An image enhancement method based on high-frequency and low-frequency adaptive fusion

Abstract

The present invention discloses an image enhancement method based on high-frequency and low-frequency adaptive fusion, comprising: performing contrast enhancement of color channels of an original image respectively to obtain an intermediate feature map; calculating the pixel mean of the intermediate feature map in the color channels respectively; performing color compensation on the intermediate feature map according to the pixel mean to obtain a color reconstructed image; decomposing the color reconstructed image into high-frequency image features and low-frequency image features; performing normalization processing and gamma correction on the high-frequency image features to obtain a high-frequency corrected image; performing sharpening processing on the high-frequency corrected image to obtain a high-frequency restored image; performing defogging processing on the low-frequency image feature weights to obtain a low-frequency restored image, and adaptively fusing the high-frequency restored image and the low-frequency restored image to obtain an output image; the high-frequency restored image provides rich details and edge information, while the low-frequency restored image provides optimized background and color information, thereby significantly improving the visualization effect of the output image.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to an image enhancement method based on high-frequency and low-frequency adaptive fusion. Background Art

[0002] In the fields of marine resource exploration, environmental monitoring and underwater robot technology, the acquisition and processing of underwater images plays a key role in ocean exploration. However, during the imaging process, underwater images often suffer from severe image quality degradation due to the complex underwater environment. The main problems faced by underwater images include light attenuation, color distortion, low contrast, blur and noise, scattering and reflection, etc. Underwater image degradation is mainly caused by factors such as light attenuation, color distortion, scattering and reflection, which greatly reduce the quality of underwater images and affect the accuracy of subsequent image analysis and target recognition.

[0003] To solve these problems, researchers have proposed a variety of underwater image enhancement technologies, including image-based defogging algorithms, color correction methods, and deep learning-based automatic restoration technologies. Defogging algorithms enhance the contrast and clarity of images by removing light scattering and atmospheric scattering effects in underwater images; color correction methods reduce the effects of light attenuation and color shift by restoring the true color of images; and deep learning-based technologies automatically learn the degradation features of images by training deep neural networks, thereby achieving adaptive image enhancement.

[0004] Although these methods have improved the quality of underwater images to a certain extent, when processing images in complex underwater environments, traditional methods still have problems such as over-enhancement, color distortion, loss of image details, and high computational complexity. Therefore, developing more efficient and robust underwater image enhancement methods is still a hot topic and challenge in current research. Summary of the invention

[0005] The present invention provides an image enhancement method based on high-frequency and low-frequency adaptive fusion, which aims to solve the problem of image quality optimization in complex underwater environments and significantly improve the visualization effect of images.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A first aspect of the present invention provides an image enhancement method based on high-frequency and low-frequency adaptive fusion, comprising:

[0008] Get the original image in the underwater environment; The contrast of the three color channels is enhanced to obtain the intermediate feature map; the intermediate feature map is calculated in The average value of pixels under three color channels, recorded as pixel mean ; Based on the pixel mean Perform color compensation on the intermediate feature map to obtain a color reconstructed image;

[0009] Decomposing the color reconstructed image into high-frequency image features and low-frequency image features;

[0010] The high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image; the high-frequency corrected image is sharpened to obtain a high-frequency restored image;

[0011] The low-frequency image features are divided into several sub-regions; by calculating the The pixel mean of the three color channels With standard deviation , according to the pixel mean With standard deviation Determine the atmospheric light value; input the low-frequency image features into the three RGB color channels to obtain the darkest channel image And the brightest channel image ; Based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image;

[0012] The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image.

[0013] Furthermore, the original images are The contrast enhancement of the three color channels is used to obtain the intermediate feature map. The specific process includes:

[0014] The original image is sequentially The contrast enhancement of the three color channels obtains the enhanced pixel value, which is expressed as:

[0015]

[0016]

[0017] In the formula, The radius of the local image block is selected in image processing; is the pixel mean of the local image block, and For the original image Pixel values ​​under three color channels; is the pixel currently being processed, and is the step size of image processing; Indicates color channel; is the red image processing channel; It is the green image processing channel; is the blue image processing channel; For the original image Enhanced pixel values ​​under three color channels;

[0018] For the original image The enhanced pixel values ​​under the three color channels are fused to obtain the intermediate feature map.

