Researches On Rate Control And Affine Prediction For HEVC

Along with the fast development of the video capture and display equipments, the video coding technology, which belongs to the middle stage of the video transmission, is under more and more pressure. High Efficiency Video Coding (HEVC), the newest international video coding standard, can achieve 50% bitrate savings compared with the previous video coding standard H.264/AVC under the same subjective quality. Facing the increasing challenges, although the current standard can reduce the pressure to some extents, it is still far from enough. How to make full use of the bandwidth under HEVC and further improve the coding efficiency based on HEVC are the key challenges.On the one hand, the optimal bit allocation and accurate rate control are the key technologies to make full use of the bandwidth. Most of the traditional researches per-form bit allocation and bitrate control based on the quantization parameter. However, since the header information increases dramatically compared with the previous video coding standards, the Q-domain based rate control algorithm is unsuitable for HEVC. The newly emerged λ-domain rate control algorithm can control the bitrate accurate-ly for HEVC. However, there is no optimal bit allocation algorithm in correspondence to the λ-domain rate control algorithm. Besides, whether the λ-domain rate control algorithm is suitable for HEVC scalable extension needs careful study.On the other hand, only the translational motion model is employed in HEVC. Therefore, the coding tools in HEVC are still unable to characterize the complex mo-tions efficiently. Since the complex motions always take the majority of the bits, how to introduce the high order motion models into HEVC to better characterize the com-plex motions becomes a key problem. Most of the traditional researches on high order motion models are unable to get a balance between the coding efficiency and encod-ing/decoding complexity. Consequently the high order motion models have not been integrated into the video coding standards yet.This dissertation tries to improve the RD performance from the following two as-pects. First, this dissertation tries to achieve the optimal bit allocation and accurate rate control to achieve more effective utilization of the bandwidth. Second, this dissertation introduces high order motion models into the video coding standard to better character-ize the complex motion for better coding efficiency.The main innovations and contributions of this dissertation are listed as follows.(1) This dissertation proposes an optimal λ-domain bit allocation scheme for HEVC. The scheme mainly has three contributions. First, this dissertation proposes a complete λ-domain R-D analysis framework. Except for the R-λmodel to represent the relation- ship between R and A, this dissertation also proposes a novel D-λ model to represent the relationship between D and A. Second, this dissertation proposes a content-related optimal picture level bit allocation algorithm based on the λ-domain R - D analysis framework and fundamental R - D theory. Third, this dissertation proposes a content-related optimal basic unit level bit allocation algorithm based on the A-domain R-D analysis framework and fundamental R- D theory. The experimental results demon-strate that the proposed scheme can achieve better coding efficiency compared with the traditional methods. The proposed scheme has already been accepted by the interna-tional standardization group JCT-VC and integrated into the HEVC reference software.(2) This dissertation proposes an optimal λ-domain rate control algorithm for HEVC scalable extension. The scheme mainly has three contributions. First, this dissertation proposes an optimal initial target bits and initial encoding parameters determination al-gorithm to achieve better R - D performance. Second, this dissertation proposes an optimal bit allocation algorithm for the inter pictures for spatial and quality scalability taking the intra and inter layer dependency into consideration. Third, this dissertation proposes an adaptive model updating algorithm to estimate the R - λ model parameters accurately for better bitrate accuracy. The experimental results demonstrate that the rate control algorithm for HEVC scalable extension can improve the R-D performance obviously under quite high bitrate accuracy compared with the traditional methods.(3) This dissertation proposes a low complexity affine motion compensation frame-work to better characterize the complex motions. First, this dissertation proposes a four parameter affine motion compensation framework supporting multiple reference frames. Second, this dissertation proposes two methods to determine the affine motion: advanced affine motion vector prediction (AAMVP) and affine model merge. Especial-ly, combined with the AAMVP, this dissertation also proposes a gradient-descent based fast affine motion estimation algorithm. Third, this dissertation proposes two coding tools to reduce the complexity of affine motion compensation. One is the one-step sub-pixel interpolation filter to reduce the interpolation times. The other is an interpolation precision based adaptive block size motion compensation to reduce the interpolation complexity. The experimental results demonstrate that the proposed algorithm can im-prove the coding efficiency for the videos with obvious rotation and zooming. The proposed algorithm has already been integrated into the JEM reference software.