Multiple connected residual network for image enhancement on smartphones

Abstract

Image enhancement on smartphones needs rapid processing speed with comparable performance. Recently, convolutional neural networks (CNNs) have achieved outstanding performance in image processing tasks such as image super-resolution and enhancement. In this paper, we propose a lightweight generator for image enhancement based on CNN to keep a balance between quality and speed, called multi-connected residual network (MCRN). The proposed network consists of one discriminator and one generator. The generator is a two-stage network: (1) The first stage extracts structural features; (2) the second stage focuses on enhancing perceptual visual quality. By utilizing the style of multiple connections, we achieve good performance in image enhancement while making our network converge fast. Experimental results demonstrate that the proposed method outperforms the state-of-the-art approaches in terms of the perceptual quality and runtime. The code is available at https://github.com/JieLiu95/MCRN.