PROJECT TITLE :

Scale-Free Single Image Deraining Via Visibility-Enhanced Recurrent Wavelet Learning

ABSTRACT:

Even in the presence of big rain streaks and rain streak accumulation, we solve a rain removal problem in this study (where individual streaks cannot be seen and thus are visually similar to mist or fog). For rain streak removal, a mismatch between the training and testing stages results in poor performance, particularly when there are huge streaks. For this, we use a wavelet transform hierarchical representation in a rain removal process: On the low-frequency component, the rain is removed, and on the high-frequency components, the recovered low-frequency component serves as a guide. Recurrent multi-scale wavelet transform-like design enables the proposed network to adapt to larger streaks, which considerably improves the removal of genuine rain streaks. Based on this concept, the recurrence recovery process may be explained. Multi-scale redundancy and the utilisation of context information across vast regions are enabled by the network's inclusion of several pathways with various receptive fields. To deal with heavy rain, we have developed a detail-appearing rain accumulation remover that not only improves vision but also enhances the details in dark areas. New models considerably outperform state-of-the-art methods based on the evaluation of synthetic and actual photographs, particularly those with huge rain streaks and high accumulations.