DEEP LEARNING NETWORK FOR LOW-LIGHT IMAGE ENHANCEMENT

Abstract

Low-light image enhancement is a challenging task that has attracted considerable attention. Pictures  taken in low-light conditions often have bad visual quality. To address the problem, regard the  low-light enhancement as a residual learning problem that is to estimate the residual between low- and  normal-light images and  propose a novel Deep Lightening Network (DLN) that benefits  from the recent development of Convolutional Neural Networks (CNNs). The proposed DLN consists  of several Lightening BackProjection (LBP) blocks. The LBPs perform lightening and darkening  processes iteratively to learn the residual for normal-light estimations. To effectively utilize the local  and global features, propose a Feature Aggregation (FA) block that adaptively fuses the results  of different LBPs and  evaluate the proposed method on different datasets. Numerical results show that  our proposed DLN approach outperforms other methods under both objective and subjective metrics.