Multi-scene image enhancement based on multi-channel illumination estimation

The current mainstream algorithms cannot effectively take into account different luminance areas for image enhancement under different lighting conditions. Although there are many excellent algorithms for low-light image enhancement, the results are less satisfactory when those algorithms are applied to different exposure scenes of images. To solve the above problems, this paper proposes an image enhancement algorithm that can adapt to different exposure conditions. First, we introduce two forward and reverse dual-channels as initial illumination maps, which can effectively take into account the different exposure areas in an image. For the lack of details in enhanced images, we use a guided filter to pre-process the illumination map for retaining more detailed information. Second, the illumination map estimated by an optimization function will cause some distortion in fused images, so we propose an improved adaptive gamma correction method. At last, these images are fused by multi-scale exposure to obtain high-quality images. The exposure fusion allows our algorithm to cope with more scenes (shadows, overexposure, etc.). Experiments show that the proposed algorithm is not only effective in processing low-light images but can also be adapted for images in different exposure environments without deliberate adjustment of parameters. The proposed algorithm has a good performance in low-light, shadow, and overexposure scenes. Compared with other algorithms, the enhanced images are subjectively more realistic, natural, and maintain more complete details. Besides, the proposed algorithm outperforms other algorithms in comparison to several objective evaluation indicators.