| Video compression is the precondition of realizing various video services. Gener-ally, the raw video data is converted through certain encoders to data with other formats for decreasing the cost of store and transmission. During the video processing, the data after compression is converted back to the raw video with the corresponding decoders. Currently the latest video coding standard is High Efficiency Video Coding-HEVC (also known as H.265), which is developed in2010by Joint Collaborative Team on Video Coding (JCT-VC) established by the International Telecommunication Union Telecom-munication Standardization Sector (ITU-T), the International Standardization Organi-zation and International Electrotechnical Commission (ISO/IEC). HEVC is finalized in January2013. In spite of the similar block-based hybrid coded video framework as H.264/AVC(Advanced Video Coding), H.265/HEVC introduced many new coding tools and designs with improving both the visual quality and compression ratio. Ac-cording to experiments, by using H.265/HEVC, the bitrate can be reduced by an average50%compared to H.264/AVC with the same subjective image quality. In the trend of digital videos towards high definition, high frame rate and high compression ratio, ex-cellent coding performance will enable the latest H.265/HEVC to become the preferred video standard in the near future.H.265/HEVC is developed on the basis of H.264/AVC, and shows better perfor-mance in compression ratio, parallel processing capability and network adaptability. However, the improvement of compression ratio is achieved at the cost of high increase of computation complexity. Experiment tests show that H.265/HEVC doubles or even quadruples complexity compared with H.264/AVC. High computational complexity is a limiting factor in the spread of the new coding standard in the areas of monitoring systems, mobile devices and so on. Therefore, how to reduce the complexity and im-prove the encoding speed of H.265/HEVC has become a hotspot research in many en-terprises and research institutions at present. In this paper, we improved the encoding of H.265/HEVC mainly from intra prediction, inter prediction and parallel encoding to solve the problem. Concretely speaking:1. Improvement of intra prediction coding:As one key technology, intra coding is indispensable in H.265/HEVC. With using the information of neighboring blocks in the current frame, intra prediction removes the spatial redundancy effectively. However, complex Coding Unit(CU) structure and decision of intra prediction mode also bring high computation complexity. To reduce the computation com-plexity, we design and implement three fast CU size decision schemes.1) Based on the analysis of spatial correlation for image, the first method classifies the CUs into three groups. And the CU size is determined in early with referencing the neighboring CU sizes and combining the comparison of luma component of pix-els.2) The second scheme improves the intra coding speed by combining the distribution of luma component and local Rate Distortion (RD) costs. First, a narrow coding size range is achieved by analyzing the luma distribution through histogram. Then, during the intra coding of the CU in the narrow range, a thresh-old used to judge the splitting is calculated based on the neighboring RD costs. By comparing the RD cost of current CU with the threshold, the splitting can be terminated in early.3) Combining feature points and local RD costs, the third scheme determines the coding size range through Support Vector Machine (S VM) algorithm, and then terminates the splitting of CU in early based the RD costs as the second scheme.2. Improvement of inter prediction coding:the inter prediction occupies the whole encoding time by60%-70%, which is the key to improve the encoding speed and the focus of the current study. In this paper, we proposed two improved inter cod-ing algorithms. The first to speed the encoding is based on RD cost by selecting the proper thresholds for CUs in different depths and avoiding some unnecessary CU partition. The division of the Largest Coding Unit(LCU) with depth being0is determined by comparing the RD costs with the RD costs of CUs encoded with64x64size in its referencing picture. For the CUs in other depths, with us-ing the RD costs of encoded CUs in the same depth and the same Coding Tree Unit(CTU) as reference, the division can be terminated. This method can reduce the computation by terminating the CU partition in early. The second mothed is based on prediction residuals. Before inter perdition, we first compute one seg-menting threshold adaptively based on Lagrange multiplier for different frames, and then detect the edge of the prediction residuals using Sobel operator for root node in the coding tree structure of H.265/HEVC. The splitting of LCU is deter-mined according to the statistic information of edge. For the nodes in other depth, we classify the CUs into three categories and select appropriate partition by com-paring the dispersion of prediction residuals and one threshold which is acquired based on the statistics of different videos and Lagrange multipliers. Therefore, the improved inter coding algorithm is realized by analyzing the prediction residuals and selecting corresponding criteria for the CUs in different depths.3. The improvement of encoding for surveillance videos. Video surveillance is one of the most widely used video applications. This paper takes advantage of the high similarity in the consecutive frames of surveillance videos and proposed a new coding scheme for surveillance videos using inter-frame difference to en-code different image areas with different encoder options. The scheme is imple-mented through the proposed fast CU size decision algorithm. With using the luma component of difference image, the proposed algorithm can segment out moving objects from background, and then select proper CU size for different areas. Experimental results show that the encoding complexity can be reduced by an average of45%with small increment in bitrate and negligible loss in the visual quality of videos. Furthermore, the proposed scheme is not only applied to surveillance videos recorded by static cameras, but also applied to regular videos with excellent coding performance.4. Parallel optimization of encoding. Parallel coding is an effective way to im-prove the speed of encoding. With analyzing the Slice-level parallel encoding of H.265/HEVC, we found that the unbalanced load limits the speed of encod-ing. Therefore, this paper provided two algorithms to realize the improvement of the original Slice-level parallelization. Considering the high similarity of consec-utive images in one video, this paper proposed one algorithm called Reference Fore-frame Prediction (RFP) to balance the CTUs in each slice of current frame based on statistics of encoding time of all CTUs in the encoded frame. Further-more, with analysis on the structures of Group Of Pictures(GOP) and the encoding order, encoding modes and reference pictures of GOP under different configura-tions, the images were categorized into three groups. For each group, we utilized the proposed Weighted Composite Prediction(WCP) algorithm to optimize the CTU allocation of each slice, overcome the unbalanced allocation of CTUs for slices and improve the parallel efficiency of Slice-level encoding.In summary, the proposed algorithms developed and utilized the temporal corre-lation, spatial correlation and statistical correlation in the videos more fully than the HEVC test model. The improvement of intra coding, inter coding and parallel coding is obvious under the guarantee of visual quality. With neglectable loss in compression ra-tio, the saved encoding time is significant. In other words, the computation complexity of H.265/HEVC is reduced effectively. |
PDF Full Text Request