Transform-based underwater image enhancement method

Abstract

The invention relates to an underwater image enhancement method based on a Transform. The invention relates to the technical field of digital image processing and underwater image processing, and the method comprises the steps: inputting a to-be-enhanced underwater image through an input end, carrying out the down-sampling through a convolution mode through employing a multi-stage down-sampling module, and obtaining images under different scales; the global feature extraction module is used for extracting features of images under different scales in a convolution and residual block combined mode, and feature maps under different sizes are obtained; inputting the feature map with the minimum size into an encoder, and processing the feature map in a multi-level feature deepening extraction mode; block decoding is carried out on the output of the decoder, and the minimum-size feature map is recovered from the sequence to the image form again through the output of the Transformer; scale recovery is carried out through a multi-stage up-sampling module, and an output enhanced image is obtained in combination with the previous multi-stage detail deepening module.

Description

technical field

[0001] The invention relates to the technical fields of digital image processing and underwater image processing, and relates to a Transformer-based underwater image enhancement method. Background technique

[0002] The ocean is rich in natural resources, and underwater optical observation is an important means for human beings to understand the ocean and use ocean resources for development. Underwater optical image is a very important part of it, it can carry rich underwater information, and it plays an important role in fishery, aquatic products and military. However, the underwater environment is different from the terrestrial environment. There will be a lot of interference in the underwater imaging process. When the light passes through the water, it will be absorbed and scattered by the water, resulting in attenuation, which will affect the image color and texture characteristics of optical imaging. produce degradation. As the water depth increases, t...

Examples

specific Embodiment 1

[0029] according to Figure 1 to Figure 4 As shown, the specific optimization technical scheme taken for solving the above-mentioned technical problems is: the present invention relates to a Transformer-based underwater image enhancement method:

[0030] A Transformer-based underwater image enhancement method, comprising the following steps:

[0031] Step 1: Input the underwater image to be enhanced through the input terminal, and use the multi-level down-sampling module to down-sample through convolution to obtain images at different scales;

[0032] The step 1 is specifically:

[0033] Convolution kernels of different sizes are used for step-by-step downsampling, and the size of the original image is gradually reduced. While obtaining the original images at different scales, the memory required for subsequent Transformer processing is limited.

[0034] Step 2: Through the global feature extraction module, the method of convolution and residual block is used to extract the ...

specific Embodiment 2

[0046] Step 1, the image to be enhanced is input to the underwater image enhancement processing network through some pre-processing;

[0047] Step 2, using the underwater image processing network to enhance this type of degraded underwater image;

[0048] Step 3. Output and obtain the final enhanced result map

[0049] The first step is specifically as follows: firstly, preprocessing is performed on the optical image to be enhanced (underwater optical image) obtained from various sources. Since the size of the input part of the network is fixed, the size of the input image needs to be processed into an appropriate size .

[0050]The second step specifically includes: using the underwater image enhancement network in the present invention to perform image enhancement. What is designed in the present invention is actually an end-to-end network, and the effect to be obtained is to directly obtain an output enhanced image after inputting the image into the network. The main str...

technical field

[0001] The invention relates to the technical fields of digital image processing and underwater image processing, and relates to a Transformer-based underwater image enhancement method. Background technique

[0002] The ocean is rich in natural resources, and underwater optical observation is an important means for human beings to understand the ocean and use ocean resources for development. Underwater optical image is a very important part of it, it can carry rich underwater information, and it plays an important role in fishery, aquatic products and military. However, the underwater environment is different from the terrestrial environment. There will be a lot of interference in the underwater imaging process. When the light passes through the water, it will be absorbed and scattered by the water, resulting in attenuation, which will affect the image color and texture characteristics of optical imaging. produce degradation. As the water depth increases, t...

Examples

specific Embodiment 1

[0029] according to Figure 1 to Figure 4 As shown, the specific optimization technical scheme taken for solving the above-mentioned technical problems is: the present invention relates to a Transformer-based underwater image enhancement method:

[0030] A Transformer-based underwater image enhancement method, comprising the following steps:

[0031] Step 1: Input the underwater image to be enhanced through the input terminal, and use the multi-level down-sampling module to down-sample through convolution to obtain images at different scales;

[0032] The step 1 is specifically:

[0033] Convolution kernels of different sizes are used for step-by-step downsampling, and the size of the original image is gradually reduced. While obtaining the original images at different scales, the memory required for subsequent Transformer processing is limited.

[0034] Step 2: Through the global feature extraction module, the method of convolution and residual block is used to extract the ...

specific Embodiment 2

[0046] Step 1, the image to be enhanced is input to the underwater image enhancement processing network through some pre-processing;

[0047] Step 2, using the underwater image processing network to enhance this type of degraded underwater image;

[0048] Step 3. Output and obtain the final enhanced result map

[0049] The first step is specifically as follows: firstly, preprocessing is performed on the optical image to be enhanced (underwater optical image) obtained from various sources. Since the size of the input part of the network is fixed, the size of the input image needs to be processed into an appropriate size .

[0050]The second step specifically includes: using the underwater image enhancement network in the present invention to perform image enhancement. What is designed in the present invention is actually an end-to-end network, and the effect to be obtained is to directly obtain an output enhanced image after inputting the image into the network. The main str...

Images