[0019] Further, the intermediate feature maps are calculated respectively in The pixel average under three color channels, the specific process includes:

[0020]

[0021] In the formula, when c is , is the pixel mean of the intermediate feature map under the red image processing channel; when c is , is the pixel mean of the intermediate feature map under the green image processing channel. When c is , is the pixel mean of the intermediate feature map under the blue image processing channel, is the intermediate feature map; is the pixel coordinate; M and N are the number of columns and rows of pixels in the intermediate feature map.

[0022] Furthermore, according to the pixel mean Perform color compensation on the intermediate feature map to obtain a color reconstructed image. The specific process is as follows:

[0023] The pixel mean , pixel mean and pixel mean Sort to get the permutation array, the permutation array includes the minimum value , median and maximum value , the expression formula is:

[0024]

[0025] In the formula, S is the permutation array, is the sorting function;

[0026] According to the minimum , median and maximum value The minimum channel color compensation and the second minimum channel color compensation are performed on the intermediate feature map to obtain the first reconstructed image and the second reconstructed image. The expression formula is:

[0027]

[0028]

[0029] In the formula, The first reconstructed image after the minimum channel color compensation is performed on the intermediate feature map; The second reconstructed image after performing sub-minor channel color compensation on the intermediate feature map; The intermediate feature map is at the minimum value Corresponding image features under color access; The intermediate feature map is at the intermediate value Corresponding image features under color access; The intermediate feature map is at the maximum value Corresponding image features under color access;

[0030] The first reconstructed image and the second reconstructed image are fused to obtain a color reconstructed image.

[0031] Furthermore, the color reconstructed image is decomposed into high-frequency image features and low-frequency image features, specifically including:

[0032]

[0033]

[0034]

[0035] In the formula, To decompose the weights, Reconstruct the image for color The maximum pixel value under the three color channels, Reconstruct images for color; Reconstruct the image for color Low-frequency features of the three color channel outputs; Reconstruct the image for color High-frequency features of the three color channel outputs; is the red image processing channel; It is the green image processing channel; is the blue image processing channel;

[0036] Reconstruct the color image in The low-frequency features of the three color channels are fused to obtain the low-frequency image features; the color reconstructed image is The high-frequency features output by the three color channels are fused to obtain high-frequency image features.

[0037] Furthermore, the high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image, specifically including:

[0038] The noise in the high-frequency image features is truncated to obtain a high-frequency denoised image. The expression formula is:

[0039]

[0040]

[0041]

[0042] In the formula, and is the pixel cutoff coefficient of high-frequency image features, is the minimum position of pixel cutoff; is the maximum position of pixel cutoff; is the lower pixel value of noise cutoff; is the upper pixel value of noise cutoff; is a high-frequency image feature; is the high-frequency denoising image feature; is the pixel classification function; is the total number of pixels in the high-frequency image features;

[0043] The high-frequency denoised image features are normalized and gamma corrected to obtain the high-frequency corrected image, which is expressed as:

[0044]

[0045] In the formula, is the minimum pixel value in the high-frequency denoised image; is the maximum pixel value in the high-frequency denoised image; num is the scaling factor; is the gamma correction coefficient; Correct the pixel values ​​in the image for high frequencies; is the pixel value in the high-frequency denoised image.

[0046] Furthermore, the high-frequency corrected image is sharpened to obtain a high-frequency restored image. The specific process includes:

[0047]

[0048]

[0049]

[0050]

[0051] In the formula, is the image after Gaussian smoothing of the high-frequency correction image; is the standard deviation of pixel values ​​in the high-frequency correction image, is pi; The Laplace value of each pixel in the high-frequency correction image; is the pixel coordinate; Compute symbols for partial differentials; For the image Image after Laplace filtering; Correct the image for high frequency; To set weights; Restore the image for high frequencies.

