An underwater image restoration method based on polarization component extreme value construction and differential evolution fusion

Abstract

This invention discloses an underwater image restoration method based on polarization component extremum construction and differential evolution fusion, belonging to the field of underwater optical imaging technology. The method includes: acquiring image sequences with different polarization angles using an active polarization imaging system, calculating the brightest and darkest images; based on the backscattered light polarization degree, setting the target light polarization degree as the physical extremum of the domain, directly constructing high and low polarization component images and performing non-negative correction; establishing a linear fusion model for the component images, constructing a fitness function using information entropy and standard deviation, and adaptively optimizing the enhancement factor using a differential evolution algorithm; finally, fusing the component images using the optimal enhancement factor and outputting the restoration result. This invention achieves effective restoration of underwater scenes with complex polarization characteristics without image segmentation, significantly improving contrast and information richness while maintaining image naturalness, providing effective technical support for underwater detection, target recognition, and navigation systems.

Description

Technical Field

[0001] This invention relates to the field of underwater optical imaging technology, and in particular to an underwater image restoration method based on polarization component extremum construction and differential evolution optimization. Background Technology

[0002] Underwater optical imaging technology has significant applications in marine resource exploration, underwater target detection, and ecological monitoring. However, the scattering effect caused by suspended particles in water severely degrades image quality, leading to blurred details and reduced contrast, which seriously limits the effectiveness of underwater optical images. To address these challenges, polarization imaging technology has been introduced into this field as an effective means of enhancing physical information. This technology effectively separates target light information from the backscattered light by analyzing the difference in polarization characteristics between the reflected light and the backscattered light from the target, providing a new technical approach for improving underwater visual clarity.

[0003] Traditional underwater polarization imaging methods typically estimate the target polarization degree as a global constant, making it difficult to fully reconstruct complex scenes with targets exhibiting significant polarization differences. In recent years, some improved methods have attempted to use image segmentation techniques to partition underwater scenes, aiming to achieve local polarization parameter estimation. However, under highly turbid water imaging conditions, due to severe attenuation of effective information and extreme weakening of target features, image segmentation-based approaches face fundamental technical bottlenecks. Their performance improvement is limited by inherent principles, making it difficult to maintain stable reconstruction results under highly turbid conditions. Furthermore, traditional underwater polarization imaging methods require manually selecting targetless background areas to calculate backscattered light polarization, necessitating human-computer interaction and limiting the application scope of underwater polarization imaging technology. Summary of the Invention

[0004] This invention aims to address the problem that existing underwater polarization imaging technology struggles to fully restore images in highly turbid waters due to the significant differences in target polarization characteristics. It proposes an underwater image restoration method based on polarization component extremum construction and differential evolution fusion.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An underwater image restoration method based on polarization component extremum construction and differential evolution fusion is proposed. This method can effectively restore underwater turbid scenes with complex polarization characteristics without image segmentation. The specific steps include the following: Step 1, Polarization Image Acquisition and Information Calculation: Using an active polarization imaging system, acquire image sequences of the scene to be restored at different polarization angles, and calculate the image with the brightest light intensity. and the darkest image ; Step 2, Construction of extrema of polarization component image: The polarization degree of the target light... The physical extrema of the domain are set respectively, and the high polarization component image and the low polarization component image are directly calculated and generated. The obtained component images are then subjected to non-negative correction processing. Step 3, Enhancement factor optimization based on differential evolution algorithm: Establish a fusion model for high and low polarization component images to fuse the information entropy of the images. with standard deviation Construct a fitness function; use differential evolution algorithm for adaptive optimization, where each individual in the algorithm population is represented as a two-dimensional vector. , representing a set of candidate combinations of enhancement factors; by iteratively performing mutation, crossover, and selection operations, the population evolves within a preset parameter space, ultimately searching for the combination of enhancement factors that optimizes the fitness function. and ; Step 4, Image Sharpening Generation and Output: The optimal high-bias enhancement factor is then used to generate and output the image. With low-biased enhancement factor Substitute the data into the fusion model to generate and output the final restored image.

