Method and apparatus for estimating motion vector fields by rejecting local outliers

Apparatus for generating sample values representing a motion vector field which describes motion of individual image components of a current image frame and corresponding image components of at least one other image frame in a sequence of image frames, the apparatus comprising:

• memory means for holding data values representing respective image components of the current image frame and the one other image frame;
• initialization means, coupled to the memory means, for generating a plurality of motion vectors, each of said plurality of motion vectors corresponding to an estimate of motion between one of the image components in the current image frame and a corresponding image component in the one other image frame;
• means for generating an energy function value for a target motion vector of the plurality of motion vectors which said target motion vector corresponds to a current image component in the current image frame, said energy function value being a function of the target motion vector and of neighboring motion vectors which said neighboring motion vectors are ones of the plurality of motion vectors which correspond to ones of the image components that surround the current image component in the current image frame;
• means for selecting ones of said neighboring motion vectors of the target motion vector for further processing, each one of the neighboring motion vectors is selected if the one neighboring motion vector differs in value from the target motion vector by less than a predetermined threshold;
• and
• means for minimizing the energy function value of each motion vector of the plurality of motion vectors as a function of the selected motion vectors to produce a corresponding optimized motion vector value, wherein the optimized motion vectors corresponding to the plurality of motion vectors define the motion vector field.