[0052] Furthermore, according to the pixel mean With standard deviation Determine the atmospheric light value. The specific process is:

[0053]

[0054]

[0055]

[0056]

[0057] In the formula, for Set weight values ​​for the three color channels; is the brightness feature; is the maximization function; is the pixel mean The total value of the weighted summation; The standard deviation The total value of the weighted summation; is the atmospheric light value.

[0058] Furthermore, based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image; specifically, the following steps are included:

[0059] Calculate the darkest channel image separately And the brightest channel image The scene depth is calculated as:

[0060]

[0061]

[0062] In the formula, The darkest channel image Depth of scene; The brightest channel image Depth of scene;

[0063] According to the darkest channel image And the brightest channel image The scene depth is calculated to obtain the fused scene depth, and the expression formula is:

[0064]

[0065]

[0066] In the formula, To integrate scene depth; is the dynamic fusion weight; is the estimated mean value of atmospheric light intensity; To set weights;

[0067] The image transmission rate is calculated based on the fusion scene depth, and the low-frequency image features are restored based on the image transmission rate and the atmospheric light value to obtain the low-frequency restored image. The expression formula is:

[0068]

[0069] In the formula, Restore images for low frequencies; is the atmospheric light value to low-frequency image features; is the low-frequency image feature, is the image transmission rate.

[0070] Furthermore, the high-frequency restored image and the low-frequency restored image are adaptively fused to obtain an output image. The specific process includes:

[0071] Gradient information, contrast difference and normalized brightness are extracted from the high-frequency restored image and the low-frequency restored image, and the fusion weight of the high-frequency restored image and the low-frequency restored image is calculated. The expression formula is:

[0072]

[0073]

[0074] In the formula, Restore the gradient information of the image for high frequency; Restore image contrast differences for high frequencies; Restore the normalized brightness of the image for high frequencies; is the fusion weight of high-frequency restored image; Restore the gradient information of the image for low frequencies; Restore image contrast differences for low frequencies; Restore the normalized brightness of the image for low frequencies; The fusion weight of the low-frequency restored image; , and To set weights;

[0075] According to the fusion weight and fusion weight The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image, which is expressed as:

[0076]

[0077] In the formula, Restore images for low frequencies; Restore images for high frequencies; For the output image.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] In the present invention, the original image is processed separately The contrast enhancement of the three color channels obtains the intermediate feature map; by performing contrast enhancement on the three color channels of RGB respectively, the overall contrast of the image can be significantly improved. Contrast enhancement makes the dark parts of the image darker and the bright parts brighter, making the image details more obvious. Color compensation of the intermediate feature map can effectively correct the common color deviations in underwater images and restore the image to a color closer to the real scene.

[0080] In the present invention, the color reconstruction image is decomposed into high-frequency image features and low-frequency image features; the high-frequency image features are normalized and gamma corrected to obtain a high-frequency correction image, which can effectively adjust the brightness and contrast of the image, while suppressing abnormal noise areas, making the high-frequency features more prominent and the details more obvious; the high-frequency correction image is sharpened to obtain a high-frequency restoration image, and the sharpened image can better retain the contour and texture of the object and improve the clarity of the image; the low-frequency image feature weight defogging is performed to obtain a low-frequency restoration image, which can effectively remove the scattering and fogging effects in the underwater image. By calculating the atmospheric light value and transmission rate, the background area of ​​the low-frequency image becomes clearer and the contrast is improved.

[0081] The present invention adaptively fuses the high-frequency restored image and the low-frequency restored image to obtain an output image; the high-frequency restored image provides rich details and edge information, while the low-frequency restored image provides optimized background and color information. Through adaptive fusion, the final output image achieves a good balance between details and background. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a flow chart of the image enhancement method provided by Embodiment 1 and Embodiment 2 of the present invention;

[0083] Figure 2 The image of seabed organisms collected according to Embodiment 2 of the present invention;

[0084] Figure 3 This is the enhanced image of seafloor organisms provided by Example 2 of the present invention;

[0085] Figure 4 The image collected by the submarine equipment provided in Embodiment 2 of the present invention;