[0006] Furthermore, in step 1 of this invention, polarization image acquisition and information processing involve acquiring four images at different polarization angles (0°, 45°, 90°, and 135°) in an underwater turbidity scene using a multi-channel real-time polarization detection system; calculating the linear polarization component using the Stokes vector method; and obtaining the brightest image by fitting the light intensity at different polarization angles. and the darkest image Calculate the polarization degree image And estimate the polarization degree of the backscattered light. ;

[0007] Furthermore, in step 1, the polarization degree of the backscattered light is estimated. This includes: utilizing gamma correction to enhance the contrast of the polarization image, taking into account the depolarization characteristics of underwater turbid media, to obtain an enhanced polarization image. ;

[0008] Take the global average value of the compensated polarization degree image. The estimated value;

[0009] Furthermore, the extreme value construction of the polarization component image in step 2 of this invention specifically includes: combining the physical meaning of polarization degree, the value range of polarization degree is 0 to 1, representing completely unpolarized light and completely linearly polarized light, respectively; based on this physical characteristic, the polarization degree of the target light is... By setting the extreme values ​​at the endpoints of their respective domains and directly substituting them into the active polarization imaging model, the optimal high polarization component image and low polarization component image are calculated respectively; by setting the target light polarization degree to 1, the high polarization component image is obtained. :

[0010] By setting the target light polarization degree to 0, a low-polarization component image is obtained. :

[0011] Furthermore, in step 2, non-negative correction processing is performed on the obtained component images. Specifically, to avoid negative pixel values ​​affecting the subsequent fusion effect, the high polarization component images are corrected. and low polarization component images Non-negative correction is performed on the image after correction. and satisfy:

[0012] Furthermore, the enhancement factor optimization based on the differential evolution algorithm in step 3 of this invention specifically includes: introducing a high-biased enhancement factor. and low-biased enhancement factor A linear fusion model for high and low polarization component images is established to calculate the fused image. :

[0013] Furthermore, step 3, the optimization of enhancement factors based on the differential evolution algorithm, specifically includes: constructing a fitness function using information entropy and standard deviation; performing adaptive optimization using the differential evolution algorithm; representing the combination of enhancement factors as two-dimensional vector individuals; generating a candidate solution population through initialization; performing mutation operations on the target vector to generate a mutated vector; performing crossover operations to generate experimental vectors; updating the population using a greedy selection strategy; and automatically searching for the optimal combination of enhancement factors within the preset parameter space through iterative evolution. and .

[0014] Furthermore, in step 4 of this invention, the generation and output of the sharpened image involves using the optimal high-bias enhancement factor. With low-biased enhancement factor Substitute the data into the fusion model to generate and output the final restored image.

[0015]

[0016] in, For the final restored image, The optimal high-biased enhancement factor, It is the optimal low-biased enhancement factor.

[0017] Compared with existing technologies, this invention discloses an underwater image restoration method based on polarization component extremum construction and differential evolution optimization, which can achieve the following beneficial technical effects: 1. This invention effectively solves the problem of uneven restoration effect of traditional methods in complex polarization characteristic scenarios by designing a polarization component extremum construction mechanism, and can achieve synergistic enhancement of high and low polarization targets without relying on image segmentation technology; 2. This invention employs an intelligent optimization strategy based on differential evolution algorithm. By establishing a fitness function with information entropy and standard deviation as joint evaluation indicators, it can adaptively balance the enhancement amplitude in different polarization characteristic regions. 3. Through an end-to-end fully automated processing flow, this invention eliminates the need for manual selection of background areas or manual adjustment of parameters, reducing reliance on operators and significantly improving the practicality and ease of use of underwater polarization imaging. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating the effect of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating the overall process of the underwater image restoration method based on polarization component extremum construction and differential evolution optimization of the present invention. Specifically, it includes the following steps: Step 1, Polarization Image Acquisition and Information Calculation: Using an active polarization imaging system, acquire image sequences of the scene to be restored at different polarization angles, and calculate the image with the brightest light intensity. and the darkest image .

[0021] Step 1.1: Place the highly polarized target (a metal coin in this invention) and the low-polarized target (a rough sticker in this invention) at the bottom of the water tank to simulate a turbid water environment (simulated by adding skim milk in this invention). Adjust the direction of the analyzer in front of the polarization camera and acquire image sequences at four analysis angles: 0°, 45°, 90°, and 135°. , , , During image acquisition, maintain stable lighting conditions, water turbidity, and the position of the target object.

[0022] Step 1.2: Calculate the polarization parameters using the Stokes vector method. The Stokes vector components are calculated as follows: in, , , , Image sequences with four bias detection angles: 0°, 45°, 90°, and 135°.

