An adaptive scheme for estimating motion

In this paper, we propose a new scheme for estimating the motion adaptively at each pixel. We assume non-stationary statistics for noise and motion, and propose a blind approach based on using the generalized cross validation (GCV) technique. We thus determine a different level of regularization at each pixel based on the local statistics using GCV. The problems addressed include the preservation of discontinuities, model/data errors, and confidence measures. Experimental results demonstrate the high accuracy of the proposed technique applied to both synthetic and real data.