A SAR image registration method using image block matching

Abstract

The present invention discloses a SAR image registration method using image block matching. The two images to be matched are divided into blocks and matched respectively to obtain a position deviation matrix and an amplitude mean ratio matrix. A threshold is set to eliminate the singular values of the two matrices, reducing the matching error. The SAR image is multiplied by the coefficient of the amplitude mean ratio, and the SAR image is interpolated according to the parameters of the position deviation. Compared with other image registration methods, it has strong robustness, realizes pixel-level registration, and has high operation efficiency, providing a prerequisite guarantee for the detection, tracking and positioning of moving targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and particularly relates to an image registration technology. Background Art

[0002] The main function of video SAR is to detect and track ground moving targets. The premise of using change detection to achieve moving target detection is high-precision image registration. After the SAR images are registered with high precision, moving targets are detected through change detection, and then moving target tracking is realized based on the detection results.

[0003] The registered multi-frame SAR images can achieve high-precision estimation of the speed of moving targets, and can also achieve image stitching and multi-view image fusion. Image registration is one of the important research contents of SAR image processing.

[0004] The existing registration methods for SAR images include:

[0005] In the "Synthetic Aperture Radar Image Registration Method Using SAR-FAST Corner Detection", a method for matching SAR images using corners is disclosed. This method first suppresses speckle noise, then extracts and screens corner regions to obtain a stable and repeatable matching technique.

[0006] In the "SAR Image Registration Method Based on Point Features under Full Affine Deformation", an image registration method based on a radiometric deformation model is disclosed. This method first uses a particle swarm optimization algorithm to estimate the scale transformation parameters of the image, and then uses scale-invariant features and the transformation operator SIFT to register the image.

[0007] In the "A New Algorithm for SAR Image Registration Based on Local Invariant Features", an image registration method is disclosed. This method first uses the accelerated segment test feature FAST to detect image corners, secondly uses the DAISY descriptor to perform invariant description of the FAST features, then uses the KD tree, Euclidean distance, and RANSAC algorithm to achieve image registration, and finally uses affine transformation to achieve image interpolation and image transformation.

[0008] In the existing SAR image registration methods, feature point extraction algorithms such as SIFT and SURF are mostly used, and complex feature screening methods are used to screen matching points, resulting in low computational efficiency and many mismatching situations. An efficient SAR image registration method is needed on the basis of ensuring the image registration accuracy. Summary of the Invention

[0009] In order to solve the problems existing in the prior art, the present invention proposes a SAR image registration method using image block matching. Through the block registration of SAR images, the position offsets of different regions of the SAR image are obtained, and high-precision registration of the SAR image is achieved through interpolation, providing guarantee for the detection and tracking of moving targets. To achieve the above object, the present invention adopts the following technical solutions.

[0010] The two images to be matched are respectively block-matched to obtain a position deviation matrix and an amplitude mean ratio matrix. A threshold is set to eliminate the singular values of the two matrices, reducing the matching error. The SAR image is multiplied by the coefficient of the amplitude mean ratio, and the SAR image is interpolated according to the parameters of the position deviation.

[0011] Further, the two images are blocked according to a certain size, and the fast Fourier transform is used to calculate the correlation coefficient between the blocks. The position corresponding to the maximum value in the correlation coefficient matrix is selected, and combined with the block size, the position deviation of the block is calculated. The corresponding blocks of the two images are matched, the position deviations of all blocks of the two images are calculated, a row position deviation matrix and a column position deviation matrix are generated, the amplitude mean ratio between the two images is calculated, and an amplitude mean ratio matrix is generated.

[0012] The purpose of block-matching the two images respectively is to calculate the position deviation and amplitude deviation between all blocks of the two images. The position deviation includes row position deviation and column position deviation.

[0013] Let the row and column sizes of the main and auxiliary SAR images be N r and N c , the row and column sizes of the image blocks after blocking are N bk , the matching step number is N step , the image block at the i-th row and j-th column of the main image I1 is I1 (i,j) , the image block at the i-th row and j-th column of the auxiliary image I2 is I2 (i,j) , the correlation coefficient matrix of the image blocks I1 (i,j) and I2 (i,j) is ρ, FFT2 is the two-dimensional Fourier transform, IFFT2 is the two-dimensional inverse Fourier transform, conj(g) is the conjugate calculation, then ρ = IFFT2[FFT2(I1 (i,j) )·conj(FFT2(I2 (i,j) ))], the row and column positions corresponding to the maximum value of ρ are [a, b] = arcmax(ρ). Let the matrices M r and M c be the row position deviation and column position deviation of each image block of the main and auxiliary images respectively, then the position deviation of the image block at the i-th row and j-th column is Let the matrix η A be the amplitude mean ratio of each image block of the main and auxiliary images, then the amplitude mean ratio of the image block at the i-th row and j-th column is

[0014] η A (i,j) = mean(I1 (i,j) ) / mean(I2 (i,j) )。

[0015] The position deviation and amplitude mean ratio are calculated for each image patch, achieving the matching between image patches, and obtaining the row position deviation matrix, column position deviation matrix, and amplitude mean ratio matrix between the main and auxiliary images.

