| 摘要 |
A system and method for estimating motion vectors between frames of a video sequence which operates in the DCT domain with improved efficiency and reduced computational requirements. The motion estimation system operates to encode a target block using pointers or motion vectors to a previously encoded block, referred to as the reference block or search block. The system first partitions the target frame into a plurality of target blocks, and DCT transforms the target blocks in the target frame. The motion estimation system then selects a candidate block from the search frame and DCT transforms the selected candidate block. The motion estimation system uses a novel method for selecting candidate blocks which allows re-use of at least a portion of the transformed values of a prior selected candidate block. After the candidate block is selected, the motion estimation system determines one or more "relevant" target blocks in the target frame which have a corresponding search window which includes the selected candidate block. The system then computes a distance between the transformed values of the selected candidate block and the transformed values of each of the determined target blocks. If a better metric is found for a respective target block, the system stores the location of the selected candidate block for the respective target block. The above steps are performed for each of the possible candidate blocks in the search frame. This method thus produces or determines "best fit" candidate blocks for each of the target blocks. The system then computes motion vectors for each of the target blocks in the target frame based on these "best fit" blocks. <IMAGE> |
