheme have to be selected appropriately so as to allow for a good adaptation of the estimated alphabet probability distribution to the actual alphabet statistics. That is, the entropy coding scheme may be configured to update a current estimate of the context's alphabet probability distribution whenever a symbol having this context is encoded/decoded, thereby approximating the actual alphabet statistics. The approximation is faster if the contexts are chosen appropriately, that is fine enough, but not with too many different contexts so as to avoid a too infrequent association of symbols with certain contexts.

Likewise, the symbolization parameter for a coefficient should be chosen dependent on the previously coded/decoded coefficients so as to approximate the actual alphabet statistics as close as possible. Too fine diversification is not a critical issue here, because the symbolization parameter is directly determined from the previously coded/decoded coefficients, but the determination should closely correspond to the correlation of the dependency of the probability curve within thesecond interval18 on the previously coded/decoded coefficients.

As will be described in more detail below, the embodiments for coding transform coefficients further described below are advantageous in that a common function is used in order to achieve the context adaptivity and the symbolization parameter determination. Choosing the correct context is, as outlined above, important in order to achieve a high coding efficiency or compression rate, and the same applies with respect to the symbolization parameter. The embodiments described below allow for achieving this aim by keeping the overhead for instantiating the dependency on previously coded/decoded coefficients low. In particular, the inventors of the present application found a way of finding a good compromise between realizing efficient dependency on previously coded/decoded coefficients on the one hand and reducing the number of proprietary logic for instantiating the individual context dependencies on the other hand.

FIG. 4 shows an apparatus for coding a plurality of transform coefficients having transform coefficient levels into a data stream in accordance with an embodiment of the present invention. It is noted that in the following description, the symbol alphabet is often assumed to be a binary alphabet although this assumption is, as outlined above, not critical for the present invention and accordingly, all of these explanations shall be interpreted as being also illustrative for an extension onto other symbol alphabets.

The apparatus ofFIG. 4 is for coding a plurality of transform coefficients entering at aninput30 into adata stream32. The apparatus comprises asymbolizer34, a contextadaptive entropy encoder36, asymbolization parameter determinator38 and aninserter40.

Thesymbolizer34 has two outputs, namely one for symbol sequences of the first symbolization scheme, and another for the symbol sequences of the second symbolization scheme. Theinserter40 has an input for receiving the second symbolization scheme'ssymbol sequences42 and the contextadaptive entropy encoder36 has an input for receiving the first symbolization scheme'ssymbol sequences44. Further, thesymbolizer34 has a parameter input for receiving thesymbolization parameter46 from an output ofsymbolization parameter determinator38.

The contextadaptive entropy encoder36 is configured to entropy encode the symbol of thefirst symbol sequences44 into thedata stream32. Theinserter40 is configured to insert thesymbol sequences42 intodata stream32.

Generally speaking, bothentropy encoder36 andinserter40 sequentially scan the transform coefficients. Obviously, inserter40 merely operates for transform coefficients, the transform coefficient level of which lies within thesecond level interval18. However, as will be described in more detail below, there are different possibilities for defining the order between the operation of theentropy encoder36 and theinserter40. In accordance with a first embodiment, the coding apparatus ofFIG. 4 is configured to scan the transform coefficients in one single scan so thatinserter40 inserts thesymbol sequence42 of a transform coefficient into thedata stream32 subsequent to the entropy encoder's entropy encoding of thefirst symbol sequence44 relating to the same transform coefficient into thedata stream32 and prior to the entropy encoder's entropy encoding thesymbol sequence44 relating to the next transform coefficient in line into thedata stream32.

In accordance with an alternative embodiment, the apparatus uses two scans, wherein within the first scan the contextadaptive entropy encoder36 sequentially encodes thesymbol sequences44 into thedata stream32 for each transform coefficient withinserter40 then. inserting thesymbol sequences42 for those transform coefficients the transform coefficient level of which lies within thesecond level interval18. There could even be more sophisticated schemes according to which, for example, the contextadaptive entropy encoder36 uses several scans in order to encode the individual symbols of thefirst symbol sequences44 into thedata stream32 such as the first symbol or bin in a first scan, followed by a second symbol or bin of thesequences44 in a second scan and so forth.

As already indicated above, the contextadaptive entropy encoder36 is configured to entropy encode at least one predetermined symbol of thesymbol sequences44 into thedata stream32 in a context adaptive way. For example, the context adaptivity could be used for all the symbols of thesymbol sequences44. Alternatively, contextadaptive entropy encoder36 may restrict the context adaptivity to the symbols at the first position and the symbol sequences of the first symbolization scheme only, or the first and second, or the first to third positions and so forth.

As described above, for context adaptivity,encoder36 manages contexts by storing and updating an alphabet probability distribution estimate for each context. Each time a symbol of a certain context is encoded, the currently stored alphabet probability distribution estimate is updated using the actual value of this symbol thereby approximating the symbols' actual alphabet statistics of that context.

Likewise,symbolization parameter determinator38 is configured to determine thesymbolization parameter46 for the second symbolization scheme and itssymbol sequences42 depending on previously coded transform coefficients.

To be more precise, the contextadaptive entropy encoder36 is configured such that same uses, or select, for the current transform coefficient a context depending, via a function parameterizable via a function parameter, and with the function parameter set to a first setting, on previously coded transform coefficients, while thesymbol parameter determinator38 is configured to determine thesymbolization parameter46 depending, via the same function, and with the function parameter set to a second setting, on the previously coded transform coefficients. The settings may differ, but nevertheless, assymbolization parameter determinator38 and contextadaptive entropy encoder36 use the same function, logic overhead may be reduced. Merely the function parameter may differ between the context selection of theentropy encoder36 on the one hand and the symbolization parameter determination of thesymbolization parameter determinator38 on the other hand.

As far as the dependency on the previously coded transform coefficients is concerned, it should be noted that this dependency is restricted to the extent to which these previously coded transform coefficients have already been coded intodata stream32. Imagine, for example, that such a previously encoded transform coefficient lies within thesecond level interval18, but thesymbol sequence42 thereof has not yet been inserted intodata stream32. In that case,symbolization parameter determinator38 and contextadaptive entropy encoder36 merely know from thefirst symbol sequence44 of that previously coded transform coefficient that same lies within thesecond level interval18. In that case, the maximum level of thefirst level interval16 may serve as a representative for this previously coded transform coefficient. Insofar, the dependency “on the previously coded transform coefficients” shall be understood in a broad way so as to encompass a dependency on “information on other transform coefficients previously encoded/inserted into thedata stream32”. Further, transform coefficients lying “beyond” the last non-zero coefficient L position may be inferred to be zero.

In order to finalize the description ofFIG. 4, the outputs ofentropy encoder36 andinserter40 are shown to be connected to acommon output48 of the apparatus via a switch50, with the same connectivity existing between inputs for previously inserted/coded information ofsymbolization parameter determinator38 and contextadaptive entropy encoder36 on the one hand and the outputs ofentropy encoder36 andinserter40 on the other hand. Switch50 connectsoutput48 with either one of the outputs ofentropy encoder36 andinserter40 in the order mentioned above with respect to the various possibilities of using one, two or more scans for coding the transform coefficients.

symbol sequences

42 and44 of transform coefficients on one diagonal of thescan140 may be coded in parallel since none of the transform coefficients on a diagonal falls into thetemplate56 of another transform coefficient in the same diagonal. Naturally, similar templates may be found for row- and column-wise scans.

In order to provide more specific examples for the commonly used function g(f(x)) and the corresponding function parameters, in the following such examples are provided using respective formulae. In particular, the apparatus ofFIG. 4 may be configured such that thefunction52 defining the relationship between a set x of previously coded transform coefficients on the one hand, and acontext index number54 indexing the context and thesymbolization parameter46 on the other hand, may be

g (f (x)) where g (x) = \sum_{i = 1}^{d_{f}} δ^{1} (x, n_{1}) | and f (x) = \sum_{i = 1}^{d} w_{i} \cdot h \cdot δ (x_{i}, t)

with

δ (x, t) = {\begin{matrix} 1 \langle x \rangle \geq t \\ 0 \langle x \rangle < t \end{matrix} and δ^{'} (x, n) = {\begin{matrix} 1 x > n \\ 0 x \leq n \end{matrix}

where

t and {x_y₂, . . . , n_d_ff}=n and, optionally w_i, from the function parameter,
x={x₁, . . . , x_d} with x_iwith i∈{1 . . . d} representing a previously decoded transform coefficient i, w_iare weighting values each of which may be equal to one or unequal to one, and h is a constant or function of x_i.

