Underwater image enhancement method and system based on energy balance and variance self-adaptation

Abstract

This invention relates to an underwater imaging enhancement method and system based on energy balance and variance adaptation, belonging to the field of image data processing technology. The invention addresses the problems of polarization channel energy imbalance and artifacts easily generated by traditional physical calculations in underwater low-light environments. The method includes: S1: underwater scene image acquisition; S2: brightness alignment of the underwater images to generate a balanced image sequence; S3: traversing the balanced image sequence and calculating normalized fusion coefficients; S4: establishing a dual-branch parallel processing architecture, performing pixel-level weighted fusion of the balanced image sequence according to the fusion coefficients, and simultaneously outputting underwater visual enhancement images and physical property maps of different dimensions. The system includes a light source module, an acquisition module, a processing module, and an output terminal.

Description

Technical Field

[0001] This invention relates to an underwater imaging enhancement method and system based on energy balance and variance adaptation, belonging to the field of image data processing technology. Background Technology

[0002] The 21st century is the century of the ocean. Underwater imaging technology, as a visual extension of humankind's exploration and understanding of the ocean, plays an irreplaceable core role in fields such as marine resource exploration, underwater engineering facility inspection, underwater security and rescue, and marine biological community observation. However, compared to the atmospheric environment, the actual underwater operating environment is extremely harsh. Deep water, low light, and highly turbid water are the constant physical challenges that underwater imaging systems must face. Due to the strong absorption and scattering of light by the water medium, images acquired underwater by traditional optical imaging equipment often suffer from extremely low contrast, blurred details, and severe backscattering veils. In recent years, polarization imaging technology has gradually become a key technological path to improve the quality of underwater visual imaging because it can effectively suppress underwater background backscattering and extract and highlight the physical texture features of the target surface.

[0003] Despite the immense potential of polarization imaging in underwater detection, existing underwater polarization imaging schemes and processing algorithms still face significant macroscopic physical challenges and technical bottlenecks under practical underwater low-light and non-uniform medium conditions. Firstly, there is the channel energy imbalance caused by the anisotropic selective absorption and scattering of light in the medium. Water's absorption and scattering of light are not uniform and isotropic, leading to varying energy losses for light with different polarization directions during underwater transmission. In practical applications, even when the system uses a theoretically highly uniform active illumination source, the polarization sensor still receives... The raw image signals from the four channels often exhibit significant light intensity imbalances. Secondly, there's the amplification effect of energy fluctuations in low-light environments. In deep water or extremely low-contrast dark environments, imaging systems often need to significantly increase the photosensitive gain of the signal to clearly see faint targets. At this time, the originally minute brightness differences between the four polarization channels are magnified many times over. This directly leads to a drastic amplification of the original light intensity deviation when subsequently calculating physical quantities such as the degree of linear polarization (DoLP) or Stokes vector, resulting in severe physical artifacts and grid noise in the generated image, severely obscuring the true physical target.

[0004] Furthermore, the limitations of existing static processing algorithms severely restrict the real-time engineering applications of underwater polarization imaging. Traditional underwater image processing and enhancement methods often rely on fixed global evaluation operators or hard region segmentation methods based on a single threshold. However, the underwater environment is extremely complex and variable, with highly uneven light curtain distribution and highly random noise. When the underwater ambient light intensity fluctuates dynamically or the water turbidity changes, these fixed static enhancement parameters can lead to excessive hardening of image edges, artifact proliferation, or loss of key texture details, failing to meet the operational requirements of autonomous underwater vehicles (AUVs) and other equipment for real-time dynamic perception and accurate target identification in complex waters. Therefore, there is an urgent need to develop an imaging system that can physically smooth out four-channel energy deviations and adaptively allocate fusion weights based on the local texture activity of the image. This is of significant practical importance for overcoming the physical limits of existing underwater polarization imaging and achieving high-definition detection in complex waters. Summary of the Invention

[0005] This invention addresses the issues of polarization channel energy imbalance and artifacts generated by traditional physical calculations in underwater low-light environments, and proposes an underwater imaging enhancement method and system based on energy balance and variance adaptation.

[0006] The technical solution adopted by this invention to solve the above problems is: the underwater imaging enhancement method based on energy balance and variance adaptation proposed in this invention includes: Step 1: Using an active illumination source and polarization imaging equipment, acquire raw sub-image sets of the same underwater scene at multiple different polarization angles; Step 2: Calculate the compensation factor based on the brightness of the original sub-images at each polarization angle, perform pixel-level light intensity compensation on the original sub-images, and obtain a balanced image sequence with brightness alignment. Step 3: Use a sliding window to traverse the balanced image sequence, calculate the local variance of each pixel in its neighborhood, and generate normalized fusion coefficients based on the local variance; Step 4: Construct a dual-branch parallel processing architecture, perform pixel-level weighted fusion of the balanced image sequence according to the fusion coefficient, and simultaneously output underwater visual enhancement images and physical property maps of different dimensions.

