A Global Sparse Model with Singular Value Decomposition and Huber Loss Function for Rain Streak Removal
Rachapalli Sunanda, Joseph Beatrice Seventline · International journal of intelligent engineering and systems · 2020
The rainy effect normally decreases the visual quality of the images that highly affects the outdoor vision system's performance.Due to light scattering, the rain streaks generate the haziness and blurring effect.So, an effective model is required to remove rain streaks from the single image that assists an extensive range of applications such as object tracking, image enhancement, etc.In this research work, a new model has been proposed for rain streak removal in the single image.In the proposed work, a global sparse model with Singular Value Decomposition (SVD) and Huber loss function were used to remove rain streaks from the synthesized images.In the proposed model, three sparse terms (characteristics of image background information, structural knowledge and intrinsic direction of rain streaks) were used for depicting the directional smoothness of rain-free and rain streak images and also described the intrinsic and latent properties of rain streaks.The rain streaks sparsity was also enhanced by employing l_1 norm that effectively avoids the undesired features on the rain free regions.Then, Alternating Direction Method of Multipliers (ADMM) was used for tackling the proposed model in order to achieve optimal solutions.In the experimental phase, the proposed model was compared with the existing models like Cascading Attention Aggregation Network (CAAN) and Directional Global Sparse Model (DGSM) in terms of Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), and Root Mean Square Error (RMSE).Through the experimental simulation, the proposed model almost showed 0.52-1 dB improvement in PSNR value.