Quality Improvement of Endoscopy Videos

Abstract

We address two quality issues in endoscopy videos originating from: (a) Colour Channel Misalignment: In current endoscope systems, the colour channel images are most commonly acquired sequentially at different time instances. Whenever the camera moves significantly between acquisition instances, the colour channels get misaligned. This misalignment degrades the quality of the video due to the appearance of stroboscopic artefacts. (b) Specular Highlights: The surfaces of human organs, visualised in endoscopy videos, usually have a glossy appearance which causes specular highlights in the images. For many image analysis algorithms, these distinct and bright artefacts can become a significant source of error. In this paper, we propose two novel algorithms to remove the aforementioned artefacts. The colour channel misalignment artefacts are removed by estimating the camera motion in the time interval between the acquisition of two colour channel images. The issue of specular highlights is addressed using: (i) a segmentation method based on nonlinear filtering and colour image thresholding and (ii) a fast inpainting method. Both algorithms are evaluated using image sets extracted from colonoscopy videos. The colour channel realignment algorithm achieved a success rate of 86% and 78% in image sets with artificial and real misalignment, respectively. When specular highlights are a priori removed by the proposed algorithm, these success rates increase to 92% and 84%, respectively. The specular highlight removal algorithm achieved an accuracy of 91.99%.

Survey of Computer Vision and Machine Learning in Gastrointestinal Endoscopy

“…The following references give additional methods from this category proposed in literature: [263]-[270] 13) Specular Highlight Detection and Removal (SHD): Although, specular highlights in the image constitute UI regions, this particular category of methods attempt to not only identify such regions, but also correct them. In [271], specular highlights is addressed using:a segmentation method based on non-linear filtering and colour image thresholding followed by a fast in painting method. The proposed method in [272], aims to decouple the specular and diffuse components of endoscopic imagery in order to suppress specular reflectance.…”