A forward-looking sonar image fusion method based on support transformation and sparse representation

Abstract

The present application relates to a kind of based on support transformation and sparse representation forward-looking sonar image fusion method, belong to digital image processing technical field, the steps of the present application are as follows;Step one, analysis forward-looking sonar original data, calculate to obtain fan-shaped sonar image;Step two, the image of step one is decomposed into multistage support image sequence and low-frequency component image;Step three, according to each level support image sequence, train each level image dictionary sequence;Step four, using the multistage image dictionary sequence trained in step three, the image decomposed in step two is sparsely represented according to each level support image sequence, and the sparse coefficient of each level support image sequence is obtained;Step five, the sparse coefficient of each level support image sequence obtained in step four is fused, and the fused coefficient is restored as support image sequence;Step six, the image fused in step five is reconstructed.The device can remove the edge joint produced in image and splicing well.

Description

Technical Field

[0001] This invention relates to a forward-looking sonar image fusion method based on support degree transformation and sparse representation, belonging to the field of digital image processing technology. Background Technology

[0002] Image fusion refers to the process of combining image data of the same target acquired from multiple sources using image processing and computer technology. This process maximizes the extraction of useful information from each source and synthesizes the data into a high-quality image. This improves the utilization of image information, enhances the accuracy and reliability of computer interpretation, and increases the spatial and spectral resolution of the original image, thus facilitating monitoring. The images to be fused are pre-registered and have consistent pixel widths. The process integrates and extracts information from two or more multi-source images and is widely used in remote sensing, aerospace, virtual reality, medical imaging, and 3D reconstruction.

[0003] Existing forward-looking sonar image fusion algorithms refer to popular multi-scale fusion algorithms in optical image fusion, performing downsampling operations on the images, such as NSCT and Laplacian pyramid. However, acoustic images themselves have low imaging quality, and the details that the image can reflect after downsampling are even less complete. In addition, the influence of interpolation operations may lead to the sampling points being interpolation points, affecting the fusion effect. Furthermore, existing image fusion algorithms do not fully consider the impact of sonar image noise on the fusion effect, and the fusion result is often still very noisy, so the fusion quality needs to be improved.

[0004] This invention proposes a forward-looking sonar image fusion method based on a combination of support transformation and sparse representation. The support transformation avoids the downsampling operations performed by pyramid-type multi-scale fusion algorithms on acoustic images where the target is already unclear. The sonar data image is divided into support sequence images and low-frequency component images. A learning dictionary for the forward-looking sonar support sequence images is constructed using the classic dictionary learning method KSVD. The OMP algorithm, combined with the learning dictionary, is used to perform sparse representation of the support sequence images. Then, appropriate fusion strategies are applied to the sparse coefficients of the support sequence images and the low-frequency component images to enhance target details and edges, reduce noise, and retain low-frequency component energy, achieving better fusion results. Finally, the image is reconstructed through inverse support transformation to obtain the final fused result. Summary of the Invention

[0005] This invention addresses the challenges of reducing noise in image fusion algorithms, enhancing image target details and edges, and improving the fusion performance of forward-looking sonar images. Therefore, it proposes a forward-looking sonar image fusion method based on support degree transformation and sparse representation.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: The forward-looking sonar image fusion method based on support degree transformation and sparse representation described in the present invention is implemented through the following steps;

[0007] Step 1: Analyze the raw forward-looking sonar data and calculate the sector sonar image;

[0008] Step 2: Decompose the image obtained in Step 1 into a multi-level support image sequence and a low-frequency component image;

[0009] Step 3: Train image dictionary sequences for each level based on the image sequences of each level of support.

[0010] Step 4: Using the image dictionary sequences trained in Step 3, perform sparse representation of the images decomposed in Step 2 based on the support image sequences at each level, and obtain the sparsity coefficients of the support image sequences at each level.

[0011] Step 5: Fuse the sparse coefficients of the support image sequences at all levels obtained in Step 4, and restore the fused coefficients to the support image sequence;

[0012] Step 6: Reconstruct the image after fusion in Step 5.

[0013] Furthermore, in step two, when decomposing the image, the support transformation theory is used to decompose the forward-looking sonar into a multi-level support image sequence and a low-frequency component image.

