Low-light Image Enhancement Algorithm Based on Adaptive Feature Decomposition and Parallel Dual Extractor

Many existing methods based on Retin-ex model mainly rely on convolution layer to process input features, which-h lack dynamic adjustment ability and are easily disturbed by redundant information. At the same time, when the local features are transferred to Transformer for modeling, the detailed information may be lost, which will lead to the decline of the visual quality of the image. In this paper, an Adaptive Feature Decomposition and Parallel Double Extractor (AFD-PDE) method is proposed to solve the problems of insufficient brightness, loss of details and increased noise in the process of weak light image enhancement. By combining the weak light image with illumination prior, this method extracts illumination features and illumination maps by convolution operation and channel attention mechanism, thus realizing adaptive feature decomposition. Then, the extracted features are denoised by Parallel Double Extractor module, and the image details are restored and noise is eliminated by down-sampling and up-sampling layer by layer. Finally, a high-quality image enhancement effect is achieved by combining Gamma correction and feature fusion. The experimental results verify the superior performance of AFD-PDE method in weak light conditions, and show its broad application prospects in the field of image enhancement.