Three stage hierarchal motion vector determination

Claim 1

In an encoder for coding a sequence of interlaced video frames to generate a compressed bit stream, a method for determining a motion vector for a macroblock in one of said frames, said method comprising the steps of (1) in a first stage of a search hierarchy, decimating said macroblock and a search area in a reference frame by a first decimation factor horizontally and a first decimation factor vertically to form a first decimated macroblock and a first decimated search area, (2) utilizing said first decimated macroblock and first decimated search area, generating in one or more motion estimation units in said encoder, a frame-based motion vector and a plurality of field-based motion vectors, (3) retaining said frame-based motion vector and a selected one of said field-based motion vectors which results in the best prediction for said macroblock, (4) in a second stage of said search hierarchy, decimating said macroblock and said search area by a second decimation factor horizontally and a second decimation factor vertically to form a second decimated macroblock and a second decimated search area, said second decimation factor horizontally and said second decimation factor vertically being smaller than said first decimation factor horizontally and said first decimation factor vertically, respectively, (5) using said second decimated macroblock and said second decimated search area, generating in said one or more motion estimation units, a refined frame-based motion vector and a refined field-based motion vector corresponding to the selected field-based motion vector, (6) determining a mode used to obtain a final motion vector for said macroblock based on whether said refined frame-based motion vector or said refined field-based motion vector results in a better prediction for said macroblock, (7) in a third stage of said search hierarchy, using said determined mode and said macroblock at full resolution and said search area at full resolution, generating in said one or more motion estimation units, a final motion vector for said macroblock, and (8) coding said macroblock in said encoder using said final motion vector to generate a portion of said compressed bitstream.