CN102567995A - Image registration method - Google Patents

Abstract

The invention relates to an image registration method, which comprises steps of obtaining characteristic point sets of a first image and a second image respectively; taking one of the characteristic point sets as a target point set and the other one as a floating point set, and obtaining a mapping point set through conversion of each point in the floating point set; calculating the distance matrix between the mapping point set and each point of the target point set, and obtaining the shortest distance between each point in the point set and the opposite point set according to the distance matrix; summing up all the shortest distances to build an objective function; calculating the minimum of the objective function, wherein a parameter obtained is the registration parameter of the characteristic point set; conducting registration on the first image and the second image according to the obtained registration parameter. The image registration method has the advantages of reduced calculation scale, concise and efficient calculation and high robustness, and is especially suitable for different source image registrations.

Description

Method for registering images

Technical field

The present invention relates to image processing field, particularly, relate to a kind of method for registering images.

Background technology

Along with continuing to bring out of novel sensor, the ability that people obtain image improves rapidly, and the image that sensor produced of different physical features also is on the increase.Because the digital picture that the different images sensor obtains exists tangible limitation and otherness, so only utilize a kind of view data often to be difficult to practical requirement.Second image mainly is through obtaining the information of visible light part, have the resolution height, the characteristics clearly that form images, but be difficult to differentiate through camouflage and the object that covers, and relatively poor at the imaging effect of light dark place.Infrared image mainly is the information that receives infrared band through infrared sensor, and the thermal-radiating difference in can reflecting regional can disclose through camouflage and the object that covers, and can reflect the sight of dark place.For this reason, need can optical sensor and the infrared image synthesis that obtains be got up to use, reach, the accurately purpose of understanding and cognition more comprehensive, clear target through image registration techniques.

In the prior art, image registration techniques mainly is divided three classes: the method for finding the solution based on the method for half-tone information, based on the method and the transform domain of characteristic.Method for registering based on gray scale is directly to utilize the gray scale of image to measure the similarity between two width of cloth images, then, adopts the method searching of search to make similarity measurement maximum or minimum point, thereby confirms the transformation model parameter between two width of cloth images.The shortcoming that exists based on the method for registering of gray scale is that calculated amount is big, between two width of cloth images disappearance can not be arranged, and is comparatively responsive to convergent-divergent rotation and distortion.Based on the method for registering of characteristic, be on two width of cloth images, to extract characteristics of image respectively, like profile, angle point etc., two stack features that extract are carried out registration, the registration parameter of obtaining is the registration parameter of two width of cloth images.The method that transform domain is found the solution is to utilize the translation character of Fourier transform under the frequency and phase propetry to come the translation vector between the detected image, but this method is responsive for convergent-divergent and distortion.

Need a kind of new method for registering images, on the basis that reduces operand, keep very strong robustness, can also be applicable to the registration problems of different source images.

Summary of the invention

The object of the present invention is to provide a kind of based on method for registering minor increment sum between feature point set, that be particularly useful for the infrared image and second image; Pass through Corner Detection extract minutiae collection at the infrared image of treating registration and second image; Through making up objective function, registration problems is converted into the minimum problem of asking objective function based on minor increment sum between point set.Find the solution the parameter that parameter that this objective function obtains is registration, with this registration parameter to infrared image and second image enforcement registration.The present invention has reduced the scale of calculating, and has computing simple and high-efficient characteristics, has very strong robustness, is particularly suitable for the registration problems of different source images.

According to a main aspect of the present invention, a kind of method for registering images is provided, it comprises the steps:

(a) obtain the feature point set of first image and second image respectively;

(b) be the target point set with one of them feature point set, another feature point set is the point set that floats, and each point obtains shining upon point set through conversion in the point set that floats;

(c) calculate the distance matrix that shines upon between point set and each point of target point set, and, obtain the bee-line of interior each point of point set apart from the other side's point set according to distance matrix;

(d) all bee-lines are sued for peace, with this establishing target function;

(e) ask the minimal value of objective function, the parameter that obtains is the registration parameter of feature point set; And

(f) utilize the registration parameter of trying to achieve that first image and second image are carried out registration.

According to an aspect of the present invention, first image is an infrared image.

According to an aspect of the present invention, second image is a visible images.

