Research On Rate-distortion Optimization Techniques For Video Coding

The rapid development of multimedia services and the increase of the users’requirement to video quality stimulate the constant improvement and update of thevideo coding technologies. Nowadays, the video coding standard of H.264/AVC hasbeen extensively utilized in various media products due to its high coding performance.Since it takes a better user experience, the ultra-HD video gradually comes into our eyes.But enormous video data make the storage and transmission more difficult. Therefore,the latest video coding standard of H.265/HEVC is formulated by joint collaborativeteam on video coding (JCTVC) for pursuing higher compression efficiency.Rate distortion optimization plays a critical role in video coding. It guides theoperations of mode decision and rate control for obtaining the target bit-rate, optimalcoding modes, and most appropriate quantization parameter for higher efficiency. In thisdissertation, the emphasis is put on the mode decision of H.264/AVC and H.265/HEVC,and the rate control of H.265/HEVC. Additionally, the no-reference quality estimationmodel is concerned. The major contributions are outlined as follows:1. A fast algorithm of bit-rate estimation is developed to reduce the complexity ofH.264/AVC encoder. The principle of context-adaptive binary arithmetic coding(CABAC) and the characteristic of the syntax elements related to residues are firstanalyzed. Then the efficient bit-rate estimation algorithm is gradually constructed forsubstituting the actual CABAC in mode decision. The proposed method, on the onehand, brings forward a highly accurate and efficient solution in bit-rate prediction. Onthe other hand, its design is well suited to parallel processing for further time-saving.Experimental results demonstrate the proposed method can reduce about57%of theencoding complexity in CABAC in H.264/AVC mode decision, with negligibledegradation of the video quality.2. A new fast mode decision algorithm is proposed for decreasing the codingcomplexity of H.265/HEVC encoder. Considering that the factors of frame texture andits related quantization step potentially affect the final CU mode, an initial CU depthprediction algorithm for large coding unit (LCU) is constructed to skip partial larger CUand an early termination algorithm of CU mode is established to avoid the unnecessarysmaller CU. The two techniques mentioned above are combined for fastCU modedecision, serving for considerable complexity reduction while well maintaining thevideo quality. Experimental results demonstrate that compared to the conventionalHEVC, the proposed method achieves an average bit-rate increment of0.69%, whereas the coding complexity has a significant reduction of about51%.3. A frame layer bit allocation algorithm considering the video contentcharacteristic is proposed for H.265/HEVC codec. Generally, the difference of videocontent characteristic would lead to the distinct of its coded bits. In order to guaranteethe consistency between the target bits and coded bits, the frame content complexityshould be considered in the frame layer bit allocation. In this paper, the relationship ofcontent characteristic and bits is first obtained by theoretical derivation. Then theparameter related to frame content complexity is extracted and employed forconstructing the efficient bit allocation algorithm. Experimental results show that thetarget bits and coded bits keep a better consistency, and the reconstruction video qualityis improved by0.128dB in average.4. An accurate rate model is proposed for H.265/HEVC inter-frame coding. Theestablishment of rate model is related to two factors: effective source description andexpected behavior of specific coding algorithms. Firstly a joint Laplacian distributionsource at the transform unit (TU) levels is constructed to describe the sourcecharacteristic according to the inter-frame residues distribution. Then the mutualinformation is predicted to measure the reduction of uncertain information in CABAC.Finally, the conditional entropy is calculated to estimate the output bit-rate ofinter-frame residues. Experimental results show that the proposed model achieves highaccuracy in bit-rate prediction. And the proposed approach can be extended to othervideo codecs using CABAC for the design of rate models.5. A no-reference quality estimation model is proposed based on analysis ofcontent feature of packets. Packet loss is a primary element to impair the quality ofnetwork speech quality, and the impairment extent is not only related to the number oflost packets, but also their content feature. This method first estimates the contentfeature of lost packets resorting to the voice active detection technology and the level ofnon-lost frame. Based on the content feature analysis, the proposed method calculatesthe speech packet loss rate and then predicts the quality of networked speech.Experimental results show that high prediction accuracy can be obtained by theproposed model. Compared with the speech quality evaluation model in G.1070, theaverage performance gain of the proposed method is8.4%in terms of relativity tosubjective results.