It follows that g(f(x)) lies within [0,d_f]. If g(f(x)) is used to define an context index offset number ctx_offsetwhich is summed-up along with at least one base context index offset number ctx_base, then the value range of resulting context index ctx=ctx_base+ctx_offsetis [ctx_base; ctx_base+d_f]. Whenever it is mentioned that differing sets of contexts are used to entropy code symbols ofsymbol sequences44, then ctx_baseis chosen differently such that [ctx_base,1ctx_base+d_f] does not overlap [ctx_base,2; ctx_base+d_f]. This is, for example, true for

- transform coefficients belonging to transform blocks of differing size;
- transform coefficients belonging to transform blocks of differing information component type such as depth, luma, chroma and so forth;
- transform coefficients belonging to differing frequency portions of the same transform block;

As mentioned before, the symbolization parameter may be a Rice parameter k. That is, (absolute) levels withininterval16, i.e. X, with X+M=x (where M is the maximum level ofinterval16 and x is the (absolute)transform coefficient level) would be mapped onto a bin string having a prefix and a suffix, the prefix being a unary code of [X·2^−k], and the suffix being a binary code of the remainder of [X·2^−k].

d_fmay also form part of the function parameter, d may also form part of the function parameter.

A difference in function parameter such as between context selection and symbolization parameter determination necessitates merely one difference in either t, {n₁, . . . , n_d_f}=n, d_f(if forming part of the function parameter), or d (if forming part of the function parameter).

As far as the symbolization parameter determinator is concerned, same may be configured such that, in the determination of the symbolization parameter, x_iis equal to the transform coefficient level of the previously coded transform coefficient i, independent from the transform coefficient level of the previously coded transform coefficient i being within the first or second level interval.

The apparatus may be further configured such that n₁≤≤n_d_fapplies in any case.
The apparatus may also be configured such that h=|x_i|−t.

In a further embodiment the apparatus may be configured to spatially determine the previously coded transform coefficients depending on a relative spatial arrangement of the transform coefficients relative to the current transform coefficient, i.e. based on a template around the current transform coefficient's position.

The apparatus may be further configured to determine a position of a last non-zero transform coefficient L among transform coefficients of atransform coefficient block10 along apredetermined scan order14, and to insert information on the position into thedata stream32, wherein the plurality of transform coefficients encompasses the transform coefficients from the last non-zero transform coefficient L to a beginning of the predetermined scan order, i.e. a DC component transform coefficient.

In a further embodiment, thesymbolizer34 may configured to use a modified first symbolization scheme for symbolization of the last transform coefficient L. According to the modified first symbolization scheme, merely non-zero transform coefficient levels within thefirst level interval16 may be mapped, while a zero level is presumed not to apply for the last transform coefficient L. For example, the first bin of the truncated unary binarization may be suppressed for coefficient L.

The context adaptive entropy encoder may be configured to use a separate set of contexts for entropy encoding the first set of one or more symbols for the last non-zero transform coefficient, separate from contexts used in entropy encoding the first set of one or more symbols of other than the last non-zero transform coefficient.

The context adaptive entropy encoder may traverse the plurality of transform coefficients in an opposite scan order leading from the last non-zero transform coefficient to the DC transform coefficient of the transform coefficient block. This may or may not also apply for thesecond symbol sequences42.

The apparatus may also be configured to code the plurality of transform coefficients into thedata stream32 in two scans, wherein the contextadaptive entropy coder36 may be configured to entropy encode thefirst symbols sequences44 for the transform coefficients into thedata stream32 in an order corresponding to a first scan of the transform coefficients, wherein theinserter40 is configured to subsequently insert thesymbol sequences42 for the transform coefficients having a transform coefficient level within thesecond level interval18 into thedata stream32 in an order corresponding to an occurrence of the transform coefficients having a transform coefficient level within thesecond level interval18 within a second scan of the transform coefficients. An example for a resultingdata stream32 is shown inFIG. 5a:it may comprise, optionally, ininformation57 on the position of L, followed by thesymbol sequences42 in entropy encoded form (at least some in context adaptive entropy encoded form) and further followed by thesymbol sequences44 inserted directly or using, for example, bypass mode (equal probable alphabet).

In a further embodiment, the apparatus may be configured to code the plurality of transform coefficients into the data stream23 sequentially in one scan, wherein the contextadaptive entropy encoder36 and theinserter40 are configured to, for each transform coefficient in a scan order of the one scan, insert thesymbol sequences42 of respective transform coefficients having a transform coefficient level within thesecond level interval18 into thedata stream32 immediately subsequent to the context adaptive entropy coder's entropy encoding of thesymbol sequence44 into thedata stream32, along with which same form the combination onto which same transform coefficients are mapped, so that thesymbol sequences42 are interspersed into thedata stream32 betweensymbol sequences44 of the transform coefficients. The result is illustrated inFIG. 5b.

Theinserter40 may be configured to insert thesymbol sequences42 into the data stream directly or using entropy encoding using a fixed probability distribution. The first symbolization scheme may be a truncated unary binarization scheme. The second symbolization scheme may be such that thesymbol sequences42 are of a Rice code.

As already noted above the embodiments ofFIG. 4 may be implemented within an image/video coder. An example of such an image/video coder or picture coder is shown in6. The picture encoder is generally indicated atreference sign60 and comprises anapparatus62 corresponding to the one shown inFIG. 4, for example. Theencoder60 is configured to, in encoding apicture64, transform blocks66 of thepicture64 into transform coefficient blocks10 which are then treated byapparatus62 so as to code, pertransform block10, a plurality of transform coefficients thereof. In particular,apparatus62 processes transformblocks10 transform block by transform block. In doing so,apparatus62 may usefunction52 forblocks10 of different sizes. For example, an hierarchical multi-tree subdivision may be used in order to decomposepicture64 or tree-root blocks thereof, intoblocks66 of different sizes. The transform blocks10 resulting from applying a transform to theseblocks66 are, accordingly, also of different size and although, accordingly, function52 may be optimized for the different block sizes by way of using different function parameters, the overall overhead for providing such different dependencies for the symbolization parameter on the one hand and the context index on the other hand is kept low.

FIG. 7 shows an apparatus for decoding a plurality of transform coefficients having transform coefficient levels from adata stream32 which fits to the apparatus outlined above with respect toFIG. 4. In particular, the apparatus of 7 comprises a contextadaptive entropy decoder80, adesymbolizer82 and anextractor84 as well as asymbolization parameter determinator86. The contextadaptive entropy decoder80 is configured to, for a current transform coefficient, entropy decode a first set of one or more symbols, i.e.symbol sequence44, from thedata stream32.Desymbolizer82 is configured to map the first set of one or more symbols, i.e. thesymbol sequence44, onto a transform coefficient level within thefirst level interval16 in accordance with a first symbolization scheme. To be more precise, contextadaptive entropy decoder80 anddesymbolizer82 operate in an interactive manner.Desymbolizer82 informs contextadaptive entropy decoder80 via asignal88 at which symbol sequentially decoded bydecoder80 from data stream32 a valid symbol sequence of the first symbolization scheme has been finalized.

Extractor

84 is configured to, if the transform coefficient level onto which the first set of one or more symbols, i.e.symbol sequence44, is mapped in accordance with the first symbolization scheme, is the maximum level of thefirst level interval16, extract a second set of symbols, i.e.symbol sequence42 fromdata stream32. Again,desymbolizer82 andextractor84 may operate in concert. That is,desymbolizer82 may informextractor84 by asignal90 when a valid symbol sequence of the second symbolization scheme has been finalized whereuponextractor84 may finish the extraction ofsymbol sequence42.

Thedesymbolizer82 is configured to map the second set of symbols, i.e.symbol sequence42, onto a position within thesecond level interval18 in accordance with the second symbolization scheme which, as already noted above, is parameterizable in accordance with thesymbolization parameter46.

The contextadaptive entropy decoder80 is configured to, in entropy decoding at least one predetermined symbol of thefirst symbol sequence44, use a context depending, viafunction52, on previously decoded transform coefficients. Thesymbolization parameter determinator86 is configured to, if the transform coefficient level onto which thefirst symbol sequence44 is mapped in accordance with the first symbolization scheme is the maximum level of thefirst level interval16, determine thesymbolization parameter46 depending, viafunction52, on the previously decoded transform coefficients. To this end, inputs ofentropy decoder80 andsymbolization parameter determinator86 are connected via aswitch92 to an output ofdesymbolizer82 at which desymbolizer82 outputs values x_iof the transform coefficients.

As described above, for context adaptivity,decoder80 manages contexts by storing and updating an alphabet probability distribution estimate for each context. Each time a symbol of a certain context is decoded, the currently stored alphabet probability distribution estimate is updated using the actual/decoded value of this symbol thereby approximating the symbols' actual alphabet statistics of that context.

Likewise,symbolization parameter determinator86 is configured to determine thesymbolization parameter46 for the second symbolization scheme and itssymbol sequences42 depending on previously decoded transform coefficients.