[0086] Figure 5 This is the enhanced image of the submarine equipment provided by Example 2 of the present invention;

[0087] Figure 6 This is the seabed rock collection image provided by Example 2 of the present invention;

[0088] Figure 7 This is the enhanced image of seabed rocks provided by Example 2 of the present invention. DETAILED DESCRIPTION

[0089] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0090] Example 1

[0091] like Figure 1 As shown, this implementation provides an image enhancement method based on high-frequency and low-frequency adaptive fusion, including:

[0092] Get the original image in the underwater environment; The contrast enhancement of the three color channels is used to obtain the intermediate feature map. By performing contrast enhancement on the three RGB color channels respectively, the overall contrast of the image can be significantly improved. The intermediate feature map is calculated in The average value of pixels under three color channels, recorded as pixel mean ; Based on the pixel mean The color reconstructed image is obtained by color compensation of the intermediate feature map; it can effectively correct the common color deviations in underwater images and restore the image to a color closer to the real scene.

[0093] Decomposing the color reconstructed image into high-frequency image features and low-frequency image features;

[0094] The high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image; the high-frequency corrected image is sharpened to obtain a high-frequency restored image;

[0095] The low-frequency image features are divided into several sub-regions; by calculating the The pixel mean of the three color channels With standard deviation , according to the pixel mean With standard deviation Determine the atmospheric light value; input the low-frequency image features into the three RGB color channels to obtain the darkest channel image And the brightest channel image ; Based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image;

[0096] The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image; the high-frequency restored image provides rich details and edge information, while the low-frequency restored image provides optimized background and color information. Through adaptive fusion, the final output image achieves a good balance between details and background.

[0097] Example 2

[0098] like Figure 1 As shown, this implementation provides an image enhancement method based on high-frequency and low-frequency adaptive fusion, including:

[0099] Get the original image in the underwater environment; The contrast enhancement of the three color channels is used to obtain the intermediate feature map. The specific process includes:

[0100] The original image is sequentially The contrast enhancement of the three color channels obtains the enhanced pixel value, which is expressed as:

[0101]

[0102]

[0103] In the formula, The radius of the local image block is selected in image processing; is the pixel mean of the local image block, and For the original image Pixel values ​​under three color channels; is the pixel currently being processed, and is the step size of image processing; Indicates color channel; is the red image processing channel; It is the green image processing channel; is the blue image processing channel; For the original image Enhanced pixel values ​​under three color channels;

[0104] For the original image The enhanced pixel values ​​under the three color channels are pixel-fused to obtain the intermediate feature map. It can effectively correct the common color deviation in underwater images and restore the image to a color closer to the real scene. Contrast enhancement helps to restore the details in the image, especially in areas with insufficient or excessive lighting. This is very important for subsequent image analysis and target recognition.

[0105] The intermediate feature maps are calculated respectively in The average value of pixels under three color channels, recorded as pixel mean The specific process includes:

[0106]

[0107] In the formula, when c is , is the pixel mean of the intermediate feature map under the red image processing channel; when c is , is the pixel mean of the intermediate feature map under the green image processing channel. When c is , is the pixel mean of the intermediate feature map under the blue image processing channel, is the intermediate feature map; is the pixel coordinate; M and N are the number of columns and rows of pixels in the intermediate feature map.

[0108] According to the pixel mean Perform color compensation on the intermediate feature map to obtain a color reconstructed image. The specific process is as follows:

[0109] The pixel mean , pixel mean and pixel mean Sort to get the permutation array, the permutation array includes the minimum value , median and maximum value , the expression formula is:

[0110]

[0111] In the formula, S is the permutation array, is the sorting function;

[0112] According to the minimum , median and maximum value The minimum channel color compensation and the second minimum channel color compensation are performed on the intermediate feature map to obtain the first reconstructed image and the second reconstructed image. The expression formula is:

[0113]

[0114]

[0115] In the formula, The first reconstructed image after the minimum channel color compensation is performed on the intermediate feature map; The second reconstructed image after performing sub-minor channel color compensation on the intermediate feature map; The intermediate feature map is at the minimum value Corresponding image features under color access; The intermediate feature map is at the intermediate value Corresponding image features under color access; The intermediate feature map is at the maximum value Corresponding image features under color access;

[0116] The first reconstructed image and the second reconstructed image are fused to obtain a color reconstructed image. Color compensation can effectively correct common color deviations in underwater images. For example, underwater images are usually blue or green, and color compensation can restore colors that are closer to the real scene. The color-compensated image can better retain the color information of the original scene while reducing the impact of lighting and scattering on the image. Color correction and detail restoration help reduce misjudgments caused by image quality problems and improve the accuracy and reliability of subsequent algorithms.