[0016] Furthermore, compare the median of each element in the row and column position deviation matrices and the amplitude mean ratio matrix with the eight connected elements around it. If the absolute value of the difference exceeds the threshold, reassign it to the median; otherwise, the value remains unchanged. Repeat the assignment three times to obtain the row and column position deviation matrices and the amplitude mean ratio matrix with outliers removed.

[0017] There is a certain probability of error in the image patch matching calculation, resulting in large errors in the row and column position deviations, that is, the outliers of the position deviation. It is necessary to remove the outliers in the matrix matching to ensure the image matching accuracy.

[0018] Let the position deviation threshold be T dif , and the median of the eight connected elements of the element in the i-th row and j-th column of the row and column position deviation matrix be V r (i,j) and V c (i,j), use and to remove the outliers of the row and column position deviation matrix. Let the amplitude mean ratio deviation threshold be T η , and let the median of the eight connected elements of the element in the i-th row and j-th column of the amplitude mean ratio matrix be V η (i,j), use to remove the outliers of the amplitude mean ratio matrix.

[0019] Repeat the calculation three times, which is sufficient to eliminate the outliers of the position deviation matrix and the amplitude mean ratio matrix.

[0020] Furthermore, perform Gaussian smoothing on the position deviation matrix and the amplitude mean ratio matrix, interpolate the row and column position deviation matrices and the amplitude mean ratio matrix to make the sizes of the position deviation matrix and the amplitude mean ratio matrix the same as the size of the SAR image. Multiply the amplitude mean ratio matrix by the corresponding elements of the SAR image to correct the amplitude of the SAR image. Interpolate the SAR image according to the parameters in the row and column position deviation matrix, and use the cubic spline interpolation method to obtain the matched SAR image.

[0021] Perform Gaussian smoothing filtering on the position deviation matrices M r and M c to reduce the position registration error, and interpolate to obtain M r ′ and M cThe row and column dimensions of ′ are the same as those of the original image, and the amplitude mean ratio matrix η A is subjected to Gaussian smoothing filtering to reduce the amplitude registration error, and η′ is obtained by interpolation A The row and column dimensions of are the same as those of the original image, and the amplitude I of the auxiliary image I2 is corrected by the amplitude ratio of η′ A 2amp , multiply the elements of the matrix of I2 and η′ A , that is, I 2_amp = I2 * η′ A , two-dimensional interpolation is performed according to the dimensions of M r ′ and M c ′ to achieve image registration of images I1 and I2.

[0022] Advantages of the present invention: Obtain the position deviation and amplitude mean ratio of each region of the image by image block matching, remove the values with large matching errors through singular value rejection and numerical smoothing, obtain high-precision position deviation parameters and amplitude mean ratio, and achieve high-precision image registration through image interpolation. Compared with other image registration methods, the present invention has strong robustness, realizes pixel-level registration, and has high operation efficiency, providing a prerequisite guarantee for the detection, tracking and positioning of moving targets. Description of the Drawings

[0023] Figure 1 is the SAR main image.

[0024] Figure 2 is the SAR auxiliary image.

[0025] Figure 3 is the difference image of the main and auxiliary images.

[0026] Figure 4 is the amplitude map of the correlation coefficient matrix.

[0027] Figure 5 is the amplitude map of the row position deviation matrix.

[0028] Figure 6 is the amplitude map of the column position deviation matrix.

[0029] Figure 7 is the amplitude map of the amplitude mean ratio matrix.

[0030] Figure 8 is the amplitude map of the row position deviation matrix after singular value rejection.

[0031] Figure 9 is the amplitude map of the column position deviation matrix after singular value rejection.

[0032] Figure 10 is the registered SAR auxiliary image.

[0033] ​Figure 11 It is the differential image of the registered main and auxiliary images. Specific implementation mode

[0034] The technical solution of the present invention will be specifically described below with reference to the accompanying drawings.

[0035] Select 2 SAR radar images at different times, and the image size is N r = 2000 and N c = 2000, and the row and column sizes of the image blocks are both N bk = 300, and the step number during image block matching is N step = 100.

[0036] The main image and the auxiliary image are respectively as Figure 1 and Figure 2 shown. There are many residues in the absolute value of the difference between the two images, as Figure 3 shown. The amplitude of the matching correlation coefficient matrix obtained by block matching is as Figure 4 shown. Calculate the amplitudes of the row position deviation matrix and the column position deviation matrix as Figure 5 and Figure 6 shown. Calculate the average amplitude ratio of the image blocks as Figure 7 shown.

[0037] Let T dif = 3, T η = 0.1, and repeat the calculation of matrix M r 、M c and η A a total of 3 times to ensure that all singular values in the matrix are eliminated, and obtain the row and column position deviation matrix and the average amplitude ratio matrix after elimination, as Figure 8 and Figure 9 shown. There are no singular values in the average amplitude ratio matrix, so there is no change.

