Low-light image processing is an important task in computer vision, which is widely used in many fields. Due to the high image noise and blurred details in low illumination environment, it is difficult for conventional image enhancement techniques to effectively recover the image quality. In this paper, a method for automatic detail enhancement and visual effect improvement of low illumination images based on hierarchical image processing algorithm is proposed. The method combines Retinex theory and convolutional neural network, and realizes the effective enhancement of low illumination images by constructing the decomposition and enhancement module of reflection component and light component. In the experiments, LOL and LOLv2 datasets were used for subjective and objective evaluation, respectively. The results show that on the LOL dataset, the PSNR and SSIM of this method reach 25.42 and 0.911, respectively, and on the LOLv2 dataset, the PSNR is 23.09 and the SSIM is 0.932, which are better than other existing algorithms. In addition, the subjective evaluation also shows that the proposed method has obvious advantages in terms of noise removal, line clarity, color naturalness and image realism. The study shows that the method can effectively improve the detail performance and visual effect of low illumination images.