[0117] Decompose the color reconstructed image into high-frequency image features and low-frequency image features, including:

[0118]

[0119]

[0120]

[0121] In the formula, To decompose the weights, Reconstruct the image for color The maximum pixel value under the three color channels, Reconstruct images for color; Reconstruct the image for color Low-frequency features of the three color channel outputs; Reconstruct the image for color High-frequency features of the three color channel outputs; is the red image processing channel; It is the green image processing channel; is the blue image processing channel;

[0122] Reconstruct the color image in The low-frequency features of the three color channels are fused to obtain the low-frequency image features; the color reconstructed image is The high-frequency features output by the three color channels are fused to obtain high-frequency image features.

[0123] The high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image, including:

[0124] The noise in the high-frequency image features is truncated to obtain a high-frequency denoised image. The expression formula is:

[0125]

[0126]

[0127]

[0128] In the formula, and is the pixel cutoff coefficient of high-frequency image features, is the minimum position of pixel cutoff; is the maximum position of pixel cutoff; is the lower pixel value of noise cutoff; is the upper pixel value of noise cutoff; is a high-frequency image feature; is the high-frequency denoising image feature; is the pixel classification function; is the total number of pixels in the high-frequency image features;

[0129] The high-frequency denoised image features are normalized and gamma corrected to obtain the high-frequency corrected image, which is expressed as:

[0130]

[0131] In the formula, is the minimum pixel value in the high-frequency denoised image; is the maximum pixel value in the high-frequency denoised image; num is the scaling factor; is the gamma correction coefficient; Correct the pixel values ​​in the image for high frequencies; is the pixel value in the high-frequency denoised image.

[0132] Normalization and gamma correction of high-frequency image features can effectively adjust the brightness and contrast of the image while suppressing abnormal noise areas. This makes the high-frequency features more prominent and the details more obvious.

[0133] The high-frequency correction image is sharpened to obtain the high-frequency restoration image. The specific process includes:

[0134]

[0135]

[0136]

[0137]

[0138] In the formula, is the image after Gaussian smoothing of the high-frequency correction image; is the standard deviation of pixel values ​​in the high-frequency correction image, is pi; The Laplace value of each pixel in the high-frequency correction image; is the pixel coordinate; Compute symbols for partial differentials; For the image Image after Laplace filtering; Correct the image for high frequency; To set weights; Restore the image for high frequencies.

[0139] The high-frequency corrected image is sharpened to further enhance the edge and detail information of the image. The sharpened image can better preserve the outline and texture of the object and improve the clarity of the image.

[0140] The low-frequency image features are divided into several sub-regions; by calculating the The pixel mean of the three color channels With standard deviation , according to the pixel mean With standard deviation Determine the atmospheric light value. The specific process is:

[0141]

[0142]

[0143]

[0144]

[0145] In the formula, for Set weight values ​​for the three color channels; is the brightness feature; is the maximization function; is the pixel mean The total value of the weighted summation; The standard deviation The total value of the weighted summation; is the atmospheric light value.