Generally, all the possible modifications and further details described above with respect to the encoding are also transferable onto the apparatus for decoding ofFIG. 7.

FIG. 8 shows as a pendant toFIG. 6. That is, the apparatus ofFIG. 7 may be implemented within thepicture decoder100. Thepicture decoder100 ofFIG. 7 comprises an. apparatus according toFIG. 7, namelyapparatus102, Thepicture decoder100 is configured to, in decoding or reconstructing apicture104, retransform blocks106 ofpicture104 from transform coefficient blocks10 the plurality of transform coefficients of whichapparatus102 decodes from thedata stream32 which, in turn, enterspicture decoder100. In particular,apparatus102 processes transformblocks10 block by block and may, as already denoted above,use function52 commonly forblocks106 of different sizes.

It should be noted that picture encoder and

decoder

60 and100, respectively, may be configured to use predictive coding with applying the transform/retransform to the prediction residual. Moreover, thedata stream32 may have subdivision information encoded therein, which signals topicture decoder100 the subdivision into the blocks individually subject to transformation,

Below, the above embodiments are again described in some other words, and with providing more details on specific aspects which details may individually transferred onto the above embodiments. That is, above embodiments related to a specific way of context modeling for the coding of syntax elements related to transform coefficients such as in block-based image and video coders, and aspects thereof are described and highlighted further below.

The embodiments may relate to the field of digital signal processing and, in particular, to a method and apparatus for image and video decoders and encoders. In particular, the coding of transform coefficients and their associated syntax elements in block-based image and video codecs may be performed in accordance with the embodiments described. In so far, some embodiments represented an improved context modeling for the coding of syntax elements related to transform coefficients with an entropy coder that employs a probability modeling. Further, the derivation of a Rice parameter that is used for the adaptive binarization of the remaining absolute transform coefficients may be done as described above with respect to the symbolization parameter. Unification, simplification, parallel processing friendly, and moderate memory usage in terms of context memory are the benefits of the embodiments compared to straight forward context modeling.

In even other words, embodiments of the present invention may reveal a new approach for context model selection of syntax elements related to the coding of transform coefficients in block-based image and video coders. Further, derivation rules for a symbolization parameter, such as a Rice parameter, that controls the binarization of a remaining value of an absolute transform coefficients have been described. Essentially, the above embodiments used a simple and common set of rules for the context model selection for all or for a part of syntax elements related to the coding of the transform coefficients.

The first symbolization scheme mentioned above may be a truncated unary binarization. If so, coeff_significant_flag, coeff_abs_greater_1, and coeff_abs_greater_2 may be called the binary syntax elements or symbols which form the first, the second, and the third bin resulting from the truncated unary binarization of a transform coefficient. As described above, the truncated unary binarization may merely represent a prefix, which may be accompanied by the suffix being itself a Rice code in case of the transform coefficient's level falling within thesecond level interval18. A further suffix may be of a Exp-Golomb code such as of 0-order, thereby forming a further level interval following the first and

second intervals

16 and18 inFIG. 2 (not shown inFIG. 2).

The derivation of the Rice parameter for the adaptive binarization of the remaining absolute transform coefficient may be done, as described above, based on the same set ofrules52 as used for the context model selection.

With respect to the scan order, it is noted that same may be varied compared to the above description. Moreover, different block sizes and shapes may be supported by the apparatuses ofFIGS. 4 and 6, with using, however, the same set of rules, i.e. with using thesame function52. Accordingly, a unified and simplified scheme for the context model selection of syntax elements related to the coding of transform coefficients combined with a haononization for the derivation of symbolization parameter may be achieved. Thus, the context model selection and symbolization parameter derivation may use the same logic which may be hardwired, programmed hardware or a software-subroutine, for example.

To achieve a common and simple scheme for context model selection and derivation of the symbolization parameter, such as Rice parameter, already coded transform coefficients of a block or a shape may be evaluated as described above. In order to evaluate the already coded transform coefficients, the separation in coding of coeff_significant_flag, which is the first bin resulting from the binarization (which could be referred to as the coding of the significance map), and the remaining absolute value of the transform coefficient level is performed using acommon function52.

The coding of the sign information may be done in an interleaved manner, i.e. by coding the sign directly after the coding of the absolute transform coefficient. Thus, the whole transform coefficients would be coded in one scan pass only. Alternatively, the sign information can be coded in a separate scanning path as long as the evaluation values f(x) rely on absolute level information only.

As denoted above, the transform coefficients may be coded in a single scan pass or in multiple scan passes. This may be enabled by, or described by, a cutoff set c the coefficients c_iof which indicate the number of symbols of the transform coefficient's (first and second) symbolization processed in scan i. In the case of an empty cutoff set, one scan would be used. In order to have improved results for the context model selection and the derivation of the symbolization parameter, the first cutoff parameter c₀of the cutoff set c should be larger than one.

Note that cutoff set c may be chosen to be c={c₀; c₁} with c₀=1 and c₁=3 and |c|=2, where co indicates the number of bins/symbols of the first binarization, encompassed in the first scan, and c₁=3 indicating the symbol position within the first binarization up to which symbols of the first binarization are covered be the second scan. Another example is given when the scheme codes first bin resulting from the binarization for a whole block or shape in a first scan pass, next the second bin for the whole block or shape in a second scan pass, with c₀equal to one, c₁equal to two, and so on.

Thelocal template56 for the coding of coeff_significant_flag, i.e. the first bin from the binarization process, may be designed as shown inFIG. 1 or as shown inFIG. 9. As a unification and simplification, thelocal template56 may be used for all block sizes and shapes. Instead of evaluating the number of neighbors with transform coefficient inequality to zero only, the whole transform coefficients are input intofunction52 in form of x_i. Note that thelocal template56 can be fixed, i.e. independent from the current transform coefficient's position or scan index and independent from the previously coded transform coefficients, or adaptive, i.e, dependent from the current transform coefficient's position or scan index and/or the previously coded transform coefficients, and the size can be fixed or adaptive. Further, when the template size and shape is adjusted allowing the coverage of all scan positions of a block or a shape, all already coded transform coefficients or all already coded transform coefficients up to a specific limit are used for the evaluation process.

As an example,FIG. 9 shows another example for thelocal template56 which may be used for an 8×8transform block10 withdiagonal scan14. L denotes the last significant scan position and the scan positions marked with an x denotes the current scan position. Note that for other scan orders, the local template may be modified to fit thescan order14. For example, in case of a forward diagonal scan, thelocal template56 may be flipped along the diagonals.

The context model selection and symbolization parameter derivation may be based on different evaluation values f(x) resulting from the evaluation of already coded neighbors x_i. This evaluation is done for all scan positions having already coded neighbors covered by thelocal template56. Thelocal template56 has a variable or fixed size and may depend on the scan order. However, the template shape and size is an adaptation to the scan order only and therefore the derivation of the values f(x) is independent from thescan order140 and the template's56 shape and size. Note that by setting the size and the shape of thetemplate56 such that the coverage of all scan positions of ablock10 for every scan position is allowed, the usage of all already coded transform coefficients in the current block or shape is achieved.

As state before, the selection of the context model indices and the derivation of the symbolization parameter use evaluation values f(x). In general, a generic set of mapping functions maps the resulting evaluation values f(x) onto a context model index and on a specific symbolization parameter. In addition to that, additional information as the current spatial position of the current transform coefficient inside of the transform block orshape10 or the last significant scan position L may be used for the selection of context models related to the coding of transform coefficients and for the derivation of the symbolization parameter. Note that the information resulting from the evaluation and spatial location or the last information may be combined and therefore a specific weighting is possible. After the evaluation and the derivation process, all parameters (context model indices, symbolization parameter) are available for the coding of a whole transform coefficient level or a transform coefficient up to a specific limit.

As an example configuration of the presented invention, the cutoff set size is empty. This means, each transform coefficient is transmitted completely before processing the next transform coefficients along the scan order.