[0007] Furthermore, step 1 specifically includes: Utilize active illumination sources to emit polarized light towards underwater target scenes; Polarization information of the target scene at four different polarization angles (0°, 45°, 90°, and 135°) at the same time was simultaneously acquired using a polarization imaging device, forming the original sub-image set. Each polarization angle corresponds to a polarization channel. It is the polarization angle, and .

[0008] Furthermore, step 2 specifically includes: Calculate the original sub-image for each polarization channel separately. global average brightness ; The global average brightness of the original sub-images under each polarization channel. The average value is used as the target reference brightness. Based on target reference brightness Generate compensation factors for each polarization channel ; Based on compensation factor Pixel-level multiplication is performed on the original sub-images under each polarization channel to achieve light intensity compensation, thereby obtaining a brightness-balanced image sequence. ; Global average brightness The calculation formula is: (1); In formula (1), These are the width and height of the image, respectively. These are the pixel coordinates; Target reference brightness The calculation formula is: (2); Brightness Balanced Image Sequence The calculation formula is: (3).

[0009] Furthermore, step 3 specifically includes: For each polarization channel, in pixels Centered Within the sliding window, the average of the sum of squares of the differences between the brightness values ​​of each pixel within the window and the average brightness value of the window is calculated in parallel to obtain the local variance of that pixel. ; Introducing variance duty cycle logic, combined with local variance Calculate the normalized fusion weight coefficient This enables continuous weight mapping across the entire graph. Local variance The calculation formula is: (4); In formula (4),, For this Average brightness value within the sliding window. This refers to the pixel brightness value. Normalized fusion weight coefficient The calculation formula is: (5).

[0010] Furthermore, step 4 specifically includes: A parallel dual-branch processing architecture is established based on the FPGA processing module, where branch one is based on the brightness balance image sequence. Solve the Stokes vector parameters of the scene According to Stokes vector parameters Calculate the linear polarization degree, extract the physical edge contour of the object under extremely turbid water based on the linear polarization degree, and obtain the physical property map; In branch two, based on the normalized fusion weight coefficients Brightness-balanced image sequence for four polarization channels Normalized energy redistribution and pixel-level weighted summation are performed sequentially to obtain the final underwater visual enhancement image. ; Underwater visual enhancement images The calculation formula is: (6).

[0011] Furthermore, this invention also proposes an underwater imaging enhancement system based on energy balance and variance adaptation, comprising: The light source module is used to emit active illumination polarized light to underwater target scenes; The acquisition module, in conjunction with the light source module, is used to simultaneously capture sub-images with multiple polarization directions containing target texture features and backscattered background. The processing module is used to receive the image data from the acquisition module and sequentially execute the multi-channel energy smoothing mechanism, the local variance adaptive evaluation algorithm, and the normalized energy redistribution fusion processing. The processing module is equipped with an FPGA module, which establishes a dual-branch parallel processing architecture to output underwater visual enhancement images and physical property maps. The output terminal is connected to the processing module to receive and present the restored underwater visually enhanced image and physical property map.

[0012] The beneficial effects of this invention are: 1. This invention innovatively constructs a multi-channel energy mitigation mechanism, which effectively mitigates the energy deviation of the four channels from a physical level, greatly suppresses the "pseudo-polarization" noise caused by uneven brightness of underwater channels, and improves the accuracy of physical information calculation.

[0013] 2. This invention innovatively constructs a multi-channel energy mitigation mechanism, which effectively mitigates the energy deviation of the four channels from a physical level, greatly suppresses the "pseudo-polarization" noise caused by uneven brightness of underwater channels, and improves the accuracy of physical information calculation.

[0014] 3. This invention completely abandons the threshold binarization region segmentation technique, which is prone to causing image artifacts and edge hardening. It utilizes a local variance operator to achieve continuous weight mapping across the entire image. In the weight map, brighter areas represent more effective texture information provided by the polarization channel at that location, thus achieving a more natural and smooth transition and fusion.

