Abstract:
Visual quality of rainy images are considerably poor due to the raindrops in camera lens and the rain streaks in the background scenes. Although the raindrops and rain streaks are appeared together in real-world rainy images, most of the previous approaches are proposed to remove either of them. In this paper, we have proposed a novel CNN model architecture to remove raindrops and rain streaks together. The proposed CNN model architecture has two branches and it consumes two formats of a rainy image via an encoder-decoder network and a dense CNN network. At the end of the architecture, outputs of both branches are combined to produce a high-visibility rain-free image with natural colours. In addition, internal and external skip connections are introduced in the blocks of these branches to improve the performance further. The proposed model is trained and then tested on RainDrop, Rain100H, Rain100L, and Rain12 benchmarks and showed excellent performance than the state-of-the-art approaches.
