Research And Application Of Low-Light Image Enhancement Algorithm Based On Retinex Theory

Under conditions such as insufficient light,uneven or shadow occlusion,the captured images generally have problems such as excessive noise and weak contrast,resulting in poor visual performance of the image.By enhancing low-light images,image quality can be effectively improved and system performance can be improved.To this end,the following work is done: the wellwall image is collected and preprocessed,and the wellwall image dataset(Wall dataset)is made;For the characteristics of low-light images,the networks RANet and RAR-UNet based on Retinex theory are proposed,and the well wall image enhancement system is constructed,as follows:(1)An enhanced network RANet based on Retinex theory was proposed in view of the uneven illumination distribution and easy overexposure of low-light images.According to Retinex theory,convolution and residual modules are first used to decompose the image into reflected component and irradiated component.Then U-Net network combined with SE attention is used to de-noise the reflected component and enhance the irradiated component.Finally,the enhanced irradiated component and the de-noised reflected component are fused.Get a new image that matches the visual senses.The experimental results show that RANet not only effectively improves the image brightness and contrast on the standard data set,but also greatly improves the objective indicators compared with other algorithms.However,in the self-made well wall data set,the enhancement effect is not very outstanding,and there is still a lot of room for improvement.(2)In order to further improve the enhancement effect of RANet on low-light image of borehole wall,the following optimization is made based on the characteristics of low-light image of borehole wall: HS-Block is used to solve the problem that the borehole wall image extracted from decomposition network contains less feature information;ECA attention mechanism and void convolution are used to solve the problems of detail loss and noise in enhanced images.CBAM attention module and RDB module are used to solve the problems of low contrast and blurred boundary of the enhanced well wall image.The experimental results show that,compared with RANet and other algorithms,RAR-UNet has the highest objective evaluation index PSNR and SSIM in both standard data and Wall data set,and is subjectively more consistent with human visual viewing effect.(3)A borehole wall image enhancement system is designed to collect low-light images from the camera,and then use the master control system to preprocess and transmit them to the software enhancement system,and use the RANet and RAR-UNet models in the software system to enhance the low-light images.Figure [52] Table [7] Reference [88]...