A priori-guided multi-layer rain-aware network for single image deraining

Abstract Rain removal from single image is a big challenge due to the limited information in a single image. In this work, we present a priori-guided multi-layer rain-aware convolutional neural network based deraining model, which can learn the density and directional details of rain efficiently. The density perception is implemented by the concatenation of multi-layer detail learning units. The direction perception is implemented by introducing the guided directional priors operators for detail learning network. In order to better restore clean image, the signal-to-noise ratio and structural similarity based loss function is utilized. Numerical experiments on synthetic and real rainy images are given to investigate the performance of proposed method. Comparison with recent state-of-the-art methods suggest that the proposed method can remove rain steaks and protect details of background efficiently.