LLISP: Low-Light Image Signal Processing Net via Two-Stage Network

Images taken in extremely low light suffer from various problems such as heavy noise, blur, and color distortion. Assuming the low-light images contain a good representation of scene content, current enhancement methods focus on finding suitable illumination adjustment but often fail to deal with noise Recently, some works try suppress reconstruct raw data. But these apply network instead an image signal processing pipeline (ISP) map data enhanced results which leads learning burden for get unsatisfactory results. In order remove correct bias enhance details more effectively, we propose two-stage Low Light Image Signal Processing Network named LLISP. The design our is inspired by traditional ISP: multiple stages according attributes different tasks. first stage, simple denoising module introduced reduce noise. second two-branch texture details. One branch aims at correcting distortion restoring while another focuses recovering realistic texture. Experimental demonstrate that proposed method can high-quality replace ISP.