Research On Bleeding Detection Of Wireless Capsule Endoscopy Images Based On Color Feature Vector

Wireless capsule endoscopy is a new and revolutionary technique for detection of gastrointestinal problems. Compared with the traditional ways, the doctor who uses the capsule endoscope system can directly observe gastrointestinal image and obtain detailed, intuitive information. In recent years, due to the capsule endoscopy with easy to use, convenient to check, no pain, the capability of inspecting the whole small intestine, etc. the capsule endoscopy has become one of the hot topics in research field of medical equipment at home and abroad. However, WCE produces too many pictures of a patient. These images require an experienced doctor to browse one by one, which is a laboring and boring work,spent a lot of time. This paper aims to use computer based techniques to reduce the workload of doctors.In this paper, we mainly study the problem of bleeding detection algorithm for wireless capsule endoscope image. This algorithm is based on support vector machine as classifier, which is supervised learning method. In the phase of image preprocessing,our work is to extract the region of interest from the WCE images, eliminate the noise of the image, use CLAHE method to improve the image's local contrast and color correction. Base on the theory of several common color spaces and image's own characteristics, we use the A channel of LAB color space and the M channel of CMYK color space for red feature recognition. Then, Combining these two channels, two different color feature vectors are proposed. The first method is a two-stage significant extraction algorithm. The first stage uses A channel and M channel to extract binary feature vector and find bleeding area. According to the bleeding area, the second stage use the visual contrast in the RGB color space to extract a six dimensional vector and set threshold, normalization. This model is able to observe the distribution of the bleeding area, but the size of the feature depends on the size of the bleeding area. The second method also use A channel and M channel to extract the 2D binary feature,compared with the visual contrast of the first method, this feature can remove nonessential color regions and use non equal interval dimension to reduce the number of dimensions. In order to prevent the feature vector depending on the size of the bleeding region in the capsule image, the feature vector is processed by two values. Compared with first method, the second method has better effectiveness, less calculation spent. In the last chapter, we list a series of process methods and compare the experiment resultsbetween them for a better enhancement of the current algorithm.