Joint desmoking and denoising of laparoscopy images

Laparoscopic images in minimally invasive surgery get corrupted by surgical smoke and noise. This degrades the quality of the surgery and the results of subsequent processing for, say, segmentation and tracking. Algorithms for desmoking and denoising laparoscopic images seem to be missing in the medical vision literature. This paper formulates the problem of joint desmoking and denoising of laparoscopic images as a Bayesian inference problem. It relies on a novel probabilistic graphical model of the images, which includes novel prior models on the uncorrupted color image as well as the transmission-map image that indicates color attenuation due to smoke. The results on simulated and real-world laparoscopic images, including clinical expert evaluation, shows the advantages of the proposed method over the state of the art.