A multiscale algorithm for closed contour matching in image sequence

An algorithm for closed deformable active contours, so called snakes, is described and applied to speckle images. Using polar coordinates, a snake sequence is represented by a scalar function (radial distance) of two variables (angle and time). The cost function to be minimized includes terms for spatial and temporal smoothness and for image forces. The latter terms make the contour attracted to bright structures and spatio-temporal edges. The snake search is carried out on a coarse-to-fine basis and relies on dynamic programming. The temporal smoothness is imposed on the entire image sequence simultaneously rather than sequentially. The method is edge orientation selective so that the snake is attracted only to positive edges in the direction of the normal to it curvature. The performance of the algorithm is investigated by sequences of simulated images and also echocardiographic images