Low Illumination Image Enhancement Algorithm Based on Attention Mechanisms and Multiscale Perception

A low-light image enhancement method combining the attention mechanism and multi-scale perception is proposed while stably enhancing the brightness and retaining the detailed information. The algorithm is based on Retinex theory to decompose the image into a reflection component and a light component, and then reconstruct the two components after enhancement. Inspired by the pyramid structure, a multi-scale perception module is designed with the characteristics of the human visual system to improve the feature extraction ability of the model for images of different scales. Combining the multi-scale perception module and the attention mechanism to construct a reflectance enhancement network using parallel convolutional kernel to enhance the model's ability to process information in the region of image luminance change. This enables the model to enhance images at different scales in different illuminated scenes. The experimental results show that the proposed algorithm has better result in structural similarity index and naturalness image quality evaluator. Processed images maintain a high degree of naturalness while improving brightness, contrast and preserving texture detail.