According to an aspect of the present invention, adopt corner detection approach to obtain the feature point set of image.

According to an aspect of the present invention, corner detection approach comprises the Harris corner detection approach.

According to an aspect of the present invention, conversion comprises translation, rotation, convergent-divergent.

According to an aspect of the present invention, registration parameter comprises the parameter of translation, rotation, convergent-divergent.

Will be appreciated that the characteristic in the above each side of the present invention is independent assortment within the scope of the invention, and do not receive the restriction of its order---as long as the technical scheme after the combination drops in the connotation of the present invention.

Description of drawings

In order to be illustrated more clearly in the technical scheme among the present invention, will do to introduce simply to accompanying drawing of the present invention below, wherein:

Fig. 1 has shown the process flow diagram of the inventive method;

Embodiment

Hereinafter will combine the preferred embodiments of the present invention that technical scheme of the present invention is elaborated.

Need to understand that the description of hereinafter (comprising accompanying drawing) only is exemplary, but not the description of limitation of the present invention property.Can relate to the concrete quantity of parts in the following description, yet also need should be appreciated that, these quantity also only are exemplary, and those skilled in the art can choose the parts of right quantity with reference to the present invention arbitrarily.And wordings such as mentioned in the present invention " first ", " second " are not the ordering of expression to parts importance, only make the difference name of parts and are referred to as to use.

Fig. 1 has shown the process flow diagram of the inventive method.

In one embodiment of the invention, its key step is following:

At first, adopt Harris angular-point detection method extract minutiae collection respectively for same target infrared image and visible images.Angle point is the important local feature of image, and its definition directly perceived is at all bigger point of both direction epigraph grey scale change at least.At the angle point place, the shade of gray of image is discontinuous, and in the contiguous zone of angle point, gradient has two or more different values.To each pixel on the gray level image, calculate in horizontal and vertical first order derivative, and the product of the two, obtain three width of cloth new images like this.The corresponding property value of each pixel is represented g respectively on three width of cloth images_x, g_y, g_xg_yCarry out gaussian filtering for three width of cloth images, calculate the interest value of each corresponding point:

$M=G\left(\stackrel{\‾}{s}\right)\⊗\left[\begin{array}{cc}{g}_x& g_x{g}_y\\ g_x{g}_y& g_y\end{array}\right]---\left(1\right)$

I＝det(M)-k·tr²(M) (2)

Wherein, k is the weights coefficient, preferably gets 0.04, g_xBe the gradient of x direction, g_yBe the gradient of y direction,Be Gauss's template, det () is a determinant of a matrix, and tr () is a matrix trace.After the interest value of having calculated each point, extract the maximum point of all partial interest values of original image and be unique point.

Secondly, the feature point set that extracts with visible images is as target point set X={x₁, x₂..., x_n, the number of point is n in the point set; , the feature point set that extracts with infrared image is as unsteady point set Y={y₁, y₂..., y_m.The number of point is m in the point set.Point x_iWith y_jBetween distance be d (x_i, y_j)=|| x_i-y_j||, || || the tolerance of expression distance.Each some process translation (Δ y that the point set Y that floats is interior₁, Δ y₂), the rotation θ, conversion such as convergent-divergent α, by former coordinate space to target point set spatial mappings.Its mid point y_iBe designated as T (y through mapping_i), the point set Y that floats is designated as T (Y) through mapping.Wherein translation transformation does

$\left[\begin{array}{ccc}{y_1}^{*}& {y_2}^{*}& 1\end{array}\right]=\left[\begin{array}{ccc}{y}_1& y_2& 1\end{array}\right]\·\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ \mathrm{\Δ}{y}_1& \mathrm{\Δ}{y}_2& 1\end{array}\right]=\left[\begin{array}{ccc}{y}_1+\mathrm{\Δ}{y}_1& y_2+\mathrm{\Δ}{y}_2& 1\end{array}\right]---\left(3\right)$

Rotational transform does

$\left[\begin{array}{ccc}{y_1}^{*}& {y_2}^{*}& 1\end{array}\right]=\left[\begin{array}{ccc}{y}_1& y_2& 1\end{array}\right]\·\left[\begin{array}{ccc}\cos \mathrm{\θ}& \sin \mathrm{\θ}& 0\\ -\sin \mathrm{\θ}& \cos \mathrm{\θ}& 0\\ 0& 0& 1\end{array}\right]---\left(4\right)$