[0023] Calculate arbitrary polarization angle Light intensity : in, Polarization angle The light intensity below, The angle of the analyzer.

[0024] By fitting the light intensity curve, the brightest and darkest images are extracted, and their polarization directions are orthogonal. According to the active polarization imaging model, the brightest and darkest images satisfy the following: in, The brightest image light intensity, The light intensity of the darkest image. Image with maximum target light intensity The image showing the maximum intensity of backscattered light. Image with minimum target light intensity This is the image with the minimum backscattered light intensity.

[0025] Calculate polarization degree image : In the applicable scenarios of this invention, linearly polarized light is used for illumination, and the backscattering medium has a depolarization effect, thus affecting the polarization degree image. Perform gamma ( The polarization image is enhanced by transformation. This process can be represented as: Take the average polarization degree after enhancement as the estimated value of the polarization degree of the backscattered light: Step 2, Construction of extrema of polarization component image: Combining the physical meaning of polarization degree, the value range of polarization degree is 0 to 1, representing completely unpolarized light and completely linearly polarized light, respectively; based on this physical characteristic, the polarization degree of the target light is... The physical extrema of the domain are set respectively, and the high polarization component image and the low polarization component image are directly calculated and generated. The obtained component images are then subjected to non-negative correction processing. Step 2.1, Construction of extrema of polarization component image: Generating high polarization component image Using the low-polarization component image and the target light polarization degree as a reference, the high-polarization component image is calculated. The formula for calculating the high-polarization component image is as follows: in, For high polarization component images, This represents the polarization degree of the backscattered light.

[0026] Similarly, setting For low-biased component images, the formula for calculating the low-biased component map is as follows: in, This is an image with low polarization components.

[0027] Step 2.2, Non-negative correction: To avoid negative pixel values ​​affecting subsequent processing, the following steps are performed: and Pixel correction is performed using the following formula: in This is the high polarization component image after non-negative correction; This is the low polarization component image after non-negative correction.

[0028] The corrected high and low bias component images are then fused pixel-by-pixel to generate a sharpened image. This process can be expressed as follows: in, It is a high-biased enhancement factor (HPEF). It is a low-biased enhancement factor (LPEF).

[0029] Step 3, Enhancement factor optimization based on differential evolution algorithm: Establish a fusion model for high and low polarization component images to fuse the information entropy of the images. with standard deviation Construct a fitness function; use differential evolution algorithm for adaptive optimization, where each individual in the algorithm population is represented as a two-dimensional vector. , representing a set of candidate combinations of enhancement factors; by iteratively performing mutation, crossover, and selection operations, the population evolves within a preset parameter space, ultimately searching for the combination of enhancement factors that optimizes the fitness function. and ; Step 3.1, for optimization and The Differential Evolution (DE) algorithm is used to optimize parameters and construct a joint evaluation index. Using joint evaluation indicators The fitness function is expressed as follows: in, The standard deviation of the image reflects the image contrast, and its calculation formula is: Image information entropy reflects the richness of image information, and its calculation formula is: in, grayscale The probability of its occurrence, This represents the total number of gray levels.

[0030] Step 3.2, Differential Evolution Algorithm Parameter Settings: Maximum number of iterations set to 100, population size set to 50, scaling factor set to 0.8, crossover probability set to 0.9, and parameter search range set to... ∈[1,10], ∈[1,10]; Step 3.3, Algorithm Execution Flow: Randomly generate the initial population, with each individual being [ , The parameters of the target vector are combined; for each target vector, three different individuals are randomly selected to generate a mutation vector: mutant = a + F×(b - c); the target vector and the mutation vector are crossed to generate a trial vector; a greedy strategy is adopted, if the fitness of the trial vector is better than that of the target vector, the target vector is replaced; the above process is repeated until the maximum number of iterations is reached; the optimal value is output. and Parameter combinations.

[0031] Step 4, Image Sharpening Generation and Output: The optimal high-bias enhancement factor is then used to generate and output the image. With low-biased enhancement factor Substitute the data into the fusion model to generate and output the final restored image. .

[0032] Specifically, the final restored image The expression is: in, For the final restored image, The optimal high-biased enhancement factor, It is the optimal low-biased enhancement factor.

