CN115908179B - Underwater image contrast enhancement method based on double priori optimization - Google Patents

Abstract

The invention provides a double priori optimized underwater image contrast enhancement method, which is characterized in that an underwater image with corrected colors is obtained, the underwater image with corrected colors is converted into an HSV color model from an RGB color model, and a base layer and a detail layer are decomposed aiming at a V channel, wherein the method is represented as follows: for the V channel, decomposing the V channel into a base layer and a detail layer by adopting a space priori and a texture priori, realizing local contrast enhancement of the base layer by utilizing an integral strategy to count local mean and variance for the base layer, and realizing enhancement of texture details by utilizing a nonlinear stretching function for the detail layer. The underwater image contrast enhancement method with double priori optimization enhances the contrast of the image, highlights texture details, and can well solve the problem of texture detail loss caused by scattering.

Description

Double-prior optimization underwater image contrast enhancement method

Technical field

The present invention relates to an underwater image processing method, and specifically, to a dual-prior optimization underwater image contrast enhancement method.

Background technique

In the field of underwater robot vision, clear underwater images are an important carrier and presentation form for obtaining valuable information. However, due to the complex physical and chemical environment under water and the absorption and scattering of light, underwater images often suffer from various quality degradation problems. Scattering can easily cause fogging and blurring of underwater images and loss of details; absorption can easily cause color distortion and reduced contrast and brightness of underwater images.

Among them, color distortion is the main quality degradation problem faced by underwater images. Currently, methods based on statistics, methods based on linear stretching, methods based on compensation, and methods based on color transmission are gradually applied to color distortion correction of underwater images and have achieved effectiveness.

Statistics-based methods rely on prior information, but it is difficult to solve for prior information. Linear stretching-based methods and compensation-based methods are prone to introduce reddish distortion and over-correction problems. Methods based on color transfer rely on reference images, so that the effectiveness and robustness of the algorithm face challenges. Overall, despite their effectiveness in underwater color correction, these methods still have some limitations.

For example, the invention patent with application number: CN202210556777.0 and invention title: A locally adaptive underwater image contrast enhancement method proposes to obtain a color-corrected underwater image, and convert the color-corrected underwater image from RGB The color model is converted to the CIELAB color model, and different strategies are executed for the brightness channel L, color channel a, and color channel b: for the brightness channel L, taking the local image block as the object, the integral map and the square integral map are used to collect statistics on the local image block. mean and variance, and use the mean and variance of the local image block to adaptively enhance the contrast of the brightness channel L; in the process of enhancing the contrast of the brightness channel L, introduce guided filtering to reduce noise; for the color channel a and color channel b , using a color balancing strategy to balance the color difference between the color channel a and the color channel b. This method enhances the image with moderate contrast, low time complexity, can suppress noise, and makes the contrast and color of the enhanced image close to that of land images. However, this method ignores texture details and does not have wide applicability.

In general, traditional linear stretching methods are prone to over-correction problems, and this type of method cannot effectively correct underwater images with diverse color distortions.

In order to solve the above existing problems, people have been seeking an ideal technical solution.

Contents of the invention

The purpose of the present invention is to address the shortcomings of the existing technology, thereby providing a dual-prior optimized underwater image contrast enhancement method that enhances the contrast of the image, highlights the texture details, and can well solve the problem of loss of texture details due to scattering. .

In order to achieve the above purpose, the technical solution adopted by the present invention is: a double-prior optimization underwater image contrast enhancement method. First, the color-corrected underwater image is obtained, and the color-corrected underwater image is obtained from the RGB color model. Convert to HSV color model, implement base layer and detail layer decomposition for V channel, expressed as follows:

For the V channel, spatial prior and texture prior are used to decompose the V channel into a base layer and a detail layer. For the base layer, the integral strategy is used to count the local mean and variance to achieve local contrast enhancement of the base layer. For the detail layer, nonlinear pulling is used. The stretch function realizes the enhancement of texture details;

Among them, the optimization model of the base layer and detail layer using spatial and texture prior decomposition of the V channel is defined as follows:

Among them, _IV , I _B and _IV - I _B represent the input image, base layer and detail layer respectively; and/> Represent the cascade vectors of I _V and I _B respectively;/> Represents an integrated vector;/> Represents the expression of the gradient operation matrix in different directions; N represents the total number of pixels in the input image;

of the first item Force the base layer to be as close to the input image as possible; then introduce the second term/> The base layer spatial prior; at the same time, the third detail layer texture prior/> Performs an element-wise nonzero operation to produce a binary vector.

Based on the above, the augmented Lagrangian model is used to redefine it as:

Among them, two auxiliary variables are introduced and/> Replace/> and/> j ₁ and j ₂ are the two variables of the Lagrangian function, and eta is the iteratively updated parameter of the objective function;

The alternating direction method of multipliers (ADMM) is used to iteratively optimize the objective function by minimizing several sub-problems and maximizing two problems. When λ ₁ =0.25 and λ ₂ =0.025, the alternating direction method of multipliers is used at least 20 times During the iteration, I _B close to the bottom layer of the input image is effectively decomposed. After solving the base layer, the detail layer is solved by I _D = _IV - I _B , and then different enhancement operations are performed on the base layer and the detail layer.

