A low-light image enhancement method based on conditional re-enhancement network

Abstract

The invention discloses a low-illuminance image enhancement method based on a conditional re-enhancement network, which belongs to the field of digital image processing. The invention aims to solve the problem that low contrast, low brightness, noise and color degradation cannot be processed simultaneously in the existing low-illuminance image enhancement method. The enhancement method proposed by the present invention includes a conditional re-enhancement network, the input of which is a low-illuminance image and its maximum value channel image and its expected maximum value channel image, and the output is the final enhanced image. The expected maximum channel image is obtained by adding blur and noise to the maximum channel image of the supervised image or tone-mapping the maximum channel image of the low-light image in the training phase, and the low-light image processed by any image enhancement method in the test phase The maximum value channel image. The invention can remarkably enhance the brightness and contrast of low-illuminance images, remove noise and reduce color distortion at the same time. The invention can be used for enhancement of low-illuminance images.

Description

technical field

[0001] The invention belongs to the field of digital image processing, and in particular relates to a low-illumination image enhancement method. Background technique

[0002] Cameras are important perceptual components of various types of unmanned equipment. However, in many low-light environments, such as nighttime, darkroom and other environments, the acquired images often have problems such as low contrast, low brightness, high noise, and color degradation. Recent deep learning-based methods have achieved good results in various image processing tasks. Currently, in the field of low-light image enhancement, deep learning-based image enhancement methods can be divided into two categories: supervised and unsupervised.

[0003] Unsupervised image enhancement method based on deep learning: This type of method does not require pairs of low-light images and normal-light images, so it does not need to invest too much in constructing training datasets, and has a h...

Examples

specific Embodiment approach 1

[0063] Embodiment 1: Combining figure 1 Describe this embodiment,

[0064] A low-light image enhancement method based on conditional re-enhancement network, comprising the following steps:

[0065] Step 1: Design a conditional re-enhancement network based on deep learning, and the conditional re-enhancement network can enhance the input low-illumination image S, and output the final enhanced image E;

[0066] Re-enhance the network by inputting the low-light image to be enhanced;

[0067] The input of the conditional re-enhancement network is the low-light image S and its maximum channel image S max and its expected maximum channel image S expect_max , S is a matrix of M*N*3, M is the number of rows, N is the number of columns, 3 is {r,g,b} three color channels, S max It is obtained by taking the maximum value of the three color channels, which is a matrix of M*N*1. In the test phase, S expect_max S can be enhanced by any image enhancement method such as histogram equaliz...

specific Embodiment approach 2

[0078] Specific implementation mode 2: Combining figure 2 Describe this embodiment,

[0079] A low-illuminance image enhancement method based on a conditional re-enhancement network described in this embodiment, the conditional re-enhancement network is specifically as follows:

[0080] The input is input to the first convolutional layer and the second convolutional layer respectively. The first convolutional layer and the second convolutional layer are 9*9 convolutional layers and 3*3 convolutional layers respectively;

[0081] The first convolution layer is connected to the third convolution unit, and the third convolution unit is a 3*3 convolution layer followed by a ReLU layer;

[0082] The third convolution unit is connected to the fourth convolution unit, the fourth convolution unit is connected to the fifth convolution unit, the fifth convolution unit is connected to the sixth convolution unit, the fourth convolution unit, the fifth convolution unit, the sixth convolu...

specific Embodiment approach 3

[0090]In the low-illumination image enhancement method based on the conditional re-enhancement network described in this embodiment, the specific process of the third step includes the following steps:

[0091] Step 31. Extract the maximum value channel image S of the low-illumination image S max :

[0092]

[0093] Among them, S max (i,j) is the maximum channel image S max The i-th row and j-th column elements; max represents the operation of taking the maximum value; c takes r, g, b, corresponding to the three color channels of red, green and blue in the rgb color space, S c (i, j) are the elements in the i-th row and the j-th column of a channel of the low-illuminance image S in the rgb color space;

[0094] Step 32: Extract the maximum value channel image H of the normal illumination image H max :

[0095]

[0096] Among them, H max (i,j) is the maximum channel image H max The i-th row and j-th column elements; max represents the operation of taking the maximu...

technical field

[0001] The invention belongs to the field of digital image processing, and in particular relates to a low-illumination image enhancement method. Background technique

[0002] Cameras are important perceptual components of various types of unmanned equipment. However, in many low-light environments, such as nighttime, darkroom and other environments, the acquired images often have problems such as low contrast, low brightness, high noise, and color degradation. Recent deep learning-based methods have achieved good results in various image processing tasks. Currently, in the field of low-light image enhancement, deep learning-based image enhancement methods can be divided into two categories: supervised and unsupervised.

[0003] Unsupervised image enhancement method based on deep learning: This type of method does not require pairs of low-light images and normal-light images, so it does not need to invest too much in constructing training datasets, and has a h...

Examples

specific Embodiment approach 1

[0063] Embodiment 1: Combining figure 1 Describe this embodiment,

[0064] A low-light image enhancement method based on conditional re-enhancement network, comprising the following steps:

[0065] Step 1: Design a conditional re-enhancement network based on deep learning, and the conditional re-enhancement network can enhance the input low-illumination image S, and output the final enhanced image E;

[0066] Re-enhance the network by inputting the low-light image to be enhanced;

[0067] The input of the conditional re-enhancement network is the low-light image S and its maximum channel image S max and its expected maximum channel image S expect_max , S is a matrix of M*N*3, M is the number of rows, N is the number of columns, 3 is {r,g,b} three color channels, S max It is obtained by taking the maximum value of the three color channels, which is a matrix of M*N*1. In the test phase, S expect_max S can be enhanced by any image enhancement method such as histogram equaliz...

specific Embodiment approach 2

[0078] Specific implementation mode 2: Combining figure 2 Describe this embodiment,

[0079] A low-illuminance image enhancement method based on a conditional re-enhancement network described in this embodiment, the conditional re-enhancement network is specifically as follows:

[0080] The input is input to the first convolutional layer and the second convolutional layer respectively. The first convolutional layer and the second convolutional layer are 9*9 convolutional layers and 3*3 convolutional layers respectively;

[0081] The first convolution layer is connected to the third convolution unit, and the third convolution unit is a 3*3 convolution layer followed by a ReLU layer;

[0082] The third convolution unit is connected to the fourth convolution unit, the fourth convolution unit is connected to the fifth convolution unit, the fifth convolution unit is connected to the sixth convolution unit, the fourth convolution unit, the fifth convolution unit, the sixth convolu...

specific Embodiment approach 3

[0090]In the low-illumination image enhancement method based on the conditional re-enhancement network described in this embodiment, the specific process of the third step includes the following steps:

[0091] Step 31. Extract the maximum value channel image S of the low-illumination image S max :

[0092]

[0093] Among them, S max (i,j) is the maximum channel image S max The i-th row and j-th column elements; max represents the operation of taking the maximum value; c takes r, g, b, corresponding to the three color channels of red, green and blue in the rgb color space, S c (i, j) are the elements in the i-th row and the j-th column of a channel of the low-illuminance image S in the rgb color space;

[0094] Step 32: Extract the maximum value channel image H of the normal illumination image H max :

[0095]

[0096] Among them, H max (i,j) is the maximum channel image H max The i-th row and j-th column elements; max represents the operation of taking the maximu...

Images