An adaptive smoothness constraint for computing optical flow on sequences of cardiac images

Different algorithms are available for processing an optical flow field between two successive frames. In most situations, two constraints are necessary to estimate such a displacement vector field. The first constraint is based on the gray level variations. The second one is usually based on the smoothness of the flow field. Unfortunately, the results of these constraints are not really adapted to the movements of the ventricle. Here, the authors propose a new method specifically adapted to sequences of cardiac images, for processing the optical flow. They calculate the amplitude image of a sequence of images by the Fourier transform. Unlike methods which identically smooth the optical flow in each direction, the authors developed a Fourier adaptive smoothness method which restricts the variation of the flow field in the directions with small variations of the Fourier amplitude. Moreover, the authors have generalized this method to sequences of non-cyclic images.