SMOKEBLITZ : Smoke Removal from Images using Generative Adversarial Networks

Abstract

"The proposed approach is done in order to eliminate smoke from the smoke-filled images in a critical situation such as fire. The presence of smoke in the images can obscure details and compromise image quality. This could make it difficult for safety monitoring and rescue operations professionals. Recent studies have focused on removing noise, enhancing resolution, and restoring damaged areas of deteriorated images, including smoke-filled ones. However, smoke's dynamic and structure less nature presents a significant challenge to these efforts. This research will focus on developing image blind inpainting technique with GANs to reconstruct larger missing portions of the image and restore colour. The goal is to improve the visibility and quality of smoke-filled images, which will aid in safety monitoring, evacuation procedures, and firefighting efforts. The initial results of the system were calculated according to the evaluation metrics, such as the SSIM, PSNR and BCE score, to calculate the system's performance. And received an average SSIM accuracy of around 0.78,PSNR score 27.46 and BCE 0.69 as of the prototype and further the author will take more actions to increase the accuracy score."