[0015] 4. The system proposed in this invention highly integrates active illumination, polarization information acquisition, and real-time image enhancement algorithms based on local feature evaluation, and provides dynamic output capability in dual-path synthesis mode. It not only has high real-time performance, but can also be widely applied to various complex operational scenarios such as underwater low-light detection, turbid water target identification, and visual perception of autonomous underwater vehicles. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an underwater imaging enhancement method based on energy balance and variance adaptation. Figure 2 This is a block diagram of the architecture of an underwater imaging enhancement system based on energy balance and variance adaptation. Figure 3 Flowchart for four-channel independent brightness balance processing; Figure 4 A detailed flowchart of the local variance adaptive fusion algorithm; Figure 5 This is the original image; Figure 6 This is the output physical property map; Figure 7 For output visual enhancement images; Figure 8-11 This is the weight mapping diagram calculated in this invention. Detailed Implementation

[0017] like Figure 1 As shown, the steps of the underwater imaging enhancement method based on energy balance and variance adaptation described in this embodiment include: S1: Underwater scene image acquisition; To overcome the strong absorption of light by the underwater medium, this invention utilizes an integrated light source module to emit blue-green polarized light adapted to underwater penetration characteristics as active illumination towards the target area. Subsequently, a polarization imaging acquisition module simultaneously acquires images of the target scene at the same time. The original polarizer image set at four different polarization angles is denoted as , specifically Figure 5 As shown, where Due to the selective absorption and anisotropic backscattering of the water medium, even with a uniform light source, these four images will still exhibit significant macroscopic energy deviations in the initial state.

[0018] S2: Align the brightness of the underwater images to generate a balanced image sequence; like Figure 3 The diagram shown illustrates the four-channel independent luminance balancing process. This step aims to mitigate energy deviations caused by media anisotropy at a physical level, and specifically includes the following steps: S201: Brightness Statistics: For each acquired raw image channel Calculate its global average brightness independently : (1); In formula (1), These are the width and height of the image, respectively. These are the pixel coordinates; S201: Reference Alignment: In order not to change the average brightness level of the overall scene, this invention takes the average of the global brightness of the four channels as the target reference brightness of the system. Then, the energy compensation factor required for a single channel is calculated. : (2); S203: Energy Compensation: Using the obtained compensation factor, perform pixel-level multiplication on the corresponding four-channel original image to output a brightness-aligned balanced image sequence. : (3).

[0019] Through the above steps, the present invention effectively suppresses the pseudo-polarization noise phenomenon caused by uneven channel brightness.

[0020] S3: Traverse the balanced image sequence and calculate the normalized fusion coefficients; like Figure 4 The flowchart shown is a refinement of the local variance adaptive fusion algorithm. Traditional underwater processing algorithms typically use rigid region segmentation, which easily leads to the loss of details. This invention abandons rigid segmentation and instead uses local variance to measure the anti-interference ability of each polarization direction to underwater scattering. Specifically, it includes the following steps: S301: Local variance calculation: For each equilibrium channel calculated in S2 In pixels Centered Within the sliding window, its local variance is calculated in parallel. : (4); In formula (4), For this Average brightness value within the sliding window. This refers to the pixel brightness value. The magnitude of local variance has a clear physical meaning: the larger the variance, the richer the target physical texture information retained by the polarization channel in the region, and the less affected by the "smoothing effect" that causes the image to become blurred due to underwater backscattering.

[0021] S302: Normalized Energy Redistribution and Fusion Coefficient Calculation: Based on the local texture activity of the image, variance duty cycle logic is introduced to calculate the normalized fusion weight coefficient. : (5).

[0022] This weight matrix implements continuous weight mapping across the entire graph. The weight mapping diagram is shown below. Figure 8-11 As shown in the weight map, the brighter the area, the larger the corresponding fusion coefficient, meaning that the polarization channel provides more effective information (texture) at that location.

[0023] S4: Establish a dual-branch parallel processing architecture, perform pixel-level weighted fusion of balanced image sequences according to the fusion coefficient, and simultaneously output underwater visual enhancement images and physical property maps of different dimensions.

[0024] To maximize the use of polarization information, the underlying FPGA processing module of the system has constructed a dual-branch parallel processing architecture, which synchronously outputs two images with different application values.

[0025] In branch one, the balanced image sequence output from S2 is used. Solve the Stokes vector parameters of the scene This allows for the calculation of linear polarization. Since the input sequence has already had its light intensity imbalance smoothed out, this calculation effectively avoids physical artifacts caused by amplified energy fluctuations under low light conditions, thus accurately extracting the physical edge contours of objects in extremely turbid water, forming images such as... Figure 6 The image shown is a DoLP physical image, where DoLP is the degree of linear polarization.

[0026] In branch two, the fusion weight coefficients are calculated in parallel using S3. For the four brightness balance channels Normalized energy redistribution and pixel-level weighted summation are performed to obtain the following: Figure 7 The final visually enhanced image shown

[0027] (6) The enhanced image output by the dual-branch parallel processing architecture effectively suppresses backscattered background, allowing target details that were originally obscured by strong scattered light to emerge, resulting in a higher dynamic range and more natural visual features.

[0028] Furthermore, this embodiment also proposes an underwater imaging enhancement system based on energy balance and variance adaptation, such as... Figure 2 As shown, it includes a light source module, a data acquisition module, a processing module, and an output terminal.