Based on the above, use the advantageous attributes of local image blocks to enhance the base layer, approximate the mean value of local block K as a low-frequency component, and approximate the information after subtracting K from the input image block as a high-frequency component; the key to improving the detail layer How to effectively improve high-frequency components; in the local image block K, use integral mapping to solve the local mean within the block, and use it to improve the base layer. The enhancement process is defined as:

in, is the mean value of the local image block, δ is the enhancement control parameter of the high-frequency component,/> To strengthen the grassroots;

To avoid artifacts or local darkening during local block enhancement, use a Gamma correction strategy with better stretching performance for low pixel values, which is redefined as:

Among them, θ is the correction factor parameter, set θ to 0.65, use local blocks to traverse the bottom layer, and obtain the final enhanced bottom layer.

Based on the above, use a nonlinear stretching function to further stretch the detail layer, and its expression is:

where σ is the stretch control reference, set to 0.88. is an enhanced layer of detail;

Using the enhanced detail layer and the enhanced base layer we get an enhanced V channel, defined as:

in, It is an enhanced V channel.

Compared with the existing technology, the present invention has outstanding substantive features and significant progress. Specifically, the present invention first uses an algorithm to obtain the base layer and detail layer of the underwater image, and uses a double-prior optimization underwater image contrast enhancement strategy: That is, in the base layer, the advantageous attributes of local image blocks are used to enhance the base layer. In the detail layer, the texture structure of the sharpened image is taken into account, and a nonlinear stretching function is used to further stretch the detail layer. The present invention is used to solve the problem that traditional linear stretching methods are prone to over-correction.

In addition, the present invention uses spatial and texture strategies: that is, further correcting the base layer and detail layer so that the enhanced image is as close as possible to the land image in terms of contrast and color, and can be widely used in the field of underwater image processing.

Description of the drawings

Figure 1 is a schematic flowchart of the underwater image contrast enhancement method of dual prior optimization of the present invention in an embodiment of the present invention.

Figure 2 is the underwater image enhancement results with other methods in the embodiment of the present invention.

Detailed ways

The technical solution of the present invention will be further described in detail below through specific embodiments.

In order to verify the effectiveness of color correction of the present invention, underwater images with different color distortion types were selected as test sets, and subjective and objective comparisons were made with IBLA, HMUC, GDCP, DTVR, BRU, ACDC, UNTV, WaterNet, FUnlEGAN and UIECNet methods. .

As shown in Figure 1, a dual-prior optimization underwater image contrast enhancement method first obtains the color-corrected underwater image, and converts the color-corrected underwater image from the RGB color model to the HSV color model. For V The channel implements base layer and detail layer decomposition, expressed as follows:

For the V channel, the spatial prior and texture prior are used to decompose the V channel into a base layer and a detail layer. For the base layer, the integral strategy is used to count the local mean and variance to achieve local contrast enhancement of the base layer. For the detail layer, nonlinear pulling is used. The stretch function realizes the enhancement of texture details.

Among them, the optimization model of the base layer and detail layer using spatial and texture prior decomposition of the V channel is defined as follows:

Among them, _IV , I _B and _IV - I _B represent the input image, base layer and detail layer respectively; and/> Represent the cascade vectors of I _V and I _B respectively;/> Represents an integrated vector;/> Represents the expression of the gradient operation matrix in different directions; N represents the total number of pixels in the input image;

of the first item Force the base layer to be as close to the input image as possible; then introduce the second term/> The base layer spatial prior; at the same time, the third detail layer texture prior/> Performs an element-wise nonzero operation to produce a binary vector.

Using the augmented Lagrangian model, it is redefined as:

This formula is mainly used to obtain each sub-item in the aforementioned formula, and then obtain the base layer and detail layer. Among them, two auxiliary variables are introduced and/> Replace/> and/> j ₁ and j ₂ are the two variables of the Lagrangian function, and eta is the iteratively updated parameter of the objective function;

The alternating direction method of multipliers (ADMM) is used to iteratively optimize the objective function by minimizing several sub-problems and maximizing two problems. When λ ₁ =0.25 and λ ₂ =0.025, the alternating direction method of multipliers is used at least 20 times During the iteration, I _B close to the bottom layer of the input image is effectively decomposed. After solving the base layer, the detail layer is solved by I _D = _IV - I _B , and then different enhancement operations are performed on the base layer and the detail layer.

