| 摘要 |
<p>A method and device compute the inverse discrete cosine transform (IDCT) in a block based motion compensated digital video decoder, such that the computational complexity is dependent on syntax cues from an input block, and the resulting overall complexity is greatly reduced on average. The method and device provide a means for selecting a set of pruned IDCT algorithms (501) based on the statistics of generic coded video data. The classification of input IDCT blocks (507) uses the last non-zero coefficient obtained as a by-product of decompression and dequentization. The mapping (508) from the last non-zero coefficient to a corresponding set of separable one-dimensional IDCT algorithms is performed efficiently with the careful consideration of memory requirements, mapping complexity, and pruned IDCT-complexity tradeoffs.</p> |
