Motion estimation with detection of occlusion areas

A method for estimation of motion vector fields that includes detection of motion discontinuities and occlusion areas is presented. A cost function is formulated that incorporates coupled Markov random field models for the motion vector field, the motion discontinuity field and the occlusion field. The cost function is minimized by a deterministic iterative process. Application of this method to an image sequence with occlusion is described. Results of motion-compensated temporal interpolation for 4:1 field subsampling are presented and compared to motion estimation without occlusion processing.<
>