ConvNet Based Single Image Deraining Methods: A Comparative Analysis

Deraining essentially is removing the visual effect of rain from the captured video or image. It is broadly classified into two fields video deraining and image deraining. Rainy images are a complex blend of base image layer and rain layer. Captured images having drops of rain suffer from uneven brightness where regions occupied by rain are overexposed compared to other regions. After the advent of deep learning, there has been a renewed interest in developing more effective approaches. Many deep CNN-based approaches have been devised to effectively derain images. We have summed up these approaches while focusing on the ConvNet based ones to establish part by part comparison of each phase of the rain removal process: image preparation, model design, basic blocks, training. In this paper, we have analyzed the performance of the considered approaches on both qualitative and quantitative grounds.