Development of progressive GAN-based image rain removal technique

Abstract

Deraining is a significant challenge in the field of computer vision, with the primary goal of removing rain streaks from images to improve visibility. This thesis introduces an innovative solution to deraining by employing a Progressive Generative Adversarial Network (GAN)-based deraining model. The proposed model comprises a generator and a discriminator that collaborate in an adversarial manner, continuously refining the deraining process. The generator is responsible for generating more realistic and visually appealing rain-free images, whille the discriminator becomes increasingly proficient at distinguishing between real and generated images. Initially, the U-Net architecture is utilized as the generator model. To further enhance the model's performance, a pair of U-Nets is intentionally used as the generator, resulting in a more effective deraining process. The Rain13K dataset is employed for training and evaluation, offering a wide range of rainy scenarios to improve the model's ability to generalize to unseen rainy images. The model's performance is evaluated on benchmark datasets, including Rain100L, Rain100H, Test100, Rain1200, and Test2800, and compared against state-of-the-art deraining methods. Objective metrics, such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), are used to quantitatively assess image quality. This comprehensive evaluation allows for an analysis of the model's effectiveness and the identification of the optimal generator architecture for deraining tasks.