Low-Light Image Enhancement Based on Local Contrast Denoising and Fusion

The purpose of low-light image enhancement is to improve the clarity of objects in low-light environments to facilitate the recognition and detection of targets later. Local Contrast Denoising and Fusion Network (LCDFN) is constructed to address the problem of poor contrast and severe noise in low light images. Images obtained in low-light environments have low contrast in the image as a whole, but usually some of the edge regions of the image have larger contrast due to the presence of the light source, and enhancement processing of low-light images often results in over-enhancement of the region, generating noise. In order to suppress the noise, firstly, determines the region with larger contrast by local inspection of low-light/bright-light image pairs and suppresses the region's weighting value; second, multiple denoising is performed on the low-light image by changing the coefficients of the weight map; then, the multiple denoising results are fused; finally, the color loss function is improved, and the luminance loss is added to the color loss function, so that the LCDFN network can better inhibit the generation of noise. The experimental results show that LCDFN has a better effect on low-light image enhancement.