부산대학교 도서관

Rain streaks in the air can be harmful for the image quality, which makes photograph blur and unaesthetic. Pictures with rain streaks are also bad resource for target detection and recognition. Therefore, we promote a method to remove rain streaks based on generative adversarial networks. We find that the rain streaks exit mainly in high-frequency component of a picture. Therefore, our work focuses on the part contains the rain streaks and avoid affecting details and color information of the picture. We also introduced a quantization method to improve network to make it thinner. Based on experiment, rains in the picture are removed effectively. Images after processing are cleaner than the original ones.