[0038] Interpolate matrix M r 、M c and η A respectively to make the size the same as that of the main and auxiliary images, and obtain the auxiliary image after amplitude correction. Use the cubic spline interpolation method on the auxiliary image to obtain the SAR image after image registration, as Figure 10 shown.

[0039] Take the difference between the main and auxiliary images and then take the absolute value to obtain the differential image, as Figure 11 shown. From the comparison between Figure 11 and Figure 3 it can be seen that after image registration, the difference between the images is significantly reduced.

[0040] The above embodiments of the present invention do not limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included in the protection scope of the present invention.

Claims

1. A SAR image registration method using image block matching, characterized in that, Including: The two images to be matched are segmented and matched respectively. The two images are segmented according to a certain size, and the fast Fourier transform is used to calculate the correlation coefficient between the segments. The position corresponding to the maximum value in the correlation coefficient matrix is selected, and combined with the segment size, the position deviation of the segment is calculated. The corresponding segments of the two images are matched, the position deviations of all segments of the two images are calculated, a row position deviation matrix and a column position deviation matrix are generated, the amplitude mean ratio between the two images is calculated, and an amplitude mean ratio matrix is generated. Let the row and column sizes of the main and auxiliary SAR images be N r and N c , the row and column sizes of the image blocks after segmentation are N bk , the matching step number is N step , the image block at the i-th row and j-th column of the main image I1 is I1 (i,j) , the image block at the i-th row and j-th column of the auxiliary image I2 is I2 (i,j) , the correlation coefficient matrix of the image blocks I1 (i,j) and I2 (i,j) is ρ, FFT2 is the two-dimensional Fourier transform, IFFT2 is the two-dimensional inverse Fourier transform, and conj(g) is the conjugate calculation. Then ρ = IFFT2[FFT2(I1 (i,j) )·conj(FFT2(I2 (i,j) ))], the row and column positions corresponding to the maximum value of ρ are [a, b] = arcmax(ρ). Let the matrices M r and M c be the row position deviation and column position deviation of each image block of the main and auxiliary images. Then the position deviation of the image block at the i-th row and j-th column is Let the matrix η A be the amplitude mean ratio of each image block of the main and auxiliary images. Then the amplitude mean ratio of the image block at the i-th row and j-th column is η A (i, j) = mean(I1 (i,j) ) / mean(I2 (i,j) ). The position deviation matrix and the amplitude mean ratio matrix are obtained, a threshold is set, the singular values of the two matrices are removed, the SAR image is multiplied by the coefficient of the amplitude mean ratio, and the SAR image is interpolated according to the parameters of the position deviation.

2. The SAR image registration method using image block matching according to claim 1, characterized in that The singular value rejection of the two matrices includes: comparing each element in the row and column position deviation matrix and the amplitude mean ratio matrix with the median of the eight surrounding connected elements. If the absolute value of the difference exceeds the threshold, it is reassigned to the median; otherwise, the value remains unchanged. The assignment is repeated three times to obtain the row and column position deviation matrix and the amplitude mean ratio matrix with singular values removed.

3. The SAR image registration method using image block matching according to claim 2, characterized in that The singular values of the two matrices are removed, and it further includes: setting the position deviation threshold as T dif , the median of the eight connected elements of the element in the i-th row and j-th column of the row-column position deviation matrix is V r (i, j) and V c (i, j), and using and Eliminate the singular values of the row-column position deviation matrix, and set the amplitude mean ratio deviation threshold as T η , and set the median of the eight connected elements of the element in the i-th row and j-th column of the amplitude mean ratio matrix as V η (i, j), and use to eliminate the singular values of the amplitude mean ratio matrix.

4. The SAR image registration method using image block matching according to claim 1, characterized in that, The interpolation of the SAR image includes: performing Gaussian smoothing on the position deviation matrix and the amplitude mean ratio matrix, interpolating the row and column position deviation matrix and the amplitude mean ratio matrix to make their sizes the same as that of the SAR image, multiplying the amplitude mean ratio matrix by the corresponding elements of the SAR image to correct the amplitude of the SAR image, and interpolating the SAR image according to the parameters in the row and column position deviation matrix using the cubic spline interpolation method to obtain the matched SAR image.

5. The SAR image registration method using image block matching according to claim 4, characterized in that, The interpolated SAR image further includes: performing Gaussian smoothing filtering on the position deviation matrices M r and M c to reduce the position registration error, and interpolating to obtain M r ′ and M c ′ with the same row and column dimensions as those of the original image. Performing Gaussian smoothing filtering on the amplitude mean ratio matrix η A to reduce the amplitude registration error, and interpolating to obtain η′ A with the same row and column dimensions as those of the original image. Correcting the amplitude I A of the auxiliary image I2 using the amplitude ratio of η′ 2_amp , multiplying the elements of I2 and η′ A matrix by matrix, i.e., I 2_amp = I2 * η′ A , and performing two-dimensional interpolation according to the dimensions of M r ′ and M c ′.

Images