A rain detection and removal algorithm based on rainy intensity for videos in heavy rainy scene

Abstract

Detection and removal of rain for videos is a challenging problem because of the difficulty to build an appropriate imaging model for rainy weather. By far, almost all the algorithms proposed for rain removal considered a certain pixel in a video frame as rain affected pixel or non-rain affected pixel determinedly, which is not reasonable for heavy rainy scene. Because in the heavy rainy scene, almost all the pixels are covered by rain more or less. Instead of strictly considering a pixel as either rain covered or non-rain covered, the proposed algorithm calculates the rainy intensity for each pixel. The rainy intensity is soft information ranging from zero to one which contains more information about the rain. Furthermore, the rainy intensity can help estimate the original value of pixels that are covered by rain. Experiment results show that the rain removal videos processed by the proposed algorithm has much better vision effect than by other existing algorithms in heavy rainy scene.