Retinex-based Illumination feature fusion network for low-light image enhancement

Vision is the main means for human beings to acquire external information, and with the arrival of the wave of informationization, digital images have become an important carrier of information dissemination. Due to the complexity of the imaging environment, as well as the lack of hardware conditions of the imaging equipment, the directly acquired images are often of poor quality, limiting their application in actual scenes, while the lack of light during the image acquisition process will greatly reduce the visibility of the image. Existing low-light image enhancement techniques are difficult to balance quality and efficiency, and are ineffective in complex scenes. In this paper, an image decomposition model is constructed based on Retinex theory to reduce the uncertainty of decomposition. In order to recover the degradation process existing in the reflection image, a light feature map is introduced to serve as a guide, and an image enhancement network module is constructed by using a light feature fusion module, and at the same time, a light tuning network is designed to adaptively adjust the lighting conditions of the light map. Experiments on a benchmark dataset show the effectiveness and rationality test of the method.