[0146] The low-frequency image features are input into the three RGB color channels to obtain the darkest channel image And the brightest channel image ; Based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image; specifically, the following steps are included:

[0147] Calculate the darkest channel image separately And the brightest channel image The scene depth is calculated as:

[0148]

[0149]

[0150] In the formula, The darkest channel image Depth of scene; The brightest channel image Depth of scene;

[0151] According to the darkest channel image And the brightest channel image The scene depth is calculated to obtain the fused scene depth, and the expression formula is:

[0152]

[0153]

[0154] In the formula, To integrate scene depth; is the dynamic fusion weight; is the estimated mean value of atmospheric light intensity; To set weights;

[0155] The image transmission rate is calculated based on the fusion scene depth, and the low-frequency image features are restored based on the image transmission rate and the atmospheric light value to obtain the low-frequency restored image. The expression formula is:

[0156]

[0157] In the formula, Restore images for low frequencies; is the atmospheric light value to low-frequency image features; is the low-frequency image feature, is the image transmission rate.

[0158] Weighted dehazing of low-frequency image features can effectively remove the scattering and fogging effects in underwater images. By calculating the atmospheric light value and transmission rate, the background area of ​​the low-frequency image becomes clearer and the contrast is improved.

[0159] The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image. The specific process includes:

[0160] Gradient information, contrast difference and normalized brightness are extracted from the high-frequency restored image and the low-frequency restored image, and the fusion weight of the high-frequency restored image and the low-frequency restored image is calculated. The expression formula is:

[0161]

[0162]

[0163] In the formula, Restore the gradient information of the image for high frequency; Restore image contrast differences for high frequencies; Restore the normalized brightness of the image for high frequencies; is the fusion weight of high-frequency restored image; Restore the gradient information of the image for low frequencies; Restore image contrast differences for low frequencies; Restore the normalized brightness of the image for low frequencies; The fusion weight of the low-frequency restored image; , and To set weights;

[0164] According to the fusion weight and fusion weight The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image, which is expressed as:

[0165]

[0166] In the formula, Restore images for low frequencies; Restore images for high frequencies; To output an image. By calculating the gradient, contrast and brightness information of the high-frequency and low-frequency images, the fusion weights are dynamically allocated. This adaptive fusion strategy can automatically adjust the contribution ratio of high-frequency and low-frequency features according to the image content to ensure that the enhancement effect can be optimized in different scenarios. The high-frequency restored image provides rich details and edge information, while the low-frequency restored image provides optimized background and color information. Through adaptive fusion, the final output image achieves a good balance between details and background. This embodiment can effectively deal with complex lighting and scattering problems in underwater environments. Whether in clear or turbid underwater environments, high-quality images can be stably output.

[0167] The image enhancement method disclosed in this embodiment is denoted as HLAE; in the experiment of this embodiment, the experimental effect is verified by UIEB dataset, UIQS dataset and UCCS dataset, and compared with the gradient domain color correction method (GDCP), the Laplacian operator-based image enhancement method (ULAP) and the underwater image enhancement method (Water-Net) based on deep learning. The UIQS dataset includes four data subsets, denoted as U1, U2, U3 and U4 respectively.

[0168] The experiment used four indicators for evaluation: edge intensity (EI), average gradient (AG), information entropy (IE), and color correction factor (CCF). A higher EI value indicates that the edge contours in the enhanced image are more obvious, AG indicates improved visibility in the enhanced image, and a higher IE value indicates enhanced details. In addition, a higher CCF value indicates an enhanced visual perception of the improved image.

[0169] Table 1. Mean values ​​of AG, EI, IE, and CCF for each method in the UIEB dataset.

[0170]

[0171] Table 2. Mean AG, EI, IE, and CCF of each method in the UIQS dataset

[0172]

[0173] The underwater images in the UIEB dataset are usually complex and contain rich details, showing a variety of objects and complex lighting conditions in the underwater environment. The images in this dataset are mainly used to evaluate the effects of image enhancement technology in detail recovery, contrast enhancement, and visual perception improvement. The underwater images in the UIQS dataset are divided into five groups, covering underwater images of different water depths, different water qualities (such as turbidity, clarity), and different lighting conditions. Each set of data reflects the challenges faced in the underwater environment, especially the changes in image quality under changes in water quality and lighting differences.

[0174] Table 3. Mean values ​​of AG, EI, IE, and CCF for each method in the UCCS dataset.

