Content-adaptive rain and snow removal algorithms for single image

Abstract

In this paper, we present two content-adaptive rain and snow removal algorithms for single image based on filtering. The first algorithm treats rain and snow removal task as an issue of bilateral filtering, where a content-based saliency prior is introduced. While the other views the same task from the perspective of guided-image-filtering, and the guidance image is derived according to the statistical property of raindrops or snowflakes as well as image background content. A comparative study and quantitative evaluation with some main existing image assessment algorithms demonstrate better performance of our proposed algorithms. The main contributions of our works are twofold: firstly, to the best of our knowledge, our algorithms are among the first to introduce image content information for single-image-based rain and snow removal; and secondly, we are also among the first to introduce quantitative assessment for single-image-based rain and snow removal tasks.