Multi-feature based fast depth decision in HEVC inter prediction for VLSI implementation

Abstract

High efficiency video coding (HEVC) is the latest international video compression standard that achieves double compression efficiency than the previous standard H.264/AVC. To increase the compression accuracy, HEVC employs the coding unit (CU) ranging from 8 × 8 to 64 × 64. However, the encoding complexity of HEVC increase a lot since the manifold partition sizes. A lot of works are focused on reducing the complexity but didn't considered the feasibility of hardware implementation. This paper proposes a hardware friendly fast depth range definition algorithm based on multiple features. Block texture feature, quantization feature and block motion feature are utilized. Block texture feature is based on the texture similarity in consecutive frames. Quantization feature is based on the compression regularity of HEVC. Block motion feature is for compensate the difference caused by the moving object. Comparing with the original HEVC, the proposed method can saved about 33.72% of the processing time with 0.76% of BD-bitrate increase on average.