Paired-Data Tranfromations for Weather-Affected Images
Jung Hwan Kim, Dong Seog Han · 2024
Extreme weather conditions may hinder predicting object detection through the camera sensors. Snow, dust, and rain particles can cause degraded areas and obstruct the camera vision, even if a human being is still recognizable. To eliminate the hindering particles, we apply the denoising techniques to restore the images without any weather-affected. However, the images may require the proper neural network training, and the hindering particles must be recognized. In this paper, we propose paired-data transformations to the pix2pix architecture to recognize the hindering particle effects. After applying paired-data transformations, the structural similarity index measure (SSIM) surpassed 92%, and the percentage showed 16.21% improvement.