$=\left[\begin{array}{ccc}\cos \mathrm{\θ}\·y_1-\sin \mathrm{\θ}\·y_2& \sin \mathrm{\θ}\·y_1+\cos \mathrm{\θ}\·y_2& 1\end{array}\right]$

Scale transformation does

$\left[\begin{array}{ccc}{y_1}^{*}& {y_2}^{*}& 1\end{array}\right]=\left[\begin{array}{ccc}{y}_1& y_2& 1\end{array}\right]\·\left[\begin{array}{ccc}\mathrm{\α}& 0& 0\\ 0& \mathrm{\α}& 0\\ 0& 0& 1\end{array}\right]=\left[\begin{array}{ccc}\mathrm{\α}\·y_1& \mathrm{\α}\·y_2& 1\end{array}\right]---\left(5\right)$

Once more, will pass through interior 1 the T (y of mapping back point set T (Y)_i) be designated as { d (T (y to the distance of the interior each point of target point set X_i), x₁), d (T (y_i), x₂) ..., d (T (y_i), x_m), wherein minor increment is designated as mind (T (y_i), X).Equally, 1 x in the target point set X_jDistance to the interior each point of point set T (Y) is { d (T (y₁), x_j), d (T (y₂), x_j) ..., d (T (y_n), x_j), wherein minor increment is designated as min d (T (Y), x_j).

Calculate the bee-line of the interior each point of mapping point set and target point set, obtain two minor increments set, be respectively to the other side's point set

{ min d (T (y₁), X), min d (T (y₂), X) ..., min d (T (y_n), X) }, with

{min?d(T(Y)，x₁)，min?d(T(Y)，x₂)，…，min?d(T(Y)，x_m)}。

Then, the point set Y that will float is through transforming to mapping point set T (Y), and the matching degree between T (Y) and the target point set X adds up with two-way minor increment and measures, and sets up objective function

$f(\mathrm{\Δ}{y}_1,\mathrm{\Δ}{y}_2,\mathrm{\θ},\mathrm{\α})=\min (\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\min d(T\left(y_{c\left(i\right)}\right),X)+\underset{j=1}{\overset{m}{\mathrm{\Σ}}}\min d(T\left(Y\right),x_{c\left(j\right)}))---\left(6\right)$

Then, ask the minimal value of this function, confirm parameter (Δ y₁, Δ y₂, θ, α).

At last, according to the registration parameter of obtaining, carry out registration for infrared image and visible images.

Above basis has preferred embodiment been done detailed description to the present invention; But it will be appreciated that; Scope of the present invention is not limited to these concrete embodiments, but comprises that those skilled in the art are according to any change and the change that openly can make of the present invention.

Claims (7)

1. method for registering images, it comprises the steps:

(a) obtain the feature point set of first image and second image respectively;

(b) be the target point set with one of them feature point set, another feature point set is the point set that floats, and each point obtains shining upon point set through conversion in the point set that floats;

(c) calculate the distance matrix that shines upon between point set and each point of target point set, and, obtain the bee-line of interior each point of point set apart from the other side's point set according to distance matrix;

(d) all bee-lines are sued for peace, with this establishing target function;

(e) ask the minimal value of objective function, the parameter that obtains is the registration parameter of feature point set; And

(f) utilize the registration parameter of trying to achieve that first image and second image are carried out registration.

2. the method for registering images above-mentioned according to claim 1 is characterized in that, said first image is an infrared image.

3. the method for registering images above-mentioned according to claim 1 is characterized in that, said second image is a visible images.

4. the method for registering images above-mentioned according to claim 1 is characterized in that, adopts corner detection approach to obtain the feature point set of image.

5. the method for registering images above-mentioned according to claim 4 is characterized in that said corner detection approach comprises the Harris corner detection approach.

6. the method for registering images above-mentioned according to claim 1 is characterized in that said conversion comprises translation, rotation, convergent-divergent.

7. the method for registering images above-mentioned according to claim 1 is characterized in that said registration parameter comprises the parameter of translation, rotation, convergent-divergent.

Images