A Method for Modeling and Segmentation of Spatio-Temporal Shapes

This paper presents a method for modeling and segmenting spatio-temporal shapes. The modeling part is based on obtaining a description of the statistical variations of spatio-temporal shape parameters by studying a representative training set of examples. A deformable model of spatio-temporal shapes is used for segmenting similar shapes in new image sequences. The deformations of the model are driven by image features and are forced to produce results that are similar to what was found in the training set. The proposed deformations of the spatiotemporal shape are chosen to minimize a cost function using a dynamic programming approach. The preliminary results presented show that the method succeeds in segmenting complex spatio-temporal shapes in synthetic noisy data.