The evaluation values f(x) may result from the evaluation of already coded neighbors covered by thelocal template56. A specific mapping function f_t(x) maps the input vector to an evaluation value used to select the context model and the Rice parameter. The input vector x may consist of transform coefficient values x_iof the neighbors covered by thelocal template56 and depends on the interleaving scheme. For example, if the cutoff set c is empty and the sign is coded in a separate scan pass, the vector x consists of absolute transform coefficients x_ionly. In general, the values of the input vector x can be signed or unsigned. The mapping function can he formulated as follows with an input vector x of dimension of d (given t as a constant input).

f_t(x)=Σ_i=1^i=dw_ig_t(x_i)δ(x_i,t)

To be more specific, the mapping function f_t(x) may be defined as follows with an input vector x of dimension of d (given t as a constant input).

f_t(x)=Σ_i=1^i=dw_ig_t(x_i)δ(x_i,t)

That is, g_t(x_i) may be (|x_i|−t). In the latter formula, the function δ is defined as follows (given t as a constant input):

\begin{matrix} δ (x, t) = {\begin{matrix} 1 \langle x \rangle \geq t \\ 0 \langle x \rangle < t \end{matrix} & (1) \end{matrix}

Another kind of evaluation value is the number of neighboring absolute transform coefficients levels larger or smaller than a specific value t defined as follows:

f_{t} (x) = \sum_{i = 0}^{i = d} w_{i} δ (x_{i}, t)

Note that for both kind of evaluation values, an additional weighting factor controls the importance of a specific neighbor is possible. For example, the weighting factor w_tis higher for neighbors with shorter spatial distance than for neighbors with larger spatial distance. Further, the weighting is neglected when setting all w_ito one.

As an example configuration of the presented invention, f₀, f₁, f₂and f₃are evaluation values with respective i of {0, 1, 2, 3} and δ(x_i) as defined in (1). For this example, f₀is used for the derivation of the context index of the first bin, f₁, for the second bin, f₂for the third bin, and f₃for the Rice parameter. In another example configuration, Jo is used for the context model selection of the first bin, while f₃is taken for the context model selection of the second, the third bin, and the Rice parameter. Here, the Rice parameter serves as a representative also for other symbolization parameters.

The context model selection for all syntax elements or bin indices in the entropy coding and the symbolization parameter uses the same logic by employing the evaluation values f(x). In general, a specific evaluation value f(x) is mapped by another mapping function g(x,n) to a context model index or a symbolization parameter. A specific mapping function is defined as follows with d as the dimension of the input vector n.

g (x) = \sum_{i = 1}^{i = d} δ (x, n_{i})

For this mapping, the function δ(x,n) can be defined as follows.

δ (x, n) = {\begin{matrix} 1 x > n \\ 0 x \leq n \end{matrix}

The dimension d of the input vector n and the values of the vector n may be variable and depend on the syntax element or bin index. Further, the spatial location inside of the transform block or shape may be used to add or subtract (or to move) the selected context model index.

The first scan position in scanning the transform coefficients when coding/decoding same, may be the last scan position L when applying the scan direction ofFIG. 1 pointing from DC to highest frequency. That is, at least the first scan of the scans for traversing the coefficients in order to code/decode same, may point from coefficient L to DC. For this scan position L, the first bin index may be neglected as the last information already signalized that this scan position consists of a transform coefficient unequal to zero. For this scan position, a separate context model index can be used for the coding of the second and the third bin resulting from the binarization of the transform coefficient.

As an example configuration of the presented invention, the resulting evaluation value f₀is used as input together with the input vector n={1,2,3,4,5}, and the resulting value is the context model index for the first bin. Note that, in case of evaluation value equal to zero, the context index is zero. The same scheme is applied with the evaluation value f₁and the input vector n={1,2,3,4} and the resulting value is the context model index for the second and the third bin of the binarization. For the Rice parameter, f₃and n={0,5,19} is used. Note that the maximum Rice parameter is three and therefore no change in the maximum Rice parameter compared to the state-of-the-art is done by the presented invention. Alternatively, f₁can be used to derive the Rice parameter. For that configuration, the input vector should be modified to n={3,9,21}. Note that the underlying set of rules are the same for all syntax element or bin indices and for the Rice parameter, only the parameters or threshold sets (input vector n) are different. Further, depending on the diagonal of the current scan position, the context model index may be modified as stated before by add or subtract a specific amount. An equivalent description for that is the selection of another disjoint context model set. In an example implementation, the resulting context model index for the first bin is moved by 2*|ctx0|if the current scan position lies on the first two diagonals. If the current scan position lies on the third and the fourth diagonal, the context model index for the first bin is moved by |ctx0|, where |ctx0| is the number of maximum context models resulting from the derivation base on the evaluation values resulting in disjoint context model sets. This concept is used for luma planes only for an example implementation, while no further offset is added in case of chroma avoiding context dilution (i.e. not enough bins are coded with an adaptive context model and the statistic cannot be tracked by the context model). The same technique may be applied to the context model index of the second and the third bin. Here, in an example configuration of the presented invention, the threshold diagonals are three and ten. Again, this technique is applied to the luma signal only. Note that it is also possible to extend this technique to the chroma signals. Further, note that the additional index offset depending on diagonals can be formulated as follows.

ctx_offset=d_i*idx_inc

In this formula, d_jdenotes the weight for the diagonal of the current scan position and idx_incdenotes the step size. Further, note that the offset index can be inverted for practical implementations. For the stated example implementation, an inversion would be set the additional index to zero if the current scan position lies on the first and the second diagonal, is moved by ctx0 for the third and the forth diagonal and is 2*|ctx0| otherwise. By using the given formula, the same behavior as for the example configuration is achieved when setting d₀and d₁to 2, d₃and d₄to 1 and the all remaining diagonal factors to 0.

Even if the context model index is equal for different block sizes or plane types (e.g. luma and chroma), the base context model index can be different resulting in different set of context models. For example, the same base index for block sizes larger than 8×8 in luma may be used, while the base index may be different for 4×4 and 8×8 in luma. In order to have a meaningful number of context models, the base index may, however, be grouped in a different way.

As an example configuration, the context models for 4×4 blocks and the remaining blocks may be different in luma, while the same base index may be used for the chroma signal. In another example, the same base index may be used for both luma and chroma signals, while the context models for luma and chroma are different. Furthermore, the context models for the second and the third bins may be grouped resulting in a smaller number of context memory. If the context model index derivation for the second and the third bin is equal, the same context model may be used to transmit the second and the third bin. By a right combination of base index grouping and weighting, a meaningful number of context models may be achieved resulting in a saving of context memory.

In an embodiment of the invention, the cutoff set c is empty. That is, merely one scan is used. For this embodiment, the sign information can be interleaved using the same scan pass or can be coded in a separate scan pass. In another embodiment, the set size c is equal to one and co, the first and the only value of the cutoff set c is equal to three. This corresponds to the example illustrated above with using two scans. In this embodiment, the context model selection may be done for all three bins resulting from the truncated unary binarization while the symbolization parameter derivation such as Rice parameter selection may be done using thesame function52.

In an embodiment, the size of the local template is five. The size of the local template may be four. For this embodiment, the neighbor with the spatial distance of two in vertical direction may be removed compared toFIG. 8. In a further embodiment, the template size is adaptive and is adjusted to the scan order. For this embodiment, the neighbor that is coded in a processing step before is not included into the template just as it is the case inFIGS. 1 and 8. By doing this, the dependency or latency is shortened, resulting in a higher processing order. In a further embodiment, the template size and shape is adjusted large enough (e.g. same block or shape size of the current block or shape). In another embodiment, two local templates may be used and they may be combined by a weighting factor. For this embodiment, the local templates may differ in size and shape.

In an embodiment, f₀may be used to select the context model index for the first bin and f₁for the second bin, the third bin, and the Rice parameter. In this embodiment, the input vector n={1,2,3,4,5} resulting in 6 context models. The input vector n for the second and the third bin index may be the same and n={1,2,3,4}, while the input vector n for the Rice parameter may be n={3,9,21}. Furthermore, in an embodiment, the afore-mentioned frequency portions of the transform block within which separate context sets are used, may be formed by disjoint sets of diagonals (or lines) of the diagonal (raster) scan. For example, different context base offset numbers may exist for the first and second diagonals, the second and third diagonals and the fourth and fifth diagonals when seen from DC component, so that the context selection for coefficients in these diagonals takes place within disjoint sets of contexts. Note that the first diagonal is one. For the second and the third bin index, diagonals lying in the range between [0,2] have a weighting factor of two and diagonals lying in the range between [3,9] have a weighting factor of one. These additional offsets are used in the case of luma signal, while the weighting factors for chroma are all equal to zero. Also for this embodiment, the context model for the second and the third bin index of the first scan position, which is the last significant scan position, is separated from the remaining context models. This means that the evaluation process can never select this separate context model.

In an embodiment, 4×4 luma blocks or shape uses a single set of context for the first bin, while the context models for the remaining block sizes or shape are the same. In this embodiment, there is no separation between the block size or shape for the chroma signal. In another embodiment of the invention, there is no separation between block sizes or shape results in the same base index or context model sets for all block sizes and shape. Note that for both embodiments, different set of context models are used for luma and chroma signals.

Below, an embodiment using a modified Rice parameter binarization according to above embodiments, but without context adaptive entropy coding is shown. According to this alternative coding scheme, the Rice binarization scheme is used only (with, optionally, addition of an Exp-Golomb suffix). Thus, no adaptive context model is necessitated to code a transform coefficient. For that alternative coding scheme, the Rice parameter derivation uses the same rule as for the above embodiments.

