Novel approach limiting non-square prediction unit (PU) estimation in HEVC

The new video coding standard (HEVC) allows non-square partition block prediction in inter frames, similar to H264/AVC, at every depth level. The current approach in HM carries out ME for all possible PU sizes including both square and non-square blocks. This achieves the highest coding efficiency with great demand for computational complexity. Though non-square partitions play a significant role in compression of given video content, estimating non-square partitions for each CU is not feasible on real-time encoding system. In this paper, we propose a novel approach to skip the non-square partition estimation/prediction for one or more CU-depths. In our approach, first we carry out depth decision only based on square blocks RD cost. Then, we determine CU depths at which non-square partitions can be estimated and/or avoided for other depths. Our Experimental results demonstrate that the proposed method can significantly reduce computational complexity while maintaining almost the same video coding efficiency as the original HEVC encoder.