Adaptive Fast PU Partition Mode Decision Algorithm For HEVC

HEVC(High Efficiency Video Coding) is a new generation video coding standard after H.264/AVC, compared to H.264/AVC standard, the video compression rate doubled. In order to achieve better coding performance, it employs a flexible block partition structure, including coding unit(CU), prediction unit(PU) and transform unit(TU). HEVC adopts eight kinds of inter prediction modes. HEVC uses the search selection method based on rate distortion optimization to find the minimum rate distortion cost model in all CU levels and all prediction modes, this will bring about higher computational complexity. A way to reduce the computational complexity and improve the coding speed in the premise of ensuring encoding quality is proposed in this thesis.There are three existing fast partition mode selection algorithm—CFM, ESD and the algorithm of Jong-Hyeok Lee. Remarkable success was achieved in any of the three algorithms, while the video encoding time and coding complexity remains to be decreased.The main technologies in HEVC are introduced firstly in this thesis. Then inter PU partition mode selection is deeply studied. After that, a fast PU partition mode adaptive decision algorithm based on spatial correlation is proposed. Because of the encoding order and the distance between the encoding picture and its reference pictures are different, the algorithm is divided into two parts— the low-delay configuration and the random-access configuration. The main idea of the algorithm is that, in the premise of the temporal homogeneity, a probability table is constructed between the partition mode of the current CU and its spatial reference CUs. The adaptive system could choose the partition mode candidates according to the partition mode of its best spatial neighbor CU in the probability table. In low-delay configuration, the experimental results show that, compared to the CFM algorithm, the encoding complexity can be reduced by up to 15.31% on average with only 0.0186 d B BD-PSNR decrement and compared to the ESD algorithm, the encoding complexity can be reduced by up to 22.72% on average with only 0.04793 d B BD-PSNR decrement. While in random-access configuration, compared to the CFM algorithm, the encoding complexity can be reduced by up to 16.01% on average with only 0.00928 d B BD-PSNR decrement and compared to the ESD algorithm, the encoding complexity can be reduced by up to 14.83% on average with only 0.00699 d B BD-PSNR decrement, and the encoding complexity can be reduced by up to 9.04% on average with 0.0256 d B BD-PSNR increment compared to the algorithm of Jong-Hyeok Lee. In conclusion, the algorithm reduce the encoding time in the premise of ensuring encoding quality.