Study On Fast Intra Prediction Algorithm In HEVC

As high definition and ultra high definition video appeared, the requirements ofpeople on the visual and all kinds of application for video are higher and higher. TheH.264/AVC video coding standard which was finalized in2003has not been able tomeet the growing demand people. In April2010, the JCT-VC (Joint CollaborativeTeam on Video Coding) put forward a new generation of Video Coding standardHEVC on the first meeting of JCT-VT. Meanwhile they formulated the correspondingtest model (Test Model under Consideration, TMuC). Through several years ofcontinuous efforts, HEVC was finalized in2013. HEVC coding standard focuses on HDand high-definition video image. It can be better able to adapt to a variety of differentnetwork environment, and it has the ability to support a multi-core parallel coding anddecoding.HEVC video coding standard introduces a lot of advanced coding technology inorder to improve the coding efficiency. But the introduction of advanced technology isat the expense of the high algorithm complexity, especially in the intra prediction.Therefore, we need to reduce the algorithm complexity to improve the speed of codingand meet the requirement of practical storage and transport.In this paper, by analyzing the HEVC quadtree coding unit splitting mode, wepropose a fast intra prediction algorithm of HEVC based on rate-distortion cost to theproblem of high computational complexity of intra prediction algorithm in HEVC. Andwe optimize this fast algorithm by introducing the depth relevance of the adjacent CU(Coding Unit). The join of these fast algorithms can simplify intra prediction algorithmcomplexity, shorten the encoding time and improve the coding efficiency of HEVC.In this paper, the specific research content and innovation work are as follows:(1) Make a deeply study on the intra coding techniques of HEVC standard,including the principle of intra prediction algorithm, the rate distortion optimizationtechnique, the intra prediction mode decision, algorithm complexity and some currentfast intra prediction algorithms. Through a large number simulation experiments, wesum up the relationship between the video image characteristics and the final selectionof coding unit type. (2) A fast intra prediction algorithm of HEVC based on rate-distortion cost isproposed. At first, the change law of rate distortion costs corresponding to thesedifferent prediction units and these rate-distortion cost thresholds of these differentprediction units are made through experimental analysis. When coding the currentcoding unit, we skip some coding unit sizes that do not fit this region after contrastbetween their rate-distortion costs and the threshold. Then the algorithm is integrated toHEVC by C++language. We use14general test sequences and two encodingconfigurations including LC (Low Complexity) and HE (High Efficiency). Finally, theexperimental results are analyzed from two aspects of the efficiency and quality of thealgorithm. The experimental results show that the average encoding time of theproposed fast algorithm decreases by23.55%and17.96%respectively in LowComplexity and High Efficiency test conditions. However the average coding bit rateincreases only by about0.32%and0.10%and the average brightness peaksignal-to-noise ratio reduces0.025dB and0.013dB respectively. In addition, it canprove the validity of the proposed fast algorithm that the proposed fast algorithm almosthas the same RD curves and reconstructed images comparing with HEVC standard.(3) We optimize this fast intra prediction algorithm based on rate-distortion cost byintroducing the depth relevance of adjacent CUs. When predicting the current CU, itdetermines whether to skip the prediction modes search of current CU using the depthbetween current CU and these encoded adjacent CUs and the corresponding CU inprevious frame. Then it continues coding using the fast intra prediction algorithm basedon rate-distortion. Through the analysis of simulation results, it shows that in LowComplexity and High Efficiency test conditions, the optimized fast algorithm makes theaverage encoding time decreased by27.04%and21.72%respectively.And the averagecoding bit rate only increases by about0.33%and0.16%respectively, the averagebrightness peak signal-to-noise ratio reduces0.028dB and0.021dB, respectively. Bycomparison of RD curves, we find that the coding performance of the optimized fastalgorithm has remained largely unchanged. Thus it can prove the effectiveness of thefast intra prediction algorithm based on the adjacent CU and rate-distortion cost.