Research on Single Image Deraining Algorithm Based on Euclidean Distance with Adaptive Progressive Residual Network
Xingzhi Chen, Ruiqiang Ma, Zehui Dong · 2021
In rainy days, rain streaks have various directions, shapes and densities. In order to achieve the purpose of rain removal, the deep network structure becomes complicated, thus reducing the efficiency of rain removal. The same model have an incomplete rain streaks removal problem for various rainy images. This paper proposes the adaptive progressive residual network (APRN) to solve the above problems. APRN used recursive method to achieve progressive rain streaks removal. We used the Euclidean distance to determine the distance between the derain images, and designed the Euclidean distance ratio (EDR) threshold to stop the network recursion and output final derain image. EDR compares the Euclidean distance of the de-rain images in adjacent stages, and controls the network recursion and helps the network processes rainy images more efficiently. Experiments show that APRN has better visual effects on different datasets, and the processing efficiency is improved.