[0014] Furthermore, in step three, a sparse representation method is used to train the image dictionary sequences at each level.

[0015] Furthermore, in step five, the maximum activity criterion is used to fuse the sparse coefficients of the support image sequences at each level.

[0016] Furthermore, in step five, the low-frequency component images are fused using a criterion that combines regional variance and regional energy.

[0017] Furthermore, in step six, the inverse support transform algorithm is used to reconstruct the fused image.

[0018] The beneficial effects of this invention are as follows: This invention utilizes forward-looking sonar parameters to construct a mapping model between expression domains and guides the stitching of consecutive frames based on measured data. Compared with existing forward-looking sonar fan-shaped image stitching algorithms based on the image domain, this invention can effectively remove edge seams generated during image stitching, eliminate spurious target values ​​that occur during fan-shaped unfolding interpolation stitching, and significantly improve the stitching effect, demonstrating superior stitching advantages. Attached Figure Description

[0019] Figure 1 It is a support transformation decomposition graph

[0020] Figure 2 This is a comparison chart of support transformation hierarchical decomposition.

[0021] Figure 3 It is a multi-level learning dictionary

[0022] Figure 4 This is a reference image for subjective evaluation of noise reduction.

[0023] Figure 5 These are the fusion results and noise reduction effect images. Detailed Implementation

[0024] Specific implementation method one: Combining Figures 1 to 5 This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation, which is implemented through the following steps;

[0025] Step 1: Analyze the raw forward-looking sonar data and calculate the sector sonar image;

[0026] Step 2: Decompose the image obtained in Step 1 into a multi-level support image sequence and a low-frequency component image;

[0027] Step 3: Train image dictionary sequences for each level based on the image sequences of each level of support.

[0028] Step 4: Using the image dictionary sequences trained in Step 3, perform sparse representation of the images decomposed in Step 2 based on the support image sequences at each level, and obtain the sparsity coefficients of the support image sequences at each level.

[0029] Step 5: Fuse the sparse coefficients of the support image sequences at all levels obtained in Step 4, and restore the fused coefficients to the support image sequence;

[0030] Step 6: Reconstruct the image after fusion in Step 5.

[0031] The above methods are used to fuse forward-looking sonar images.

[0032] Specific Implementation Method Two: Combining Figures 1 to 5 This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation. In step two, when decomposing the image, support transformation theory is used to decompose the forward-looking sonar into a multi-level support image sequence and a low-frequency component image. Each level of the support image sequence can separately represent the image target and background, better reflecting the characteristics of each part of the image.

[0033] Specific implementation method three: Combining Figures 1 to 5This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation. Step three employs sparse representation to train image dictionary sequences at various levels. Sparse representation can denoise image sequences at various support levels, effectively preserving the detailed information of the target portion of each image while removing background noise. The denoising effect is particularly good for high-level support images.

[0034] Specific implementation method four: Combination Figures 1 to 5 This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation. In step five, the maximum activity criterion is used to fuse the sparse coefficients of the support image sequences at each level. The maximum activity fusion criterion better reflects the detailed content of the image targets.

[0035] Specific Implementation Method Five: Combining Figures 1 to 5 This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation. In step five, the low-frequency component image is fused using a criterion combining regional variance and regional energy. This fusion rule, combining maximum regional variance and regional energy, maximizes the retention of useful low-frequency information and image energy in the image, ensuring the fusion quality of the image after the inverse support transformation.

[0036] Specific Implementation Method Six: Combination Figures 1 to 5 This embodiment describes a forward-looking sonar image fusion method based on support transformation and sparse representation. In step six, the inverse support transformation algorithm is used to reconstruct the fused image.

[0037] Example

[0038] This invention achieves sonar image fusion through the following steps;

[0039] Step 1: Adjust the two images by referring to the displacement and rotation between the forward-looking sonar images to be fused, and align the target details in the two images;

[0040] Step 2: Based on the principle of support transformation algorithm, the image is decomposed according to formula (1) and formula (2) to achieve support decomposition.

