This paper investigates the problem of detecting bleeding regions in images acquired from endoscopic examinations of gastrointestinal tract. The purpose is to identify the characteristic features of bleeding areas' contours in order to develop an accurate method for discriminating between true bleeding regions and missed detections, which could lead to a significant reduction of the false alarm rate of existing blood-detection algorithms. The research was carried out on a set of 40 unique images representing clear, bleeding regions, supplemented with precise markup of bleeding areas drawn by a physician. Another 40 images without blood presence were used, from which non-bleeding regions were acquired by gathering incorrect detections of an exemplary blood- detection algorithm. The contours of both types of regions were analysed by extracting the image area within a given distance from the region border. For describing the nature of the contours, Canny edge detector was used along with several statistical measures. The most descriptive feature enabling discrimination between true and false bleeding areas was identified. Finally, the possibility of utilization of the discovered relations for improving existing blood- detection algorithms is investigated and a method of their application is presented.
Authors
Additional information
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2012