[0175]

[0176] The underwater images in the UCCS dataset are divided into three groups: severely color-cast blue, blue-green, and green images. These images reflect the common color distortion problems of underwater images and are suitable for studying how to restore and unify the colors of underwater images to make them more natural and realistic.

[0177] As shown in Tables 1 to 3, for the above different image enhancement requirements, the HLAE method is higher than other methods in any data set, especially in EI, AG, and CCF data indicators. The HLAE method shows extremely high visibility in the visual effect of the image, such as Figures 2 to 7 As shown, this embodiment can effectively retain a large amount of detail information in the image, and can still achieve high-definition image restoration even when the underwater image is severely distorted.

[0178] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0179] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0180] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0181] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0182] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An image enhancement method based on high-frequency and low-frequency adaptive fusion, characterized in that: include: Get the original image in the underwater environment; The contrast of the three color channels is enhanced to obtain the intermediate feature map; the intermediate feature map is calculated in The average value of pixels under three color channels, recorded as pixel mean ; Based on the pixel mean Perform color compensation on the intermediate feature map to obtain a color reconstructed image; Decomposing the color reconstructed image into high-frequency image features and low-frequency image features; The high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image; the high-frequency corrected image is sharpened to obtain a high-frequency restored image; The low-frequency image features are divided into several sub-regions; by calculating the The pixel mean of the three color channels With standard deviation , according to the pixel mean With standard deviation Determine the atmospheric light value; input the low-frequency image features into the three RGB color channels to obtain the darkest channel image And the brightest channel image ; Based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image; The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image.

2. The image enhancement method according to claim 1, characterized in that: The original images are separated into The contrast enhancement of the three color channels is used to obtain the intermediate feature map. The specific process includes: The original image is processed sequentially The contrast enhancement of the three color channels obtains the enhanced pixel value, which is expressed as: ； ； In the formula, The radius of the local image block is selected in image processing; is the pixel mean of the local image block, and For the original image Pixel values ​​under three color channels; is the pixel currently being processed, and is the step size of image processing; Indicates color channel; is the red image processing channel; It is the green image processing channel; is the blue image processing channel; For the original image Enhanced pixel values ​​under three color channels; For the original image The enhanced pixel values ​​under the three color channels are fused to obtain the intermediate feature map.

3. The image enhancement method according to claim 2, characterized in that: The intermediate feature maps are calculated respectively in The pixel mean under the three color channels. The specific process includes: ； In the formula, when c is , is the pixel mean of the intermediate feature map under the red image processing channel; when c is , is the pixel mean of the intermediate feature map under the green image processing channel. When c is , is the pixel mean of the intermediate feature map under the blue image processing channel, is the intermediate feature map; is the pixel coordinate; M and N are the number of columns and rows of pixels in the intermediate feature map.

4. The image enhancement method according to claim 3, characterized in that: According to the pixel mean Perform color compensation on the intermediate feature map to obtain a color reconstructed image. The specific process is as follows: The pixel mean , pixel mean and pixel mean Sort to get the permutation array, the permutation array includes the minimum value , median and maximum value , the expression formula is: ； In the formula, S is the permutation array, is the sorting function; According to the minimum , median and maximum value The minimum channel color compensation and the second minimum channel color compensation are performed on the intermediate feature map to obtain the first reconstructed image and the second reconstructed image. The expression formula is: ； ； In the formula, The first reconstructed image after the minimum channel color compensation is performed on the intermediate feature map; The second reconstructed image after performing sub-minor channel color compensation on the intermediate feature map; The intermediate feature map is at the minimum value Corresponding image features under color access; For the intermediate feature map at the intermediate value Corresponding image features under color access; The intermediate feature map is at the maximum value Corresponding image features under color access; The first reconstructed image and the second reconstructed image are fused to obtain a color reconstructed image.

5. The image enhancement method according to claim 1, characterized in that: Decompose the color reconstructed image into high-frequency image features and low-frequency image features, including: ； ； ； In the formula, To decompose the weights, Reconstruct the image for color The maximum pixel value under the three color channels, Reconstruct images for color; Reconstruct the image for color Low-frequency features of the three color channel outputs; Reconstruct the image for color High-frequency features of the three color channel outputs; is the red image processing channel; It is the green image processing channel; is the blue image processing channel; Indicates color channel; Reconstruct the color image in The low-frequency features of the three color channels are fused to obtain the low-frequency image features; the color reconstructed image is The high-frequency features output by the three color channels are fused to obtain high-frequency image features.

