CN110852963B - A Correlation Maximum-Based Turbulence Degraded Image Restoration Method - Google Patents

Abstract

Translated fromChinese

本发明公开了一种基于相关最大性的湍流退化图像复原方法，包括：对采集的多幅湍流退化图像做平均及边缘增强处理，设为参考图像；将退化图像和参考图像分割为多个子模块图像，再将对应位置的子模块图像组成一个子图像集；计算每个子图像集中的每一子模块图像与其参考图像集中的子模块图像之间的欧几里得距离，将小于中值距离的子模块图像再组成新图像集；由新图像集，基于主成分分析法和图像最大相似性特征，得到最终复原的子模块图像；再将最终复原的子模块图像，按照提取子图像集对应的位置组成一幅图像，即为得到的复原图像。本方法将采集的多帧图像和与之对应的参考图像分割成多个模糊图像空不变的子模块图像集，适用于实际的湍流退化图像。

The invention discloses a turbulent degraded image restoration method based on correlation maximization. image, and then group the sub-module images at the corresponding positions into a sub-image set; calculate the Euclidean distance between each sub-module image in each sub-image set and the sub-module image in the reference image set, which will be smaller than the median distance. The sub-module images are then formed into a new image set; the final restored sub-module image is obtained from the new image set based on the principal component analysis method and the maximum similarity feature of the image; then the final restored sub-module image is extracted according to the corresponding sub-image set. The positions form an image, which is the resulting restored image. The method divides the collected multi-frame images and the corresponding reference images into multiple sub-module image sets with blurred image space unchanged, which is suitable for actual turbulent degraded images.

Description

Turbulence degradation image restoration method based on correlation maximum

Technical Field

The invention relates to a degraded image restoration method, in particular to an atmospheric turbulence image restoration problem, and belongs to the field of multi-frame turbulence image restoration.

Background

In a medium-long distance imaging system, due to the atmospheric turbulence phenomenon caused by the influence of wind speed, temperature and the like, the refractive index of a propagation medium is irregularly changed, so that optical waves are distorted when propagating in the medium, and the geometric deformation and the blurring of an acquired image are caused. Therefore, effectively restoring the original target image from the turbulence-degraded image is one of the key issues in achieving the processing for target detection, recognition, and the like.

The commonly used method for restoring the turbulence degradation image is mainly based on the restoration method of single-frame and multi-frame turbulence images. Due to strong randomness of turbulence, the fuzzy degree of the image acquired by the static object has uncertainty in time and space, and the restoring effect based on the complementarity of the multi-frame image is better than that based on the restoring result of the single-frame image. Therefore, a restoration method based on a plurality of frames of images is often used, and the main methods proposed at present include:

the possibility of "lucky images" in a large number of acquired short-exposure images is exploited for restoration. The method comprises the steps of firstly registering an acquired turbulence image by using an image registration method, eliminating geometric deformation, extracting a high-quality image from a plurality of frames of registered images, fusing the images into one image, and processing the fused image based on an image deconvolution or image enhancement mode to improve the definition of the image. For example, patent CN103310486B discloses an atmospheric turbulence degradation image reconstruction method. The technical scheme is that a multi-frame registration is firstly carried out to eliminate a distorted image, then a diffraction fuzzy image based on space-time neighbor combination is reconstructed, and finally a global uniform deconvolution is adopted to eliminate the diffraction fuzzy. The method has complex calculation and long time consumption, and the final recovery effect is directly influenced by the registration precision.

Estimating an initial value of a point spread function by using a multi-frame image, and obtaining a restored image by deconvolution on a time domain or a space domain through continuous cyclic iteration, such as a maximum likelihood estimation iteration blind deconvolution restoration method, a Richardson-Lucy algorithm and the like. However, the noise cannot be effectively removed by estimating the point spread function and then deconvolving, so that the ringing effect is easily generated in the recovery result.

Li, R.M.Meraereau et al propose a turbulent degraded image restoration algorithm based on principal component analysis, which restores the degraded image based on the idea of high frequency component enhancement, and obtains a restored image by using the first principal component with the largest variance as the largest high frequency component (IEEE geosci.Remote S.2007; 4(3): 340-. The method is essentially a blind deconvolution recovery method based on a principal component analysis method, has the characteristics of high speed and strong noise resistance, and can be used for not only multi-frame degraded image recovery but also single-frame degraded image recovery. Since the feature decomposition used in the principal component analysis method has uncertainty, the direction consistency between the calculated high-frequency component and the mean image cannot be ensured, and an expected restored image may not be obtained. In addition, the method is to regard the whole degraded image as a blurred space-invariant image, and is not suitable for an actual turbulent degraded image.