[0041] S j =SV j *P j 1

[0042] P j+1 =P j -S j ,j=1,2,…,r,P1=P2

[0043] In formula (2), r represents the number of decomposition layers, P is the given image, SV represents the support filter bank, * represents the convolution operation, and the support sequence obtained by support decomposition of P is {S1,S2,…,S…}. r}, the last P j+1 The image shows the low-frequency components after decomposition. Support transform decomposition is as follows: Figure 1 As shown, the comparison charts at each level are decomposed as follows: Figure 2 As shown.

[0044] Step 3: Perform coefficient decomposition on the support image sequences at each level, treating the images as vectors in an N-dimensional Euclidean space. Sparse representation can be expressed as:

[0045]

[0046] In formula (3), For a set of orthogonal bases, i.e., through a complete dictionary, d i Atoms from the dictionary; The sparse representation of x is called the sparse coefficient, which is the unique linear representation α reconstructed from the dictionary. i The vector formed; thus, the image x is sparsely decomposed into a combination of an overcomplete dictionary D and sparse coefficients α; multi-level learning dictionaries such as Figure 3 As shown.

[0047] Step 4: Using the multi-level sparse dictionary generated above, perform sparse decomposition on the support sequence images of each level of the multi-frame images in the dataset that have undergone support decomposition, and solve for their respective sparse coefficients.

[0048] Step 5: Merge the sparse coefficients of the support sequence images at each level of the image to be fused according to the principle of maximum activity.

[0049]

[0050] In formula (4), V i =||Coef i ||1 indicates activity level, Coef i Let represent the sparse coefficient matrix of the sparse decomposition of the i-th level support image sequence.

[0051] Step 6: Fuse the low-frequency components of the image to be fused by combining regional variance and regional energy.

[0052]

[0053] In formula (5), image 1iLet represent the i-th neighborhood of the low-frequency component image 1 to be fused, and so on. Q(m,n) represents the fusion index obtained by combining the regional variance and regional energy, which is obtained by formula (5):

[0054] Q(m,n)=a×V(m,n)+b×E(m,n) 6

[0055] In formula (6), Q(m,n) represents the fusion index obtained by combining regional variance and regional energy, m and n are the pixel indices within the region, and their values ​​range from the neighborhood size [M,N]. In this paper, the neighborhood size is taken as 4×4. a and b are the weights, and generally a+b=1. In this paper, a=b=0.5. V(m,n) represents the regional variance, and E(m,n) represents the regional energy, which are calculated according to formulas (7) and (8).

[0056]

[0057] E(m,n)=∑ m≤M,n≤N C(m,n) 2 8

[0058] In formulas (7) and (8), It is the average value of the pixels in the neighborhood.

[0059] Step 7: When reconstructing the image, the Inverse Support Transform (ISVT) is used, and its reconstruction formula is as follows:

[0060]

[0061] In formula (9), the restored fusion effect and noise reduction effect are as follows: Figure 4 , Figure 5 As shown.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A forward-looking sonar image fusion method based on support transformation and sparse representation, characterized in that: The aforementioned forward-looking sonar image fusion method based on support transformation and sparse representation is implemented through the following steps; Step 1: Analyze the raw forward-looking sonar data and calculate the sector sonar image; Step 2: Decompose the image obtained in Step 1 into a multi-level support image sequence and a low-frequency component image; Step 3: Train image dictionary sequences for each level based on the image sequences of each level of support. Step 4: Using the image dictionary sequences trained in Step 3, perform sparse representation of the images decomposed in Step 2 based on the support image sequences at each level, and obtain the sparsity coefficients of the support image sequences at each level. Step 5: Fuse the sparse coefficients of the support image sequences at all levels obtained in Step 4, and restore the fused coefficients to the support image sequence; In step five, the maximum activity criterion is used to fuse the sparse coefficients of the support image sequences at each level; In step five, the low-frequency component images are fused using a criterion that combines regional variance and regional energy. Step 6: Reconstruct the image after fusion in Step 5; In step six, the inverse support transform algorithm is used to reconstruct the fused image.

2. The forward-looking sonar image fusion method based on support transformation and sparse representation according to claim 1, characterized in that: In step two, when decomposing the image, the support transformation theory is used to decompose the forward-looking sonar into a multi-level support image sequence and a low-frequency component image.

3. The forward-looking sonar image fusion method based on support transformation and sparse representation according to claim 1, characterized in that: In step three, a sparse representation method is used to train the image dictionary sequences at each level.

Images