6. The image enhancement method according to claim 1, characterized in that: The high-frequency image features are normalized and gamma corrected to obtain a high-frequency corrected image, including: The noise in the high-frequency image features is truncated to obtain a high-frequency denoised image. The expression formula is: ； ； ； In the formula, and is the pixel truncation coefficient of high-frequency image features, is the minimum position of pixel cutoff; is the maximum position of pixel cutoff; is the lower pixel value of noise cutoff; is the upper pixel value of noise cutoff; is a high-frequency image feature; is the high-frequency denoising image feature; is the pixel classification function; is the total number of pixels in the high-frequency image features; The high-frequency denoised image features are normalized and gamma corrected to obtain the high-frequency corrected image, which is expressed as: ； In the formula, is the minimum pixel value in the high-frequency denoised image; is the maximum pixel value in the high-frequency denoised image; num is the scaling factor; is the gamma correction coefficient; Correct the pixel values ​​in the image for high frequencies; is the pixel value in the high-frequency denoised image.

7. The image enhancement method according to claim 1, characterized in that: The high-frequency correction image is sharpened to obtain the high-frequency restoration image. The specific process includes: ； ； ； ； In the formula, is the image after Gaussian smoothing of the high-frequency correction image; is the standard deviation of pixel values ​​in the high-frequency correction image, is the ratio of pi; The Laplace value of each pixel in the high-frequency correction image; is the pixel coordinate; Compute symbols for partial differentials; Restore images for high frequencies; For the image Image after Laplace filtering; Correct the image for high frequency; To set the weight.

8. The image enhancement method according to claim 7, characterized in that: According to the pixel mean With standard deviation Determine the atmospheric light value. The specific process is: ； ； ； ； In the formula, for Set weight values ​​for the three color channels; is the brightness feature; is the maximization function; is the pixel mean The total value of the weighted summation; The standard deviation The total value of the weighted summation; is the atmospheric light value.

9. The image enhancement method according to claim 8, characterized in that: Based on the darkest channel image And the brightest channel image The image transmission rate is calculated, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain a low-frequency restored image; specifically, the following steps are included: Calculate the darkest channel image separately And the brightest channel image The scene depth is calculated as: ； ； In the formula, The darkest channel image Depth of scene; The brightest channel image Depth of scene; According to the darkest channel image And the brightest channel image The scene depth is calculated to obtain the fused scene depth, and the expression formula is: ； ； In the formula, To integrate scene depth; is the dynamic fusion weight; is the estimated mean value of atmospheric light intensity; To set weights; The image transmission rate is obtained by calculating the depth of the fused scene, and the low-frequency image features are restored according to the image transmission rate and the atmospheric light value to obtain the low-frequency restored image. The expression formula is: ； In the formula, Restore images for low frequencies; is the atmospheric light value; is the low-frequency image feature, is the image transmission rate.

10. The image enhancement method according to claim 9, characterized in that: The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image. The specific process includes: Gradient information, contrast difference and normalized brightness are extracted from the high-frequency restored image and the low-frequency restored image, and the fusion weight of the high-frequency restored image and the low-frequency restored image is calculated. The expression formula is: ； ； In the formula, Restore the gradient information of the image for high frequency; Restore image contrast differences for high frequencies; Restore the normalized brightness of the image for high frequencies; is the fusion weight of the high-frequency restored image; Restore the gradient information of the image for low frequencies; Restore image contrast differences for low frequencies; Restore the normalized brightness of the image for low frequencies; The fusion weight of the low-frequency restored image; , and To set weights; According to the fusion weight and fusion weight The high-frequency restored image and the low-frequency restored image are adaptively fused to obtain the output image, which is expressed as: ； In the formula, Restore images for low frequencies; Restore images for high frequencies; For the output image.

Images