An efficient all-zero-block detection method for general orthogonal transformations

摘要

The latest video coding standards achieve high coding gains but result in high complexity. To reduce the computation involved and speed up the encoder, various All- Zero Block (AZB) early detection algorithms are designed to skip transform coding and quantization. In this paper, the orthogonal transformations are analyzed using the Euclid Space theory first, and then the sufficient and necessary conditions for AZB detection and non-AZB detection are derived. Based on sufficient conditions, new features for the refined classification and the corresponding thresholds are found. Finally, a general AZB detection method with a constrained false detection rate is proposed. The proposed detection scheme is applied to the 4x4 and 8x8 transforms in H.264/AVC. The experiment results show that up to 80% of the computations for transform and quantization are saved as compared to those for the encoder without AZB detection1.