| 摘要 |
In a digital video motion estimation subsystem (1), pixel block-matching is preceded by an analysis of block detail (40) to dynamically select variably-sized pixel blocks for matching. A metric for quantification of block detail is provided. Given an externally specified level of required detail (22), blocks are recursively subdivided as long as at least one subdivision child retains sufficient detail (60). The variably-sized blocks are classified into "sharp" (91) or high-detail blocks, and "flat" (92) of low-detail blocks. Flat blocks are prone to spurious matches, while sharp blocks are more likely to match unambiguously or to fail to match due to occlusion or changes of scene content. Block-matching (search) resources may then be concentrated on sharp blocks. |
