Lossy adaptive coding of motion vectors with global camera motion compensation

We propose a motion estimation and coding scheme that combines global camera motion estimation with a scene adaptive coder based on a quadtree structure with entropy-constrained vector quantization (ECVQ). Global camera motion parameters are first estimated before the local motion estimation. A quadtree structure is then constructed that finds the best block size to perform adaptive local motion estimation. Motion vectors of each block size are coded using an entropy-constrained vector quantizer specific for that block size. Our simulation results have shown that this adaptive coding algorithm has superior rate-distortion performance over the traditional fixed-block-size full-search method and the global camera motion compensation gives additional performance improvement when there is camera zoom and rotation.