Accurate analysis of wireless capsule endoscopy (WCE) videos is vital but tedious. Automatic image analysis can expedite
this task. Video segmentation of WCE into the four parts of the gastrointestinal tract is one way to assist a physician. The
segmentation approach described in this paper integrates pattern recognition with statiscal analysis. Iniatially, a support
vector machine is applied to classify video frames into four classes using a combination of multiple color and texture
features as the feature vector. A Poisson cumulative distribution, for which the parameter depends on the length of segments,
models a prior knowledge. A priori knowledge together with inter-frame difference serves as the global constraints
driven by the underlying observation of each WCE video, which is fitted by Gaussian distribution to constrain the transition
probability of hidden Markov model.Experimental results demonstrated effectiveness of the approach.