A Survey: Fast Inter CU selection algorithms for HEVC
R. S. Hingane, Sanskar Agrawal · 2015
A hybrid quad tree structure in which current CU recursively divided into four equal sized CU's, is adopted by HEVC. At each decomposition level (depth), the RD cost for each of the following modes, SKIP mode, merge mode, inter 2N × 2N, inter 2N × N, inter N × 2N, inter 2N × nU, inter 2N × nD, inter nL × 2N, inter nR × 2N, inter N × N (only available for the smallest CU), intra 2N × 2N, and intra N × N (only available for the smallest CU) in inter-frame prediction is calculated. Among the all possible depth levels and prediction modes, the one which having least RD cost is selected. These inter CU selection process poses high computational complexity which is main obstacle in implementing HEVC for real time applications. Recently huge research has carried out on reducing computational complexity of HEVC encoders. This paper presents different methods for fast inter CU selection algorithms which reduces the encoding time of HEVC coding process. A survey of these various algorithms and an evaluation of their pros and cons may provide valuable leads for the improvement in HEVC inter-prediction. Keywords— H.264/AVC;H.265/HEVC;CU; inter prediction; RD cost. I. INTRODUCTION THE Joint Collaborative Team on Video Coding (JCT-VC) formed collaboratively by ISO-IEC/MPEG and ITU-T/VCEG, has developed new video coding standard, High Efficiency Video Coding (HEVC)/H.265. HEVC adopted a quad tree coding structure with Coding Unit (CU) as a basic coding element. Unlike macroblock (MB) in AVC/H.264 standard, CU size is flexible (from 64 X 64 to 8 X 8 luma samples). A CU is recursively divided into four equally sized CU's which enables content adaptive coding in HEVC. CU splitting continues until it could not be divided further into smaller CU's. The Prediction Unit (PU) carries the information related to prediction process. The size of PU is not restricted to be square to occupy boundaries of the objects in the pictures. Fig. 1 shows CU partitioning up to depth level 3. The dark black lines shows CU at level 0, blue lines split CU into depth level 1 and so on. CU is encoded by deciding its optimal prediction mode out of 10. At each depth level, RD cost for each of the prediction modes is calculated. The prediction mode having least RD cost value is selected. After deciding on RD costs for all CU's at depth level, say X, all four cost values are added and this addition is compared with RD cost of corresponding CU in previous depth level X-1. If addition value is less than cost value at depth X-1, then CU splitting is continued (up to level 3 ), else splitting is stopped at depth level X-1. For deciding on CU splitting up to particular depth level and PU partitioning mode, HEVC encoder needs to calculate large number of RD costs. The computing process of RD cost contains a series of operations as prediction, transform/ inverse transform, quantization/ dequantization, and entropy coding for each possible PU mode which makes major contribution in computational complexity. Recently heavy research has been carried on reducing the computational complexity of HEVC encoding process. This paper presents survey on different methods for inter CU selection algorithms. The paper proceeds as; Section II Overview of HEVC Encoding, Section III gives Literature Review of different methods for fast inter CU selection, Section IV gives summary of survey and conclusions are given in Section V.