An improved fast feature-based block motion estimation

A fast block matching algorithm in the feature domain was proposed by Fok and Au (1993) which achieves a computation reduction factor of N/2 with search block size of N/spl times/N. This paper presents an improved fast block matching algorithm in the integral projections feature domain. This algorithm is based on the fast block matching algorithm in the feature domain and a new motion field sub-sampling scheme. By utilizing the properties of the features and the motion field sub-sampling scheme, with a search block size of N/spl times/N, the improved algorithm can achieve a computation reduction factor of N while retaining close-to-optimal performance in the mean absolute difference (MAD) sense.<<ETX>>