[0029] The light source module is used to emit active illumination polarized light to underwater target scenes; The acquisition module works in conjunction with the light source module to simultaneously capture sub-images with multiple polarization directions that contain target texture features and backscattered background. The processing module is used to receive the image data from the acquisition module and sequentially execute the multi-channel energy smoothing mechanism, the local variance adaptive evaluation algorithm, and the normalized energy redistribution fusion processing. The processing module is equipped with an FPGA module, which establishes a dual-branch parallel processing architecture to output underwater visual enhancement images and physical property maps. The output terminal is connected to the processing module to receive and present the restored underwater visual enhancement image and physical property map.

[0030] 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. An underwater image enhancement method based on energy equalization and variance self-adaption, characterized in that, include: Step 1: Using an active illumination source and polarization imaging equipment, acquire raw sub-image sets of the same underwater scene at multiple different polarization angles; Step 2: Calculate the compensation factor based on the brightness of the original sub-images at each polarization angle, perform pixel-level light intensity compensation on the original sub-images, and obtain a balanced image sequence with brightness alignment. Step 3: Use a sliding window to traverse the balanced image sequence, calculate the local variance of each pixel in its neighborhood, and generate normalized fusion coefficients based on the local variance; Step 4: Construct a dual-branch parallel processing architecture, perform pixel-level weighted fusion of the balanced image sequence according to the fusion coefficient, and simultaneously output underwater visual enhancement images and physical property maps of different dimensions.

2. The method of claim 1, wherein, Step 1 specifically includes: Utilize active illumination sources to emit polarized light towards underwater target scenes; The polarization information of a target scene under four different polarization angles of 0°, 45°, 90° and 135° at the same time is synchronously collected by a polarization imaging device to form the original sub-image set wherein each polarization angle corresponds to a polarization channel, is a polarization angle, and .

3. The underwater imaging enhancement method based on energy balance and variance adaptation according to claim 1, characterized in that, Step 2 specifically includes: Calculate the original sub-image for each polarization channel separately. global average brightness ; The global average brightness of the original sub-images under each polarization channel. The average value is used as the target reference brightness. Based on target reference brightness Generate compensation factors for each polarization channel ; Based on compensation factor Pixel-level multiplication is performed on the original sub-images under each polarization channel to achieve light intensity compensation, thereby obtaining a brightness-balanced image sequence. ; Global average brightness The calculation formula is: （1）； In formula (1), These are the width and height of the image, respectively. These are the pixel coordinates; Target reference brightness The calculation formula is: （2）； Brightness Balanced Image Sequence The calculation formula is: （3）。 4. The underwater imaging enhancement method based on energy balance and variance adaptation according to claim 1, characterized in that, Step 3 specifically includes: For each polarization channel, in pixels Centered Within the sliding window, the average of the sum of squares of the differences between the brightness values ​​of each pixel within the window and the average brightness value of the window is calculated in parallel to obtain the local variance of that pixel. ; Introducing variance duty cycle logic, combined with local variance Calculate the normalized fusion weight coefficient This enables continuous weight mapping across the entire graph. Local variance The calculation formula is: （4）； In formula (4),, For this Average brightness value within the sliding window This refers to the pixel brightness value. Normalized fusion weight coefficient The calculation formula is: （5）。 5. The underwater imaging enhancement method based on energy balance and variance adaptation according to claim 1, characterized in that, Step 4 specifically includes: A parallel dual-branch processing architecture is established based on the FPGA processing module, where branch one is based on the brightness balance image sequence. Solve the Stokes vector parameters of the scene According to Stokes vector parameters Calculate the linear polarization degree, extract the physical edge contour of the object under extremely turbid water based on the linear polarization degree, and obtain the physical property map; In branch two, based on the normalized fusion weight coefficients Brightness-balanced image sequence for four polarization channels Normalized energy redistribution and pixel-level weighted summation are performed sequentially to obtain the final underwater visual enhancement image. ; Underwater visual enhancement images The calculation formula is: （6）。 6. An underwater imaging enhancement system based on energy balance and variance adaptation, applied to the underwater imaging enhancement method based on energy balance and variance adaptation as described in any one of claims 1-5, characterized in that, include: The light source module is used to emit active illumination polarized light to underwater target scenes; The acquisition module, in conjunction with the light source module, is used to simultaneously capture sub-images with multiple polarization directions containing target texture features and backscattered background. The processing module is used to receive the image data from the acquisition module and sequentially execute the multi-channel energy smoothing mechanism, the local variance adaptive evaluation algorithm, and the normalized energy redistribution fusion processing. The processing module is equipped with an FPGA module, which establishes a dual-branch parallel processing architecture to output underwater visual enhancement images and physical property maps. The output terminal is connected to the processing module and is used to receive and present the restored underwater visual enhancement image and physical property map.

Images