Low Latency Mitigation of Rain Induced Noise in Images

Sometimes, weather conditions, particularly rain and snow, makes it difficult to obtain clear image or videos. Partial occlusion by moving rain drops makes it difficult to process such videos for example in object recognition and feature tracking. The difficulty further increases in presence of camera and objects motion. Motion analysis is typically performed to avoid distortions in camera and object motion. This paper proposes an image processing procedure to carry out rain removal in image sequences without explicit motion analysis. The proposed method analysis temporal changes (with 1 frame latency) and selects rain corrupted image pixels. The selected pixels are further filtered for camera and object motion by using spatio-temporal consistency constrains. The corrupted pixels are recovered by intensity subtraction. A naive selection of algorithms shows comparable results. The proposed scheme has various applications for example outdoor surveillance, feature tracking and object recognition.