Motion estimation of the endoscopy capsule using region-based Kernel SVM classifier

Wireless Capsule Endoscopy (WCE) allows physicians to examine the entire digestive system without any surgical operation. Although it provides a noninvasive imaging approach to access the gastrointestinal (GI) tract, the biggest drawback of this technology is its incapability of localizing the capsule when an abnormality is found by the video source. Existing localization methods based on radio frequency (RF) and magnetic field suffer a great error due to the non-homogeneity of the human body and uncertain movement of the endoscopic capsule. In this paper, we developed a novel image classification technique to analyze the motion of the capsule. The proposed method segments the endoscopic images into sub-regions and classified them using Kernel Support Vector Machine (K-SVM). Our method performs better than the traditional pixel based classification methods since the quantized feature vector is able to better represent the image due to its natural resistant characteristic against the noises. Besides, the Kernel function is able to map the low dimensional feature vectors to higher dimensional space to form a non-linear decision hyper-plane. Experimental results show that the proposed method is able to reach a high accuracy of 92%.