Disclosure of Invention

The invention provides a turbulence degradation image restoration method based on correlation maximization, which has the basic idea that a plurality of acquired images are divided into a plurality of fuzzy image space-invariant sub-module image sets, a finally restored sub-module image is obtained based on a principal component analysis method and the maximum similarity characteristic of the images, and the restoration of a turbulence image is realized, and the details are described in the following:

a method for turbulence degradation image restoration based on correlation maximization, the method comprising the steps of:

averaging and edge enhancement processing are carried out on the collected multiple turbulence degradation images, and the multiple turbulence degradation images are set as reference images;

dividing the degraded image and the reference image into a plurality of sub-module images, and forming a sub-image set by the sub-module images at corresponding positions;

calculating the Euclidean distance between each sub-module image in each sub-image set and the sub-module images in the reference image set, and recombining the sub-module images smaller than the median distance into a new image set;

obtaining a final restored sub-module image from the new image set based on a principal component analysis method and the maximum similarity characteristic of the image; and then, forming an image by the finally restored sub-module image according to the position corresponding to the extracted sub-image set, namely, obtaining the restored image.

The step of obtaining the finally restored sub-module image by the new image set based on the principal component analysis method and the maximum image similarity characteristic specifically comprises the following steps:

converting the A number of sub-module image matrixes in the new image set into a column matrix with the size of B, and forming a B multiplied by A matrix C by all the column matrices;

calculating a mean value column matrix of the matrix C, and subtracting each column matrix in the new image set from the mean value column matrix to obtain a matrix X;

and acquiring a result row matrix according to the matrix X, the mean value column matrix and the weight coefficient, converting the result row matrix into a matrix with the size corresponding to the size of the sub-module image matrix, namely a result sub-image, and taking the result sub-image as a finally restored sub-module image.

The obtaining of the row matrix according to the matrix X, the mean image, and the weight coefficient specifically includes:

and obtaining a product of the weight coefficient and the transposition of the matrix X, wherein the product and a modulus of the weight coefficient are subjected to quotient, and the quotient is added with the transposition of the mean value column matrix to obtain a row matrix.

In addition, when processing a set of images that are greatly affected by turbulence, the method further comprises:

setting the restored sub-images as new reference images, calculating the Euclidean distance between each sub-module image in the new image set and the reference images, recombining the sub-module images smaller than the median distance into a new sub-module image set, and processing the image set to obtain the restored sub-module images. The processing is performed a plurality of times in sequence, the number of repetitions depending on the degraded image being processed.

The technical scheme provided by the invention has the beneficial effects that:

1. the method combines the maximum similarity characteristics of the principal component and the original image, ensures the direction consistency of the calculated high-frequency component and the mean image, and obtains an expected restored image;

2. the method divides the collected multi-frame image and the reference image corresponding to the multi-frame image into a plurality of sub-module image sets with fuzzy images unchanged, and is particularly suitable for actual turbulence degradation images.

Drawings

FIG. 1 is a flow chart of a method for restoring a turbulence degradation image based on correlation maximization according to the present invention;

FIG. 2 is a detailed flowchart of a method for restoring a degraded turbulent image based on correlation maximization according to the present invention;

FIG. 3 is a set of 15 simulated turbulence-affected lunar surface images, reference images, and their segmented sub-module images in an example of the present invention;

wherein (a) is simulated 15 turbulence degradation images; (b) 16 sub-image sets which are obtained by dividing and combining 15 degraded images; (c) a reference image obtained from 15 turbulence degraded images; (d) 16 sub-images are segmented for the reference image.

Fig. 4 is a restored image of an example of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in further detail below.