In other words, in order to reduce the complexity and context memory and to improve the latency in the coding pipeline, an alternative coding scheme that is based on the same set of rules or logic is described. For this alternative coding scheme, the context model selection for the first three bins resulting from the binarization is disabled and the first three bins resulting from the Truncated Unary binarization, i.e. the first symbolization scheme, may be coded with the fixed equal probability (i.e. with a probability of 0.5). Alternatively, the Truncated Unary binarization scheme is omitted and the interval bounds of the binarization scheme are adjusted. In this usage, the left bound of the Rice interval, i.e.interval18, is 0 instead of 3 (withinterval16 vanishing). The right/upper bound for this usage can be unmodified or can be subtracted by 3. The derivation of the Rice parameter can be modified in terms of evaluation values and in terms of the input vector n.

Thus, in accordance with the just-outlined modified embodiments, an apparatus for decoding a plurality of transform coefficients of different transform blocks, each having a transform coefficient level, from adata stream32, may be constructed and operate as shown in, and described with respect toFIG. 10.

The apparatus ofFIG. 10, comprises anextractor120 configured to extract a set of symbols orsymbol sequence122 from thedata stream32 for a current transform coefficient. The extraction is performed as described above with respect toextractor84 ofFIG. 7.

Adesymbolizer124 is configured to map theset122 of symbols onto a transform coefficient level for the current transform coefficient in accordance with a symbolization scheme which is parameterizable in accordance with a symbolization parameter. The mapping may solely use the parametrizable symbolization scheme such as a Rice binatization, or may use this parametrizable symbolization scheme merely as a prefix or suffix of an overall symbolization of the current transform coefficient. In case ofFIG. 2, for example, the parametrizable symbolization scheme, i.e, the second one, formed the suffix relative to the first symbolization scheme's symbol sequence.

To present more examples, reference is made toFIG. 11aandb.According toFIG. 11a,theinterval range20 of the transform coefficients is subdivided into three

intervals

16,18 and126, together covering theinterval range20 and overlapping each other at a respective maximum level of the respective lower interval. If the coefficient level x is within thehighest interval126, the overall symbolization is a combination ofsymbol sequence44 of thefirst symbolization scheme128 symbolizing levels withininterval16, the symbol sequence forming a prefix, followed by a first suffix, namely asymbol sequence42 of thesecond symbolization scheme130 symbolizing levels withininterval18, and further followed by a second suffix, namely asymbol sequence132 of athird symbolization scheme134 symbolizing levels withininterval126. The latter may be an Exp-Golomb code such as oforder 0. If the coefficient level x is within the mid interval18 (but not within interval126), the overall symbolization is a combination of merely prefix44 followed by thefirst suffix42. If the coefficient level x is within the lowest interval16 (but not within interval18), the overall symbolization is merely composed ofprefix44. The overall symbolization is composed such that same is prefix-free. Without the third symbolization, the symbolization according toFIG. 11amay correspond to the one ofFIG. 2. Thethird symbolization scheme134 may be a Golomb-Rice binarization, Thesecond symbolization scheme130 may form the parametrizable one, although same could also be the first128.

An alternative overall symbolization is shown inFIG. 1. Here, merely two symbolization schemes are combined. Compared toFIG. 11a,the first symbolization scheme has been left away. Depending on x within theinterval136 ofscheme134, or theinterval138 of scheme130 (outside of interval136), the symbolization of x comprisesprefix140 andsuffix142, or merely prefix140.

Further, the apparatus ofFIG. 10 comprises asymbolization parameter determinator144 connected between the desymbolizer's output and a parameter input ofdesymbolizer124.Determinator144 is configured to determine thesymbolization parameter46 for the current transform coefficient depending, viafunction52, on previously processed transform coefficients (as far as derivable from the desymbolized fragments or portions desymbolized/processed/decoded so far).

Theextractor120, thedesymbolizer124 and thesymbolization parameter determinator144 are configured to sequentially process the transform coefficients of the different transform blocks as it was described above. That is, scan140 may be traversed in opposite direction within atransform block10. Several scans may be used such as, for example, for the different symbolization fragments, i.e. prefix and suffix(es).

The function parameter varies depending on a size of the current transform coefficient's transform block, an information component type of the current transform coefficient's transform block and/or a frequency portion the current transform coefficient is located within the transform block.

The apparatus may be configured such that the function defining the relationship between the previously decoded transform coefficients on the one hand, and the symbolization parameter on the other hand, is g(f(x)), which function has already been described above.

As has also been discussed above, spatial determination of the previously processed transform coefficients depending on a relative spatial arrangement relative to the current transform coefficient may be used.

The apparatus may operate very easily and fast, as theextractor120 may be configured to extract the set of symbols from the data stream directly or using entropy decoding using a fixed probability distribution. The parametrizable symbolization scheme may be such that the set of symbols is of a Rice code, and the symbolization parameter is a RICE parameter.

In other words, thedesymbolizer124 may be configured to restrict the symbolization scheme to a level interval such as 18 or 138 out of arange interval20 of the transform coefficients so that the set of symbols represents a prefix or suffix with respect to other portions of an overall symbolization of the current transform coefficient such as 44 and 132, or 142. As to the other symbols, same may also be extracted from the data stream directly or using entropy decoding using a fixed probability distribution, butFIGS. 1 to 9 showed that entropy coding using context adaptivity may also be used.

Apparatus ofFIG. 10 may be used asapparatus102 in thepicture decoder102 ofFIG. 8.

For sake of completeness,FIG. 12 shows, an apparatus for coding a plurality of transform coefficients of different transform blocks, each having a transform coefficient level, into adata stream32, fitting to apparatus ofFIG. 10.

The apparatus ofFIG. 12 comprises asymbolizer150 configured to map a transform coefficient level for a current transform coefficient in accordance with a symbolization scheme which is parameterizable in accordance with a symbolization parameter, onto a set of symbols or symbol sequence.

Aninserter154 is configured to insert the set of symbols for the current transform coefficient into thedata stream32.

Asymbolization parameter determinator156 is configured to determine thesymbolization parameter46 for the current transform coefficient depending, via afunction52 parameterizable via a function parameter, on previously processed transform coefficients, and may, to this end, be connected between an output ofinserter152 and parameter input ofsymbolizer150, or, alternatively, between output and input ofsymbolizer150.

Inserter

154,symbolizer150 andsymbolization parameter determinator156 may be configured to sequentially process the transform coefficients of the different transform blocks, and the function parameter varies depending on a size of the current transform coefficient's transform block, an information component type of the current transform coefficient's transform block and/or a frequency portion the current transform coefficient is located within the transform block.

As stated above with respect to the decoding apparatus ofFIG. 10, the apparatus ofFIG. 12 may be configured such that the function defining the relationship between the previously decoded transform coefficients on the one hand, and the symbolization parameter on the other hand, is g(f(x)), and the previously processed transform coefficients may spatially be determined depending on a relative spatial arrangement relative to the current transform coefficient. The inserter may be configured to insert the set of symbols into the data stream directly or using entropy encoding using a fixed probability distribution, and the symbolization scheme may be such that the set of symbols is of a Rice code, and the symbolization parameter is a RICE parameter. The symbolizer may be configured to restrict the symbolization scheme to a level interval out of arange interval20 of the transform coefficients so that the set of symbols represents a prefix or suffix with respect to other portions of an overall symbolization of the current transform coefficient.

As mentioned above, in an implementation ofFIGS. 10 to 12 embodiments, the context model selection for the first three bins is disabled as compared to the embodiments ofFIGS. 1 to 9. For this embodiment, the resulting bins from the TruncatedUnary binarization128 are coded with a fixed probability of 0.5. In a further embodiment, the TruncatedUnary binarization128 is omitted as shown inFIG. 11band the bounds for the Rice interval is adjusted resulting in the same interval range as in the state of the art (i.e, left and right bounds minus 3). For this embodiment, the Rice parameter derivation rule is modified compared to the embodiment ofFIGS. 1 to 9. instead using f₁as evaluation value, f₀may be used, for example. Further, the input vector may be adjusted to n={4,10,22}.

Thus, in accordance with this embodiment, the following is performed by the context adaptive entropy decoder/encoder in order to select the context of significant_coeff_flag, i.e. a flag which is part of the significance map and signals for a certain transform coefficient of a sub-block for which coded_sub_block_flag signals that therespective sub-block200 contains non-zero transform coefficients, as to whether the respective coefficient is significant, i.e. non-zero, or not.

Inputs to this process are the colour component index cldx, the current coefficient scan position (xC, yC), the scan order index scanIdx, the transform block size log2TrafoSize. Output of this process is ctxIdxInc.

