Deep Learning-Based Single Image De-Raining Using Discrete Hartley Transformation

Abstract

In computer vision, the removal of rain streaks from individual photographs has drawn a lot of interest. A saturated picture is broken down to an underlying image without any rain and a map of the rain streak to represent the de-raining challenge as an image decomposition assignment. This study introduced a DLD: Deep Learning-Based Single Image De-Raining Using Discrete Hartley Transformation, which is different from the majority of de-raining techniques now in use. The data cleansing phase of this study uses contrast-limited adaptive histogram equalization to smooth out the image and lower noise. Then, we introduced a new method to reconstruct a rainy image called Discrete Hartley Transformation (DHT). Following that feature extraction is carried by the proposed Deep Learning-based Enhanced Share-Source Residual Module (SSRM) which improves image performance also its shortcut connections. Finally, the Inverse Discrete Hartley Transformation (IDHT) is used to de-rained images. As a result, our proposed method achieves a high accuracy of 93.6, PSNR of 41.9, and SSIM of 0.96 compared to the existing techniques.