DWT-RT: A Lightweight Image Deraining Model Based on Discrete Wavelet Transforms
Yongliang He, Qinyou Zhou, Yong Chen, Yingjie Cui · 2024
This study introduces a novel image deraining model based on the Transformer architecture, named DWT-RT, which incorporates Discrete Wavelet Transform (DWT), a recursive Transformer structure, and gating mechanisms to enhance deraining performance while reducing computational resource consumption. The model was tested on three public datasets: Rain800, Rain100L, and Rain100H. In these tests, DWT-RT was compared with current state-of-the-art models using PSNR and SSIM metrics, demonstrating superior performance in most cases, particularly achieving the highest PSNR value on the Rain100L dataset. Ablation experiments confirmed the effectiveness of the discrete wavelet decomposition module and the residual gated recursive module. The findings indicate that DWT-RT can achieve efficient image deraining with fewer parameters and lower computational resource requirements, highlighting the model's potential application in edge computing scenarios.