Method and apparatus for motion field estimation, segmentation and coding

Claim 1

An improved method of motion field estimation for use in motion compensated frame-to-frame prediction coding comprising the steps of: dividing a frame having a plurality of pixels into a plurality of smaller regions to form a first segmentation level; assigning to each of said plurality of smaller regions an initial motion vector; and performing for each of said plurality of smaller regions a motion vector updating routine which updates the motion vector of a smaller region by assigning to it a best motion vector selected from among the initial motion vector assigned to the smaller region, a matched motion vector obtained by performing a block matching technique for the smaller region, and motion vectors of the smaller region's neighboring regions, wherein the best motion vector is selected according to a priority scheme and a predetermined threshold value, and wherein the threshold value reflects the amount of improvement, obtained by assigning a particular motion vector as the best motion vector, relative to a smallest matching error.

Claim 12

An apparatus for use in improved motion field estimation of a video image, said apparatus comprising: a previous decoded frame memory unit for storing a monochrome intensity for each pixel in a preceding frame; a current frame memory unit for storing a monochrome intensity for each pixel in a current frame; a motion field memory unit for storing a motion vector for each pixel in the current frame; a motion refinement unit for performing a block matching technique for a block of pixels in the current frame and for providing a matched motion vector; a candidate motion vector memory unit for storing motion vectors which are candidates for updating the motion vector of said block of pixels, wherein the candidate motion vector memory unit is connected to said motion field memory unit for receiving and storing an initial motion vector assigned to said block of pixels and motion vectors of regions neighboring said block of pixels, and wherein the candidate motion vector memory unit is further connected to said motion refinement unit for receiving and storing the matched motion vector; a matching error computing unit for computing matching errors obtained by assigning to said block of pixels the motion vectors stored in the candidate motion vector memory unit; a matching error memory unit for storing the matching errors computed by said matching error computing unit; a minimum detector unit for determining the value of the smallest matching error among the matching errors stored in the matching error memory unit; a best motion vector selection unit, connected to the motion field memory unit, for determining, according to a priority scheme and a predetermined threshold value, a best motion vector for refining the motion vector of the block of pixels, wherein the best motion vector is selected from among the motion vectors stored in the candidate motion vector memory unit; and a control unit for controlling the other units and their interaction, wherein the control unit is connected to each of the other units in the apparatus.

Claim 14

A system for performing motion field estimation and coding of a video image, said system comprising: a motion estimation unit for refining a motion vector assigned to a block of pixels in a current frame by assigning to the block of pixels a best motion vector selected from among a plurality of candidate motion vectors, wherein said best motion vector is selected according to a priority scheme and a predetermined threshold value that reflects the amount of improvement, obtained by assigning a particular motion vector as the best motion vector, relative to a smallest matching error, and wherein said plurality of candidate motion vectors includes an initial motion vector assigned to said block of pixels, a matched motion vector obtained by performing a block matching technique for said block of pixels, and motion vectors of regions neighboring said block of pixels; a merge and label unit for merging adjacent regions of the current frame that have similar motion vectors to form merged regions and for assigning to each pixel in the current frame a region label; a first region label memory unit for storing a motion vector associated with each merged region; a second region label memory unit for storing the region label assigned to each pixel in the current frame; a region shape representation unit for forming a set of contour pixels that defines the merged regions; a contour coding unit for encoding the set of contour pixels; and a control unit for controlling the other units and their interaction, wherein the control unit is connected to each of the other units in the system.