DEPTH-WISE CONVOLUTION ATTENTION AND MULTI-SCALE FEATURE FUSION NETWORK FOR LOW-LIGHT IMAGE ENHANCEMENT

This invention relates to the field of image processing technology, particularly to a low-light image enhancement method that utilizes deep-wise convolutional attention and multi-scale feature fusion. This invention presents a new low-light attention block (LLAB) and a multi-scale feature compensation block. The LLAB consists of a low-light multi-head self-attention block, a dual-branch equalization block, and two normalization layers. The low-light multi-head self- attention block is designed to extract semantic information from different channels and balance the feature weights between different channels by calculating the attention maps between channels, thereby improving the visibility and contrast of the image. The dual-branch equalization block further enhances the contrast of the image. A multi-scale feature compensation block is proposed to compensate for the loss of detail information in the image during the low-light attention block and down-sampling stage, and to fuse the deep spatial information of different scale images.