Clustering of a set of identified points on images of dynamic scenes, based on the principle of minimum description length

This paper discusses the task of separating a set of identified reference points on a pair of images of a dynamic scene, using clusters that correspond to observed moving objects. Based on the principle of minimum description length, a criterion is proposed that makes it possible to choose between different classes of transformations and to estimate the clustering quality. This criterion can be used to choose the optimum model of a spatial transformation for each discriminated cluster and to avoid being influenced by outliers in the form of incorrectly identified points. (C) 2010 Optical Society of America.