| 摘要 |
Using a multi-stage operation, a motion vector is detected between a base block in a present frame and the best matching position of an inspection block in a reference frame. In an initial stage (Fig. 9C), constant and transient components, such as a mean value (m) and a standard deviation ( sigma ), are extracted from the base block and compared with constant and transient components extracted from the inspection block as its position is varied through a predetermined search range. A coarse motion vector to a first best matching block in the predetermined search range is obtained. In a next stage (Fig. 9B), the base block is formed into small blocks, and the inspection blocks at positions in the neighborhood indicated by the coarse motion vector are formed into small blocks. Constant and transient components (e.g. m1, sigma 1) for each of the small blocks are obtained and compared. A next motion vector to a second best matching block in the neighborhood indicated by the coarse motion vector is obtained. In a final stage (Fig. 9A), image data for the base block and the inspection blocks at positions in the neighborhood indicated by the next motion vector are compared to obtain a final motion vector. <IMAGE> |
