Colour Variation Minimization Retinex Decomposition and Enhancement with a Multi-Branch Decomposition Network

This paper proposes a colour variation minimization retinex decomposition and enhancement with multi-branch network (CvmD-net) to remove single image darkness. The overcomes the problem that deep learning model relies on matching bright images process dark images. Specifically, our method takes two stages light up darkness in initial images: brightness optimization. We propose an input constant feature prior mechanism (ICFP) based reflection features. extracts structure from constrains reflected output reduce color distortion artifacts. noise amplification during is addressed by network. Sub-networks different structures are employed focus prediction tasks. reference for brightness. optimizes distribution calculating of Experimental results benchmark datasets, namely, LOL ZeroDCE, demonstrate proposed can better balance dense interference restoration. For evaluation real images, we collect Skynet at night verify performance approach. Compared state-of-the-art non-reference decomposition-enhancement models, this has best