Efficient motion estimation using multiple log searching and adaptive search windows

Most current standards for video compression employ block matching motion compensation to reduce temporal redundancy, although they do not specify how the block matching is to be implemented. Exhaustive search gives the optimal compressed video signal quality for block matching but at the expense of high computational complexity. This computation can be decreased by well-known reduced search techniques, such as log search, but this can be at the expense of reduced signal to noise ratio. This paper presents two techniques which maintain signal quality close to that of full search but with significant savings in computation. The first technique employs multiple searches focused on the predominant small motion vectors combined with pixel decimation to reduce comparisons. The second dynamically adjusts the search window size to avoid wasting search effort on areas where it is considered unlikely that the optimal motion vector will be found.