Template matching used in time contextual segmentation

Presents a time contextual segmentation algorithm based on construction of an `object model´ using evidence from several consecutive frames. The aim is to obtain more `correct´ and stable segments in low contrast image sequences. Roughly speaking, for each object to be segmented the algorithm generates an object model as an aggregate of segments derived from the most recent frames. Due to object motion, the frame-to-frame displacement must be determined. This is done by a template matching technique. The segmentation is terminated when the object model has converged to a stable segment. Good results have been obtained in situations where single frame based algorithms due to either low local contrast or partial occlusion, have failed. The method may be implemented for real time applications