[0033] In embodiments of this invention, an underwater image restoration method based on polarization component extremum construction and differential evolution fusion is used for image processing. The processed image dataset is a polarization image dataset collected in a real underwater environment, and the optimization objective is to maximize the joint evaluation index F(SE). After processing, a test set is used to verify the effectiveness of the method. Furthermore, a comparison is made with traditional active polarization imaging methods, such as... Figure 2 As shown in Table 1, the underwater image restoration method based on polarization component extremum construction and differential evolution fusion proposed in this invention can effectively improve the quality of underwater images. Compared with traditional methods, the method of this invention has significant improvements in EME and contrast indicators, and can achieve balanced restoration for targets with different polarization characteristics.

[0034] Table 1 Comparison of Objective Evaluation Indicators method EME Contrast Original image 1.998 0.16 Traditional active polarization imaging methods 5.437 1.068 This invention 14.5481 25.7076 Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An underwater image restoration method based on polarization component extremum construction and differential evolution fusion, characterized in that, Includes the following steps: Step 1, Polarization Image Acquisition and Information Calculation: In an underwater turbid scene, four images with different polarization angles (0°, 45°, 90°, and 135°) were acquired. The Stokes vector method was used to calculate the linear polarization component, and the brightest image was obtained by fitting the light intensity at different polarization angles. and the darkest image Calculate the polarization degree image And estimate the polarization degree of the backscattered light. ; Step 2, Construction of extrema of polarization component image: Based on the polarization degree of backscattered light, the polarization degree of target light is... The physical extrema of the domain are set respectively, and the high-biased component image and the low-biased component image are directly calculated and generated. The obtained component images are then subjected to non-negative correction processing. Step 3, Enhancement factor optimization based on differential evolution algorithm: Establish a fusion model for high and low polarization component images to fuse the information entropy of the images. with standard deviation Construct a fitness function; use differential evolution algorithm for adaptive optimization; Step 4, Image Sharpening Generation and Output: The optimal high-bias enhancement factor is then used to generate and output the image. With low-biased enhancement factor Substitute the data into the fusion model to generate and output the final restored image.

2. The method as described in claim 1, characterized in that, In step 1, the polarization degree of the backscattered light is estimated. This includes: utilizing gamma correction to enhance the contrast of the polarization image, taking into account the depolarization characteristics of underwater turbid media, to obtain an enhanced polarization image. ; The global average value of the compensated polarization degree image is taken as the backscattered polarization degree of the background light. The estimated value; 。 3. The method as described in claim 1, characterized in that, Step 2 involves constructing the extrema of the polarization component image, specifically including: considering the physical meaning of polarization degree, the value range of polarization degree is 0 to 1, representing completely unpolarized light and completely linearly polarized light, respectively; based on this physical characteristic, the polarization degree of the target light is... By setting the extreme values ​​at the endpoints of their respective domains and directly substituting them into the active polarization imaging model, the optimal high polarization component image and low polarization component image are calculated respectively; by setting the target light polarization degree to 1, the high polarization component image is obtained. : By setting the target light polarization degree to 0, a low-polarization component image is obtained. : 。 4. The method as described in claim 1, characterized in that, In step 2, the obtained component images undergo non-negative correction processing. Specifically, to avoid negative pixel values ​​affecting the subsequent fusion effect, the high polarization component images are corrected. and low polarization component images Non-negative correction is performed on the image after correction. and satisfy: 。 5. The method as described in claim 1, characterized in that, Step 3, the enhancement factor optimization based on the differential evolution algorithm, specifically includes: introducing a high-biased enhancement factor. and low-biased enhancement factor A linear fusion model for high and low polarization component images is established to calculate the fused image. : 。 6. The method as described in claim 1, characterized in that, Step 3, the enhancement factor optimization based on differential evolution algorithm, specifically includes: constructing a fitness function using information entropy and standard deviation; performing adaptive optimization using differential evolution algorithm; representing the enhancement factor combination as a two-dimensional vector individual; generating a candidate solution population through initialization; performing mutation operation on the target vector to generate a mutation vector; performing crossover operation to generate an experimental vector; updating the population using a greedy selection strategy; and automatically searching for the optimal enhancement factor combination within a preset parameter space through iterative evolution. and .

7. The method as described in claim 1, characterized in that, The image sharpening and output in step 4 involves using the optimal high-bias enhancement factor. With low-biased enhancement factor Substitute the data into the fusion model to generate and output the final restored image. in, For the final restored image, The optimal high-biased enhancement factor, It is the optimal low-biased enhancement factor.

Images