The variable sigCtx depends on the current position (xC, yC), the colour component index cIdx, the transform block size and previously decoded bins of the syntax element coded_sub_block_flag. For the derivation of sigCtx, the following applies.

- If log2TrafoSize is equal to 2, sigCtx is derived using ctxIdxMap[ ] specified in table 1 as follows.

sigCtx=ctxIdxMap[(yC»2)+xC]

- Otherwise, if xC+yC is equal to 0, sigCtx is derived as follows. sigCtx=0
- Otherwise, sigCtx is derived using previous values of coded_sub_block_flag as follows.
  - The horizontal and vertical sub-block positions xS and yS are set equal to (xC»2) and (yC»2), respectively.
  - The variable prevCsbf is set equal to 0.
  - When xS is less than (1«(log2TrafoSize−2))−1, the following applies. prevCsbf +=coded_sub_block_flag[xS+1][yS]
  - When yS is less than (1»(log2TrafoSize−2))−1, the following applies. prevCsbf +=(coded_sub_block_flag[xS][yS+1]«1)
  - The inner sub-block positions xP and yP are set equal to (xC & 3) and (yC & 3), respectively.
  - The variable sigCtx is derived as follows.
    - If prevCsbf is equal to 0, the following applies. sigCtx=(xP+yP==0)?2:(xP+yP<3)?1:0
    - Otherwise, if prevCsbf is equal to 1, the following applies. sigCtx=(yP==0)?2:(yP==1)?1:0
    - Otherwise, if prevCsbf is equal to 2, the following applies. sigCtx=(xP==0)?2:(xP==1)?1:0
    - Otherwise (prevCsbf is equal to 3), the following applies. sigCtx=2
    - The variable sigCtx is modified as follows.
      - If cldx is equal to 0, the following applies.
      - When (xS+yS) is greater than 0, the following applies. sigCtx+=3
      - The variable sigCtx is modified as follows.
      - If log2TrafoSize is equal to 3, the following applies. sigCtx+=(scanIdx==0)?9:15
      - Otherwise, the following applies. sigCtx+=21
    - Otherwise (cIdx is greater than 0), the following applies.
      - If log2TrafoSize is equal to 3, the following applies. sigCtx+=9
      - Otherwise, the following applies. sigCtx+=12

The context index increment ctxIdxInc is derived using the colour component index cIdx and sigCtx as follows.

- If cIdx is equal to 0, ctxIdxInc is derived as follows. ctxIdxInc=sigCtx
- Otherwise (cIdx is greater than 0), ctxIdxInc is derived as follows. ctxIdxInc=27+sigCtx

TABLE 1

Specification of ctxIdxMap [i]

i	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14
ctxIdxMap [i]	0	1	4	5	2	3	4	5	6	6	8	8	7	7	8

As described above, for each significant transform coefficient, further syntax elements or sets of symbols may be conveyed within the data stream in order to signal the levels thereof. In accordance with the embodiment outlined below, for one significant transform coefficient the following syntax elements or sets of transform coefficients are transmitted: coeff_abs_level_greater1_flag, coeff_abs_level_greater2_flag (optional), and coeff_abs_level_remaining so that the level of the currently coded/decoded significant transform coefficient level TransCoeffLevel is

TransCoeffLevel=(coeff_abs_level_remaining+baseLevel)*(1−2*coeff_sign_flag])

With

baseLevel=1+coeff_abs_level_greater1_flag+coeff_abs_level_greater2_flag

Please note that significant_coeff_flag is, per definition, 1 for significant transform coefficients, and accordingly, may be regarded as part of the coding of the transform coefficient, namely part of the entropy coded symbols thereof.

The context adaptive entropy decoder/encoder would, for example, perform the context selection for coeff_abs_level_greater1_flag as follows. For example, the current sub-block scan index i would increase alongscan path202 into the direction of DC, and the current coefficient scan index n would increase within the respective sub-block within which the currently coded/decoded transform coefficient position is located, alongscan path204, wherein, as outlined above, different possibilities exist for the

scan paths

202 and204, and same may actually be variable according to an index scanIdx.

Inputs to this process of selecting the context of coeff_abs_level_greater1_flag are the colour component index cldx, the current sub-block scan index i and the current coefficient scan index n within the current sub-block.

Output of this process is ctxIdxInc.

The variable ctxSet specifies the current context set and for its derivation the following applies.

- If this process is invoked for the first time for the current sub-block scan index i, the following applies.
  - The variable ctxSet is initialized as follows.
  - If the current sub-block scan index i is equal to 0 or cldx is greater than 0, the following applies. ctxSet=0
  - Otherwise (i is greater than 0 and cldx is equal to 0), the following applies. ctxSet=2
  - The variable lastGreaterICtx is derived as follows.
    - If the current sub-block with scan index i is the first one to be processed in this subclause for the current transform block, the variable lastGreaterICtx is set equal to 1.
    - Otherwise, the variable lastGreaterICtx is set equal to the value of greaterICtx that has been derived during the last invocation of the process specified in this subclause for the syntax element coeff_abs_level_greaterI_flag for the previous sub-block with scan index i+1.
  - When lastGreaterICtx is equal to 0, ctxSet is incremented by one as follows. ctxSet=ctxSet+1
  - The variable greaterICtx is set equal to 1.
- Otherwise (this process is not invoked for the first time for the current sub-block scan index i), the following applies.
  - The variable ctxSet is set equal to the variable ctxSet that has been derived during the last invocation of the process specified in this subclause.
  - The variable greaterICtx is set equal to the variable greaterICtx that has been derived during the last invocation of the process specified in this subclause.
  - When greaterICtx is greater than 0, the variable lastGreaterIFlag is set equal to the syntax element coeff_abs_level_greaterI_flag that has been used during the last invocation of the process specified in this subclause and greaterICtx is modified as follows.
    - If lastGreaterIFlag is equal to 1, greaterICtx is set equal to 0.
    - Otherwise (lastGreaterIFlag is equal to 0), greaterICtx is incremented by 1.

The context index increment ctxIdxInc is derived using the current context set ctxSet and the current context greaterICtx as follows.

ctxIdxInc=(ctxSet*4)+Min(3, greaterICtx)

When cldx is greater than 0, ctxIdxInc is modified as follows.

ctxIdxInc=ctxIdxInc+16

The process of selecting the context of coeff_abs_level_greater2_flag could be made the same as coeff_abs_level_greater2_flag with the following difference:

The context index increment ctxIdxInc is set equal to the variable ctxSet as follows.

ctxIdxInc=ctxSet

When cIdx is greater than 0, ctxIdxInc is modified as follows.

ctxIdxInc=ctxIdxInc+4

For the symbolization parameter selection, the following would be performed by the symbolization parameter determinator in order to determine the symbolization parameter which, here, comprises cLastAbsLevel and cLastRiceParam.

Input to this process is a request for a binarization for the syntax element coeff_abs_level_remaining[n], and baseLevel.

Output of this process is the binarization of the syntax element.

The variables cLastAbsLevel and cLastRiceParam are derived as follows.

- If n is equal to 15, cLastAbsLevel and cLastRiceParam are set equal to 0.
- Otherwise (n is less than 15), cLastAbsLevel is set equal to baseLevel+coeff_abs_level_remaining[n+1] and cLastRiceParam is set equal to the value of cRiceParam that has been derived during the invocation of the binarization process as specified in this subclause for the syntax element coeff_abs_level_remaining[n+1] of the same transform block.

The variable cRiceParam is derived from cLastAbsLevel and cLastRiceParam as:

cRiceParam

Min cLastRiceParam+(cLastAbsLevel>(3*(1«cLastRiceParam))?1:0), 4)

The variable cTRMax is derived from cRiceParam as:

cTRMax=4«cRiceParam

The binarization of coeff_abs_level_remaining may consist of a prefix part and (when present) a suffix part.

The prefix part of the binarization is derived by invoking, for example, Rice binarization process for the prefix part Min(cTRMax, coeff_abs_level_remaining[n]).

When the prefix bin string is equal to the bit string oflength4, for example, with all bits equal to 1, the bin string may consists of a prefix bin string and a suffix bin string. The suffix bin string may be derived using an Exp Golomb order-k binarization for the suffix part (coeff_abs_level_remaining[n]−cTRMax) with the Exp-Golomb order k set equal tocRiceParam+1, for example.

It should be noted that above embodiments may be varied. For example, the dependency on the colour component index cIdx could be left away. Merely one color component would, for example, be considered. Further, all of the explicit values could be varied. In so far, the just-outlined examples are to be interpreted broadly so as to also incorporate variations.

