First order absolute moment in contour tracking

A procedure to outline contours automatically in temporal sequences of cardiovascular images is presented. The contour determined on the nth frame of the sequence is used as the starting contour to determine the contour on the (n+1)th frame. When given a starting contour, standard edge detectors can efficiently locate only ideal discontinuities with step shape. First and second Gaussian derivatives, computed at points pi of the starting contour, directly provide amplitude and direction of the vectors which join pi to the respective points of the final contour. Conversely, if the discontinuity does not show an ideal profile with step shape, the computation of the Gaussian derivatives must be carried out at all points between the starting and the final contour. In the paper, a property of the first order absolute moment is exploited to develop a robust and efficient iterative procedure which can be used to locate discontinuities that do not show an ideal step profile.