GastroIntestinal Color Space and Its Application to Image Enhancement

概要

論文の詳細を見る
Color is a robust feature in computer assisted diagnosis of Video Capsule Endoscopy (VCE). This paper proposes a color space, specific to the human Gastrointestinal (GI) tract. A statistical model is constructed from a large VCE dataset, in which color components are classified into three groups: GI wall, darkness, and noise. The main cues for the clustering task are visualizing the appearance of color components and the frequency thereof. To utilize these features, a Self-Organizing Map (SOM) is deployed in a two stage semisupervised scheme. We show the effectiveness of the proposed GI space in an application of image enhancement. With prior information about available colors from the GI color space, the enhancement task turns to increasing to the maximum color contrasts in the color range of the original image. In this sense, the enhanced image preserves the natural color tones of the original image as far as possible. The experimental results show that 84% of the enhanced images exceed the baseline level and are feasible for deployment in clinical applications.
2010-08-29