抄録
The block matching technique, which is the most widely utilized technique for motion vectors (MVs), often estimates noisy MVs in matching blocks (MBs), which have less variation in brightness or contain a single edge. To improve accuracy and coherence of MVs, this paper proposes a new estimation technique. This technique comprises MB classification based on a reliability metric for MVs, MB integration, and recalculation of MVs for integrated MBs. Several examples have shown that it can accurately estimate MVs even for flat or edge MBs due to the integration. It was applied to an MPEG encoder and a retrieval system for motion pictures. An evaluation of the coding and retrieval efficiencies demonstrated the validity of the proposed approach.
