SINGLE IMAGE RAIN REMOVAL BASED ON IMPROVED ALTERNATING DIRECTION METHOD OF MULTIPLIERS AND SINGULAR VALUE DECOMPOSITION

In recent decades, the performance of outdoor vision systems are degraded due to the blurring, visibility obstruction and deformation caused by raindrops. The rain streak removal in a single image is a more challenging task, due to the unavailability of spatial and temporal information. In this research paper, a new method proposed for sparse and heavy rain streak removal is presented. The method uses an improved Alternating Direction Method of Multipliers (ADMM) to eliminate rain streaks from images in combination with Singular Value Decomposition (SVD), texture matching loss and edge aware loss. The ADMM technique uses the characteristics of structural knowledge, intrinsic direction and background information of the rain streaks for characterizing the smoothness of rain streak and free images. The combined SVD, edge aware loss and texture matching loss are utilized for ensuring the overall perceptual quality and texture generation of the image and also to describe the latent and intrinsic characteristics of the rain streaks. In the experimental section, it was observed that with the proposed work an improvement in PSNR which is minimum of 0.6164 dB and maximum of 1.5354 dB on rain100L and rain100H datasets was achieved compared to existing work.