Abstract
The H.265 (high efficiency video coding, HEVC) standard was finalized in April 2013, currently being as the newest video coding standard of the ITU-T Video Coding Experts Group and the ISO/IEC Moving Picture Expert Group. When HEVC has been emerged, several rate control algorithms are being developed for practical applications, especially in real-time streaming services. There are some of the existing rate control algorithms based on the R-Q model, which is considered having a solid relationship between streaming bitrate R and playback quality Q. Recently, a promising R-λ model, which having a more robust correspondence between R and Langrance multiplier λ, was proposed and adopted by HEVC reference software HM10.0. However, this general R-λ model does not perform well in the sense of frame rate prediction especially for low bitrate sequences. That is because the content (image frame) variation does not being included in the model function. In this study, an enhanced formula is proposed to remedy those ill-conditioned situations. By observation of the number of non-merge PU in all of coded frames, there has a great connection on the real bitrate cost noticeably. Another way of exploring the content variation is to uncover the ROIs (typically, people or faces) with saliency map. In our experience, the plot of non-merged PUs is quite similar to that of ROIs. In this thesis, an enhanced R-λ rate control modelling using non-merge information is proposed. Also, an alternative bit allocation coding tree unit (CTU) level is developed for enhancing the quality of ROIs. The experimental results show that the proposed method can reduce the bit-rate estimation error and improve the video quality in video conference test cases. Index Terms－HEVC, R-λ model, rate control, ROI, saliency map, merge mode