In the above example, the embodiments outlined above may advantageously be used in the following way. In particular, the determination of CtxIdxInc for coeff_abs_level_greater1_flag on the one hand and the symbolization parameter determination for coeff_abs_level remaining is harmonized exploiting the above functions f and g by setting the function parameters in the following way.

To this end,FIG. 16 exemplarily shows a “current transform coefficient” illustrated with across206. Same is representative for any transform coefficient with which any of the subsequently mentioned syntax elements are associated. It is positioned at (xP,yP)=(1,1) and (xC,yC)=(1,5) within current sub-block (xS,yS)=(0,1). Right-neighbor sub-block is at (xS,yS)=(1,1), bottom-neighbor sub-block is at (xS,yS)=(0,2) and the immediately previously coded sub-block depends on thescan path202. Here, exemplarily, adiagonal scan202 is shown, and the sub-block coded/decoded immediately preceding the current sub-bock is at (xS,yS)=(1,0).

Let's, again, rewrite the formulas for the common parametrizable function

g(f(x))=Σ_i=1^d^fδ′(f(x), (n_i) (1)

f(x)=Σ_tw_i×h(x_i)×δ(x_i, t) (2)

For selecting the context of significant_coeff_flag forcurrent coefficient206, the following could be computed by the entropy encoding/decoding apparatus. That is, same would use function (1) with (2) with having the function parameters t, h and w set as follows:

For function (2), w_i=1 for all x_iwithin the neighboring sub-blocks to the right and to below of, the current sub-block, and w_i=0 elsewhere inblock10;

h(x_i)=1 for all x_i, within the neighboring sub-block to the right of the current sub-block; if present, same has been previously scanned in thesub-block scan202; in case, more than onescan202 is available, all may be such that, independent of scanIdx, the neighboring sub-block to the right has its coefficients coded/decoded prior to the current sub-block;

h(x_i)=2⁴+1 for all x_iwithin the neighboring sub-block below the current sub-block previously scanned in the sub-block scan (independent of scanIdx);

h(x_i)=0 otherwise;

t=1;

If the value off equals 0, this signals the case that none of the neighboring sub blocks to the right and below the current sub-block Nachbarn comprises any significant transform coefficient;

If the value of f falls between 1 and 16, both inclusively, this corresponds to the fact that coded_sub_block_flag equals 1 in the right neighbor sub-block

If the value off is a multiple of 2⁴+1 (without reminder), This corresponds to the fact that coded_sub_block_flag equals 1 in the bottom neighbor sub-block

If the value of f is a multiple of 2⁴+1, but with reminder, this means that coded_sub_block_flag equals 1 for both neighboring sub-blocks, namely the one to the right of, and the one to below of, the current sub-block;

For function (1), n is set as follows with d_fbeing 3:

n=(0,2⁴,m)

with m = {\begin{matrix} 2^{16} if f (x) \leq 2^{4} \\ f (x) - f (x) % (2^{4} + 1) else \end{matrix}

By this measure, the variable component of the context index is determined using g(f) with the above function parameters based on already coded(decoded coefficients.

For selecting the context of coeff_abs_greater1_flag, the following could be computed by the entropy encoding/decoding apparatus. That is, same would use function (1) with (2) with having the function parameter set as follows:

For function (2), the parameters are set as follows:

w_i=1 is set for all x_i; in the immediately preceding sub-block and the current sub-block, and zero for all the others.

- h(x)=1 for all x_iin the current sub-block with |x_i|=1 h(x_i)=2⁴for all x_iin the current sub-block with |x_i|>1 h(x)=2¹⁶for all x_iin the immediately preceding sub-block t=2

For function (1) n is set as follows with d_fbeing 8:

n=(0,1,2,2⁴,2¹⁶,2¹⁶,+2¹⁶+2, 2¹⁶+2⁴)

For selecting the context of coeff_abs_greater2_flag, the following could be computed by the entropy encoding/decoding apparatus. In particular, same would use function (1) with (2) with having the function parameter set as described above with respect to coeff_abs_greater2_flag, but with d_fbeing 1:

n=(2¹⁶)

For determining the symbolization parameter for coeff_abs_level remaining, the symbolization parameter determiner could use the common function (1) with the function parameters set as follows:

For function (2), the parameters are set as follows:

w_i=1 for all x_iin the current cub-block, but zero elsewhere

h(x_i)=1 fort he most recently - in accordance with the internal coefficient scan204—visited coefficient x_ifor which coeff_abs_level_remaining has been coded, i.e. the level of which fell into the interval corresponding to the symbolization scheme;

h(x_i)=0 elsewhere in the template

t=0

For function (1) n is et as follows:

n = (2^{m} with m = {\begin{matrix} k if k < 4 \\ 2^{16} if k = 4 \end{matrix}

where k is the symbolization parameter, e.g. the Rice Parameter, for the afore-mentioned most recently—in accordance with the internal coefficient scan204—visited coefficient. Using the resulting g(f), the symbolization parameter for thecurrent coefficient206 is determined.

The following syntax could be used to transfer the just-outlined syntax elements.


residual_coding( x0, y0, log2TrafoSize, cIdx ) {
if( transform_skip_enabled_flag && !cu_transquant_bypass_flag && (log2TrafoSize = = 2
))
transform_skip_flag[ x0 ][ y0 ][ cIdx ]
last_significant_coeff_x_prefix
last_significant_coeff_y_prefix
if( last_significant_coeff_x_prefix > 3 )
last_significant_coeff_x_suffix
if( last_significant_coeff_y_prefix > 3 )
last_significant_coeff_y_suffix
lastScanPos = 16
lastSubBlock = ( 1 << (log2TrafoSize - 2))*(1<<( 1og2TrafoSize - 2)) - 1
do {
if( lastScanPos = = 0 ){
lastScanPos = 16
lastSubBlock- -
}
lastScanPos- -
xS = ScanOrder[ log2TrafoSize - 2 ][ scanIdx ][ lastSubBlock ][ 0 ]
yS = ScanOrder[ log2TrafoSize - 2][ scanIdx ][ lastSubBlock ][ 1 ]
xC = ( xS <<2 ) + ScanOrder[ 2 ][ scanIdx ] [ lastScanPos ] [ 0 ]
yC = ( yS <<2 ) + ScanOrder[ 2 ][ scanIdx ] [ lastScanPos ][ 1 ]
} while( ( xC != LastSignificantCoeffX ) \| \| ( yC != LastSignificantCoeffY ))
for( i = lastSubBlock; i >= 0; i- - ) {
xS = ScanOrder[ log2TrafoSize - 2 ][ scanIdx ][ i ][ 0
] yS = ScanOrder[ log2TrafoSize - 2 ][ scanIdx ][ i ][
1 ] inferSbDcSigCoeffFlag = 0
if( (i <lastSubBlock) && (i > 0)) {
coded_sub_block_flag[ xS ][ yS ]
inferSbDcSigCoeffFlag = 1
}
for( n = (i = = lastSubBlock) ? lastScanPos - 1 : 15; n >= 0; n- -) {
xC = ( xS <<2 ) + ScanOrder[ 2 ] [ scanIdx ] [ n ] [ 0 ]
yC = ( yS <<2 ) + ScanOrder[ 2 ] [ scanIdx ] [ n ] [ 1 ]
if( coded_sub_block_flag[ xS ][ yS ] && ( n >0 \| \| !inferSbDcSigCoeffFlag )) {
significant_coeff_flag[ xC ][ yC ]
if( significant_coeff_flag[ xC ][ yC ] )
inferSbDcSigCoeffFlag = 0
}
}
firstSigScanPos =16
lastSigScanPos = -1
numGreater1Flag = 0
lastGreater1ScanPos = -1
for( n = 15 n >= 0; n- - ) {
xC = ( xS <<2 ) + ScanOrder[ 2 ] [scanIdx ] [ n ] [ 0
] yC = ( yS <<2 ) + ScanOrder[ 2 ] [scanIdx ] [ n ] [
1 ] if( significant_coeff_flag[ xC ][ yC ]) {
if( numGreater1Flag <8 ) {
coeff_abs level_greater1_flag[ n ]
numGreater1Flag++
if( coeff_abs_level_greater1_flag[ n ] && lastGreater1ScanPos = = -1)
lastGreater1ScanPos = n
}
if( lastSigScanPos = = -1)
lastSigScanPos = n
firstSigScanPos = n
}
}
signHidden = (lastSigScanPos - firstSigScanPos >3 && !cu_transquant_bypass_flag )
if( lastGreater1ScanPos != -1)
coeff_abs_level_greater2_flag[ lastGreater1ScanPos ]
for( n = 15; n >= 0; n- - ) {
xC = ( xS <<2 ) + ScanOrder[ 2 ][ scanIdx ][ n ][ 0 ]
yC = ( yS <<2 ) + ScanOrder[ 2 ][ scanIdx ][ n ][ 1 ]
if( significant_coeff_flag[ xC ][ yC ] &&
( !sign_data_hiding_flag \| \| !signHidden \| \| n != firstSigScanPos ))
coeff_sign_flag[ n ]
}
numSigCoeff = 0
sumAbsLevel = 0
for( n = 15; n >= 0; n- - ) {
xC = ( xS <<2 ) + ScanOrder[ 2 ][ scanIdx ][ n ][ 0 ]
yC = ( yS <<2 ) + ScanOrder[ 2 ][ scanIdx ][ n ][ 1 ]
if( significant_coeff_ flag[xC ][ yC ]) {
baseLevel = 1 + coeff_abs_level_greater1_flag[ n ] +
coeff_abs_level_greater2_flat[ n ]
if( baseLevel = = (( numSigCoeff <8 ) ? ( (n = = lastGreater1ScanPos) ? 3:2 ) :
1 ) )
coeff_abs_level_remaining[ n ]
TransCoeffLevel[ x0 ][ y0 ][ cIdx ][ xC ] [ yC ] =
( coeff_abs_level_remaining[ n ] + baseLevel) * ( 1 - 2 * coeff_sign_flag[ n ]
)
if( sign_data_hiding_flag && signHidden ) {
sumAbsLevel += ( coeff_abs_level_remaining[ n ] + baseLevel)
if( n = = firstSigScanPos && (( sumAbsLevel % 2) == 1))