The following explains the specific processing procedure of the degraded image restoration method of the present invention, taking the degraded image shown in fig. 3 as an example;

step 1, simulating 15 lunar surface images g affected by turbulence_i(i is 1,2, …,15, 256 × 256 pixels in size), and the average image is calculated as shown in fig. 3(a)

Performing enhancement processing on the mean image phi by using a laplacian filter, and setting the mean image phi as a reference image psi as shown in fig. 3 (c);

step 2, for each g_iThe image and the reference image Ψ are divided into 16 sub-module images having a size of 106 × 106 pixels, and the interval between two adjacent sub-modules is nexp — 50 pixels, as shown in fig. 3 (a). Then the sub-module images at the corresponding positions are combined into a sub-image set which is respectively marked as G_jAnd Ψ_jWherein j is 1,2, …,16, each G_jContains 15 sub-module images, each Ψ is shown in FIG. 3(b)_jContains only one sub-block image, as shown in fig. 3 (d);

step 3, for G₁Calculates the sum of Ψ and Ψ for each sub-module image in the image₁Euclidean distance z of_1,iThen will be

z_1,i＜median(z₁) The sub-module images form a new image set, which is recorded as

Contains 7 sub-module images, wherein, mean (z)₁) For the calculated Euclidean distance z_1,iThe median value of (1);

step 4, mixing

The 7 submodule image matrixes in the process are converted into an 11236 multiplied by 1 column matrix, and then all the column matrices are combined into a 11236 multiplied by 7 matrixMatrix array

Step 5, calculating the matrix obtained in the step 4

Mean value image of

Then will be

Each column matrix in

Phi and phi₁Subtracting to obtain a matrix X₁＝[x_1,1,x_2,1,…,x_7,1]Wherein p is 1,2, …, 7;

step 6, the formula

Calculating Y₁Wherein, in the step (A),

is X₁The transpose matrix of (a) is,

is phi₁The transposed matrix of (2), the weight coefficient mu₁Comprises the following steps:

the obtained row matrix Y₁Conversion into a matrix of 106 x 106 size, i.e. the resulting sub-image f_g1. The result is sub-image f_g1As the final restored sub-module image f_g1；

Step 7, then for G₂，G₃，...，G₁₆In turn, a method ofProcessing the image by the steps 3 to 6 to obtain a corresponding finally restored sub-image f_g2，f_g3，...，f_g16；

Step 8, recombining the image into an image according to the position corresponding to the extracted sub-image set, and superposing and averaging the images of the superposed part to obtain a restored image f_gAs shown in fig. 4.

Those skilled in the art will appreciate that the drawings are only schematic illustrations of preferred embodiments, and the above-described embodiments of the present invention are merely provided for description and do not represent the merits of the embodiments.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (3)

1. A method for restoring a turbulence degradation image based on correlation maximization, the method comprising the steps of:

averaging and edge enhancement processing are carried out on the collected multiple turbulence degradation images, and the multiple turbulence degradation images are set as reference images;

dividing the degraded image and the reference image into a plurality of sub-module images, and forming a sub-image set by the sub-module images at the same position;

calculating the Euclidean distance between each sub-module image in each sub-image set and the sub-module images in the reference image set, and recombining the sub-module images smaller than the median distance into a new image set;

obtaining a final restored sub-module image from a new image set based on a principal component analysis method and the maximum similarity characteristic of the image, and forming an image from the finally obtained sub-module image according to the position corresponding to the extracted sub-image set, namely the obtained restored image;

the method for obtaining the final restored sub-module image by the new image set based on the principal component analysis method and the maximum image similarity characteristic specifically comprises the following steps:

converting the A number of sub-module image matrixes in the new image set into a column matrix with the size of B, and forming a B multiplied by A matrix C by all the column matrices;

calculating a mean value column matrix of the matrix C, and subtracting each column matrix in the new image set from the mean value column matrix to obtain a matrix X;

and acquiring a result row matrix according to the matrix X, the mean value column matrix and the weight coefficient, converting the result row matrix into a matrix with the size corresponding to the size of the sub-module image matrix, namely a result sub-image, and taking the result sub-image as a finally restored sub-module image.

2. The method for restoring turbulence degradation images based on correlation maximum as claimed in claim 1, wherein the obtaining of the row matrix according to the matrix X, the mean column matrix, and the weight coefficient specifically includes:

and obtaining a product of the weight coefficient and the transposition of the matrix X, wherein the product and a modulus of the weight coefficient are subjected to quotient, and the quotient is added with the transposition of the mean value column matrix to obtain a row matrix.

3. A method of turbulence degradation image restoration based on correlation maximum as claimed in any one of claims 1-2, wherein when processing image sets that are greatly affected by turbulence disturbance, the method further comprises:

setting the restored sub-images as new reference images, calculating Euclidean distances between each sub-module image in the new image set and the reference images, recombining the sub-module images smaller than the median distance into a new sub-module image set, processing the image set to obtain restored sub-module images, and sequentially processing for multiple times, wherein the number of times of repetition is determined according to the processed degraded images.

Images