Intra prediction header bits estimation algorithm for RDO in H.265/HEVC

Abstract

The amount of video information to store and transmit is continuously increasing. The new video compression standard H.265/HEVC provides the ability to halve video bitrate at the cost of higher computational expenses. The most time-demanding part of a compression system is decision making, especially when based on rate-distortion optimization (RDO). There are a lot of papers that tend to decrease the complexity of decision making by either reducing the number of coding options to compare or by reducing the complexity of RDO estimation. The aim of the latter approach is to eliminate from rate estimation the binary arithmetic coder (BAC), as it is the most computationally expensive step. Most of works focus on proper estimation of the residual data, while proper estimation of prediction information is either neglected or is still performed with BAC.In our work we elaborate the proper rate estimation of the prediction header without the use of BAC. The suggested algorithm counts the symbols to be coded with BAC and estimates the bit size of the intra prediction header based on the entropy of the symbols, obtained in the statistical experiments. The proposed algorithm provides the average Maxim P. Sharabayko and Oleg G. Ponomarev bitrate overhead of 0.09% on JCT-VC test video sequences, and can be used as a part of decision making algorithm in H.265/HEVC video compression systems.Mathematics Subject Classification: 94A08

Modified fast search for intra prediction mode in H.265/HEVC standard

“…Those modes are a part of the most probable modes (MPM) list that additionally includes some modes that are considered the most probable for the current PU [10]. Coding of MPM takes less bits than coding of non MPM [11], which means that potentially MPM may have least RD cost . The results in Table 2 show that this approach provides 2.60% bitrate overhead on average and 5.98% at maximum.…”

Fast Rate Estimation for RDO Mode Decision in HEVC

“…While the total bit rate overhead is 1.54%, the prediction header estimation errors provide 0.09% overhead [22], while the rest is provided by the residuals estimation errors.…”