Use the advantageous attributes of local image blocks to enhance the base layer, approximate the mean value of local block K as a low-frequency component, and approximate the information after subtracting K from the input image block as a high-frequency component; the key to improving the detail layer is how to effectively improve it High-frequency components; in the local image block K, integral mapping is used to solve the local mean within the block, and it is used to improve the base layer. The enhancement process is defined as:

in, is the mean value of the local image block, δ is the enhancement control parameter of the high-frequency component,/> To strengthen the grassroots;

To avoid artifacts or local darkening during local block enhancement, a Gamma correction strategy is used to have better stretching performance for low pixel values, which is redefined as:

Among them, θ is the correction factor parameter, set θ to 0.65, use local blocks to traverse the bottom layer, and obtain the final enhanced bottom layer.

Use a nonlinear stretch function to further stretch the detail layer, its expression is:

Among them, σ is the stretch control reference, set to 0.88; is an enhanced layer of detail;

Using the enhanced detail layer and the enhanced base layer we get an enhanced V channel, defined as:

in, It is an enhanced V channel.

As shown in Figure 2, the present invention demonstrates the enhancement results of testing underwater images with color distortion compared with other methods. As can be seen from Figure 2, the GDCP and FUnlEGAN methods have poor performance in contrast enhancement and detail highlighting. Other methods are better than the GDCP and FUnlEGAN methods, but they are worse than the present invention in detail highlighting. The subjective results of the enhanced image of the present invention are distributed wider and more uniformly than other methods.

This embodiment compares different methods with UIQM and CCF. From the data in Table 1 and Table 2, it can be seen that the present invention has the highest UIQM and the second highest CCF value, which shows that the present invention also has better enhancement performance in terms of objective evaluation indicators. . Therefore, this scheme is superior to the comparative method in terms of both subjective and objective evaluation.

Table 1 Comparison of UIQM of images corrected by the method of the present invention and other methods

Table 2 Comparison of CCF of images corrected by the method of the present invention and other methods

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it; although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the present invention can still be modified Modifications to the specific embodiments of the invention or equivalent substitutions of some of the technical features without departing from the spirit of the technical solution of the present invention shall be covered by the scope of the technical solution claimed by the present invention.

Claims (4)

1. A dual-prior optimization underwater image contrast enhancement method, characterized by: first obtaining the color-corrected underwater image, converting the color-corrected underwater image from the RGB color model to the HSV color model, aiming at V The channel implements base layer and detail layer decomposition, expressed as follows: For the V channel, spatial prior and texture prior are used to decompose the V channel into a base layer and a detail layer. For the base layer, the integral strategy is used to count the local mean and variance to achieve local contrast enhancement of the base layer. For the detail layer, nonlinear pulling is used. The stretch function realizes the enhancement of texture details; Among them, the optimization model of the base layer and detail layer using spatial and texture prior decomposition of the V channel is defined as follows: Among them, _IV , I _B and _IV - I _B represent the input image, base layer and detail layer respectively; and/> Represent the cascade vectors of I _V and I _B respectively;/> Represents an integrated vector;/> Represents the expression of the gradient operation matrix in different directions; N represents the total number of pixels in the input image; of the first item Force the base layer to be as close to the input image as possible; then introduce the second term/> The base layer spatial prior; at the same time, the third detail layer texture prior/> Performs an element-wise nonzero operation to produce a binary vector. 2. The double-prior optimized underwater image contrast enhancement method according to claim 1, characterized in that: the augmented Lagrangian model is used to redefine it as: Among them, two auxiliary variables are introduced and/> Replace/> and/> j ₁ and j ₂ are the two variables of the Lagrangian function, and eta is the iteratively updated parameter of the objective function; The alternating direction method of the multiplier is used to iteratively optimize the objective function. When λ ₁ =0.25 and λ ₂ =0.025, the alternating direction method of the multiplier effectively decomposes I _B close to the bottom layer of the input image in at least 20 iterations and solves the base layer. Finally, the detail layer is solved through I _D = _IV - I _B , and then different enhancement operations are performed on the base layer and the detail layer. 3. The underwater image contrast enhancement method of double prior optimization according to claim 2, characterized in that: using local image blocks to enhance the base layer, approximating the mean value of the local block K to a low-frequency component, and converting the input image block into The information after subtracting K is approximately a high-frequency component; in the local image block K, integral mapping is used to solve the local mean within the block and use it to improve the base layer. The enhancement process is defined as: in, is the mean value of the local image block, δ is the enhancement control parameter of the high-frequency component,/> To strengthen the grassroots; Correction is performed using the Gamma correction strategy, which is redefined as: Among them, θ is the correction factor parameter, set θ to 0.65, use local blocks to traverse the bottom layer, and obtain the final enhanced bottom layer. 4. The double-prior optimized underwater image contrast enhancement method according to claim 3, characterized in that: a nonlinear stretching function is used to further stretch the detail layer, and its expression is: Among them, σ is the stretch control reference, set to 0.88; is an enhanced layer of detail; Using the enhanced detail layer and the enhanced base layer we get an enhanced V channel, defined as: in, It is an enhanced V channel.