Rain Streak Removal via Spatio-Channel Based Spectral Graph CNN for Image Deraining

Abstract

Removing rain streaks from the captured single rainy images plays a dominant role in high-level Computer Vision (CV) applications. Since, many existing deraining methods ignores long range contextual information and utilize only local spatial information. To address this issue, a Spatio-channel based Spectral Graph Convolutional Neural Network (SCSGCNet) for image deraining was proposed and two new modules were introduced to extract representations along spatial and channel wise dimensions. Therefore, we integrate deep Convolutional neural network (CNN) with spatial based spectral graph convolutional neural network (SSGCNN) and channel based spectral graph convolutional network (CSGCNN) modules into a single end-to-end network. Therefore, our network was able to model feature representations from local, global spatial patterns and channel correlations. Experimental results on five synthetic and real-world datasets shows that the proposed network achieves state-of-the-art (SOTA) results.