A CORRELATION-RELAXATION-LABELING FRAMEWORK FOR COMPUTING OPTICAL-FLOW - TEMPLATE MATCHING FROM A NEW PERSPECTIVE

Optical flow estimation is discussed based on a model for time-varying images more general than that implied in the work of Horn and Schunk [21]. The emphasis is on applications where low contrast imagery, non- rigid or evolving object patterns movement, as well as large interfram e displacements are encountered. Template matching is identified as ha ving advantages over point correspondence and the gradient-based appro ach in dealing with such applications. The two fundamental uncertainti es in feature matching procedures, whether it is template matching or feature point correspondences, are discussed. Correlation template mat ching procedures are established based on likelihood measurement. A ne w method for determining optical flow is developed by combining templa te matching and relaxation labeling. In this method, a number of candi date displacements for each template and their respective likelihood m easures are first determined. Then, relaxation labeling is employed to iteratively update each candidate's likelihood by requiring smoothnes s within a motion field. Real cloud images taken from meteorological s atellites are used to test the usefulness of this method. It is shown in this application that the new method can deal effectively with the uncertainty of multiple peak (multi-modal) correlation surfaces encoun tered in template matching. The results show significant improvement w hen compared to that of the maximum cross correlation (MCC), which has been operationally used for cloud tracking, and to that of the method of Barnard and Thompson, which estimates displacements based on combi ning point correspondences with relaxation labeling.