Low-light image enhancement method and device based on wavelet transform and Retinex-Net

A low-light image enhancement method based on wavelet transform and Retinex-Net, comprising:

• decomposing a to-be-processed image into a low-frequency component and a high-frequency component by using discrete wavelet transform;
• inputting the low-frequency component and a normal light image into a decomposition network to obtain an illumination component and a reflection component of the low-frequency component and the normal-light image respectively, wherein the normal-light image is a normally-exposed image captured for a same scene, same object, and same lighting environment as the to-be-processed image;
• denoising the reflection component of the low-frequency component to obtain a denoised reflection component;
• inputting the reflection component and the illumination component of the low-frequency component into an enhancement network to enhance the illumination component of the low-frequency component, to obtain an enhanced illumination component;
• multiplying the enhanced illumination component and the denoised reflection component element by element to obtain an enhanced low-frequency component, and transferring the enhanced low-frequency component from a RGB space to a HSV space to obtain a first hue component H1, a first saturation component S1, and a first value component V1;
• fusing the high-frequency component of the to-be-processed image based on regional characteristics and transferring the fused high-frequency component from the RGB space to the HSV space to obtain a second hue component H2, a second saturation component S2, and a second value component V2;
• using a value fusion technology of the wavelet transform to fuse the first value component V1 and the second value component V2 to obtain a third value component V and stretching the third value component V to obtain a fourth value component V′;
• and
• using a wavelet inverse transform to fuse the first hue component H1 and the second hue component H2 into a third hue component H, and to fuse the first saturation component S1 and the second saturation component S2 into a third saturation component S, and reducing the third hue component H, the third saturation component S, and the fourth value component V′ into the RGB space to obtain an enhanced image.