TransCoeffLevel[x0][y0][cIdx][xC][yC]

=

−

TransCoef1Level[x0][y0][cIdx][xC][yC]

}

numSigCoeff++

}

The syntax indicates that the level of the transformation coefficient is composed of coeff_abs_level_remaining and haseLevel, wherein haseLevel is composed of 1+coeff_abs_level_greater1_flag[n]+coeff_abs_level_greater2_flag[n]. 1 is used, as at this location (or at the time where the levels are reconstructed in the decoder) the syntax element is significant_coeff_flag=1. “First set” would then be the TU code (Rice code with parameterization equal to 0)—from this the first 3 syntax elements are formed. “Second set” then forms the syntax element coeff_abs_level_remaining.

As the boundary is shifted between “first” and “second set” the maximum value is either defined by coeff_abs_greater1_flag, coeff _abs_greater2_flag or by .significant_coeff_flag, hence the branches depending on the syntax elements in the table.

The above settings of the function parameters are still motivated a little in the following.

g(f) forms the sum of the neighboring coefficients and using the result a context and a desymbolization parameter are derived, wherein a later modification may be executed depending on the spatial position.

g(x) acquires one single value. This value corresponds to the result of the function f(x). Knowing this, the context selection and also the parameterization of the Rice parameter may be derived. significant_coeff_flag: As h may itself be a function of x, f(x) or any other function may be chained again and again. Function f(x) with w_i−1 for all positions in theright hand 4×4 sub- block, t=1 and h a function which is configured just like f(x) but inverted, so that in the end the

value

0 or 1 results, i.e. h(x)=min(1, f(x)).

Equivalently, for the second entry this is applied to the bottom 4×4 sub-block. Then, prevCsbf=h₀±2×h₁, wherein prefCsbf may also be a function h within f(x).

If t=∞ is set, the values of the syntax element coded_sub_block_flag may be derived. Thus, a value between 0 and including 3 is acquired as a result for the outermost f(x). The parameter n for g(x) would then either be (xP+yP), xP, yP, or (0,0). If f(x)=0 results, then n=(xP+yP,xP+yP+3), for f(x)=l n=(yP,yP+1) results, for f(x)=2 n=(xP,xP+1) results, and for f(x)=3 n=(0,0) results. So to speak, f(x) may be evaluated directly in order to determine n. The remaining formulae above merely describe an adaption depending on luma/chroma and a further dependency on the global position and scan. In case of a pure 4×4 block, f(x) may be configured so that the value for prevCsbf=4 (may also be different) and thus the mapping table may be reproduced.

coeff_abs_level_greater1_flag: Here, the evaluation of the sub-blocks is similar, wherein only the preceding sub-block is evaluated. The result is, e.g. 1 or 2 (it only has to be two different values), wherein t=2. This corresponds to the selection of a base index depending on already decoded levels in the preceding sub-block. The direct dependency on the levels located within the sub-block may thus be acquired. Effectively, switching on by one index is executed when a 0 was decoded (limited to 3 starting by 1) and as soon as a 1 was decoded it is set to 0. If the arrangement is not considered, parameterization may be executed as follows, starting from 0. w_i=1 for all levels in the same sub-block and t=3, i.e. f(x) provides the number of levels with coeff_abs_greater1_flag=1. For a further function f(x) t=2, i.e. the number of positions with an encoded syntax element coeff_abs_greater1_flag. The first function is limited, i.e. h₀=f(x)=min(f₀(x),2) and the second function is limited with h₁=f(x)=max(f₁(x),1). All of this chained with a delta function (0 if h₁=1, ho otherwise). For coeff_abs_greater2_flag only the derivation of the set is used (w_iis set to 0 for the chained inner function). coeff_abs_level_remaining: The selection is only limited to the current sub-block and n is derived as described above.

With regard to the just outlined embodiment, the following is noted. In particular, in compliance with the above description, different possibilities exist with regard to the definition of the template: the template could be a moving template, the position of which is determined depending on the position of thecurrent coefficient206. The outline of such an exemplary moving template is depicted inFIG. 13 with a dashedline208, The template is composed of the current sub-block, within which thecurrent coefficient206 is located, the neighboring sub-blocks to the right and below the current sub-block, as well as the one or more sub-blocks which immediately precede the current sub-block in thesub-block scan202 or any of thesub-block scans202 if there are several of them among which one is selectable using a scan index as explained above. As an alternative, thetemplate208 may simply encompass all transformcoefficients12 ofblock10.

In the above example, there are further different possibilities for selecting the values of h and n. These values may, accordingly, set differently. This is somehow also true with respect to as far as those weights are concerned which are set to one. Same may be set to another non-zero value. They do not even have to be equal to each other. As w_iis multiplied with h(x_i), the same product value may be achieved by differently setting non-zero w_i's. Moreover, the symbolization parameter does not have to be a Rice parameter or, differently speaking, the symbolization scheme is not restricted to be a Rice symbolization scheme. As to the context index selection, reference is made to the above description where it was already noted that a final context index may be obtained by adding the context index as obtained using function g(f) to some offset index which is, for example specific for the respective type of syntax element, i.e. specific for significant_coeff_flag, coeff_abs_level_greater1_flag, and coeff_abs_level_greater2_flag.

Although some aspects have been described in the context of an apparatus, it is clear that these aspects also represent a description of the corresponding method, where a block or device corresponds to a method step or a feature of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding apparatus. Some or all of the method steps may be executed by (or using) a hardware apparatus, like for example, a microprocessor, a programmable computer or an electronic circuit. In some embodiments, some one or more of the most important method steps may be executed by such an apparatus.

Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or in software. The implementation can be performed using a digital storage medium, for example a floppy disk, a DVD, a Blu-Ray, a CD, a ROM, a PROM, an EPROM, an EEPROM or a FLASH memory, having electronically readable control signals stored thereon, which cooperate (or are capable of cooperating) with a programmable computer system such that the respective method is performed. Therefore, the digital storage medium may be computer readable.

Some embodiments according to the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.

Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for performing one of the methods when the computer program product runs on a computer. The program code may for example be stored on a machine readable carrier.

Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine readable carrier.

In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.

A further embodiment of the inventive methods is, therefore, a data carrier (or a digital storage medium, or a computer-readable medium) comprising, recorded thereon, the computer program for performing one of the methods described herein. The data carrier, the digital storage medium or the recorded medium are typically tangible and/or non-transitionary.

A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein. The data stream or the sequence of signals may for example be configured to be transferred via a data communication connection, for example via the Internet.

A further embodiment comprises a processing means, for example a computer, or a programmable logic device, configured to or adapted to perform one of the methods described herein

A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.

A further embodiment according to the invention comprises an apparatus or a system configured to transfer (for example, electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may, for example, be a computer, a mobile device, a memory device or the like. The apparatus or system may, for example, comprise a file server for transferring the computer program to the receiver

in some embodiments, a programmable logic device (for example a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor in order to perform one of the methods described herein, Generally, the methods are performed by any hardware apparatus.

While this invention has been described in terms of several advantageous embodiments, there are alterations, permutations, and equivalents which fall within the scope of this invention. It should also be noted that there are many alternative ways of implementing the methods and compositions of the present invention. It is therefore intended that the following appended claims be interpreted as including all such alterations, permutations, and equivalents as fall within the true spirit and scope of the present invention.