Motion-Based Post Processing of Deformable Contours

We present a framework for tracking contours in 2D image sequences by taking into account the motion information and correspondence between contour points. With this framework, The tracking problem can be divided into two steps: i) Contour detection using general internal and external energies in separate frames, and ii) Tracking and refining contours using motion information of the entire image sequence. In addition to the traditional internal and external energies, we introduce two new constraints in the temporal domain which we call temporal internal energy and temporal external energy in the post processing stage. The temporal internal energy smoothes the contours by minimizing the velocity difference between neighboring contour points and the temporal external energy pushes the contour points toward their best location according to their corresponding points in the preceding image. We show the performance of the new constraints with synthetic image sequences and also present the post processing results for medical image sequences.