Desmoking Laparoscopy Surgery Images Using an Image-to-Image Translation Guided by an Embedded Dark Channel

Abstract

In this paper, a method to remove the smoke effects in laparoscopic images is presented. The proposed method is based on an image-to-image conditional generative adversarial network endowed with a dark channel's embedded guide mask. The obtained experimental results were evaluated and quantitatively compared with desmoking state-of-art methods using the Peak Signal-to-Noise Ratio (PSNR) metrics and Structural Similarity (SSIM) index. Those results throw an improved performance compared with relevant works. Also, the processing time required by our method is 92 frames per second; a processing time that sets the foundation for a possible real-time implementation in a more modest embedded system.