【摘要】：新一代視頻編碼標準HEVC(High Efficiency Video Coding)的編碼效率比H.264/MPEG-4AVC提高了一倍以上。但是對其編碼工具的靈活選擇使得HEVC編碼器復雜度急劇增加,這嚴重阻礙了HEVC的應用和發(fā)展。因此針對HEVC編碼快速算法的研究至關重要。 本文在介紹了視頻編碼技術的概況和視頻編碼標準的發(fā)展歷史之后,對HEVC的編碼框架進行了概括。然后針對HEVC編碼工具的特點,指出了HEVC編碼優(yōu)化的方向：幀內編碼單元快速選擇算法、幀間編碼單元快速選擇算法和運動估計快速算法。 針對HEVC幀內編碼單元劃分復雜度高的問題,提出了一種基于統(tǒng)計學習的幀內編碼單元快速選擇算法。將幀內編碼單元的劃分建模為k-means分類問題,通過分析四個子編碼單元覆蓋區(qū)域像素的方差組合而成的四維向量的特征,用簡單而有效的k-means分類方法進行編碼單元劃分的預測,從而避免了基于率失真優(yōu)化的全搜索算法,降低了編碼器的計算復雜度。 針對HEVC幀間編碼單元劃分復雜度高的問題,提出了基于深度時空相關性的幀間編碼單元快速選擇算法。在分析了時空相鄰編碼樹單元之間的相關性之后,依據相關性強弱選擇最佳相鄰編碼樹單元,并利用最佳相鄰編碼樹單元的深度,提前預判當前編碼樹單元的深度搜索范圍。同時根據前一幀中相鄰編碼單元的深度關系和當前幀中已編碼相鄰編碼單元的深度,預判當前編碼單元的深度搜索范圍,從而進一步提高了幀間編碼單元選擇的速度。 HEVC中的多參考幀技術以及靈活的數據劃分方式,大幅度增加了運動估計的復雜度。針對多參考幀選擇,提出了基于不同預測單元最佳參考幀相關性和層間編碼單元最佳參考幀相關性的多參考幀選擇算法。利用劃分為2N×2N的預測單元中各個參考幀的率失真代價,減少同一編碼單元中其它劃分模式的候選參考幀數目,加速參考幀選擇過程。同時當父編碼單元的模式為SKIP時,將當前編碼單元中所有劃分模式的參考幀限定為父編碼單元的最佳參考幀,從而進一步降低多參考幀選擇的復雜度。另一方面,搜索范圍在影響運動搜索復雜度的同時也影響數據搬運帶寬,通過分析不同分辨率視頻設置不同搜索范圍的編碼結果,為不同分辨率的視頻推薦不同的搜索范圍,能夠有效的降低數據搬運帶寬。最后總結了本論文的研究成果,并提出了該領域下一步研究的方向和任務。
[Abstract]:The coding efficiency of the new generation video coding standard HEVC (High Efficiency Video Coding) is more than double that of H.264/MPEG-4AVC. However, the flexible choice of encoding tools makes the complexity of HEVC encoder increase dramatically, which seriously hinders the application and development of HEVC. So it is very important to study the fast algorithm of HEVC coding. After introducing the general situation of video coding technology and the development history of video coding standards, this paper summarizes the coding framework of HEVC. Then, according to the characteristics of HEVC coding tools, the paper points out the direction of HEVC coding optimization: fast selection algorithm of intra coding unit, fast selection algorithm of interframe coding unit and fast algorithm of motion estimation. In order to solve the problem of high complexity of HEVC intra coding unit partitioning, a fast selection algorithm of intra coding unit based on statistical learning is proposed. The partition of intra coding units is modeled as k-means classification problem. By analyzing the characteristics of four dimensional vectors formed by the variance combination of four subcoding units covering the region pixels, a simple and effective k-means classification method is used to predict the division of coding units. Thus the full search algorithm based on rate-distortion optimization is avoided and the computational complexity of encoder is reduced. In order to solve the problem of high complexity of HEVC inter-frame coding unit partition, a fast selection algorithm of inter-frame coding unit based on depth space-time correlation is proposed. After analyzing the correlation between the adjacent coding tree units in time and space, the best adjacent coding tree units are selected according to the relative strength, and the depth of the best adjacent coding tree units is determined in advance to determine the depth search range of the current coding tree units. At the same time, according to the depth relation of adjacent coding unit in the previous frame and the depth of the adjacent coding unit in the current frame, the depth search range of the current coding unit is forecasted. The multi-reference frame technology in HEVC and the flexible data partition method greatly increase the complexity of motion estimation. For multi-reference frame selection, a multi-reference frame selection algorithm based on optimal reference frame correlation of different prediction units and optimal reference frame correlation of interlayer coding unit is proposed. Using the rate-distortion cost of each reference frame in the prediction unit divided into 2N 脳 2N, the number of candidate reference frames in other partition modes in the same coding unit is reduced, and the selection process of reference frames is accelerated. At the same time, when the mode of the parent coding unit is SKIP, the reference frames of all partition modes in the current coding unit are limited to the best reference frames of the parent coding unit, thus further reducing the complexity of multi-reference frame selection. On the other hand, the search range not only affects the complexity of motion search, but also affects the bandwidth of data transportation. By analyzing the coding results of different search range for different resolution video, it recommends different search range for different resolution video. Can effectively reduce the data handling bandwidth. Finally, the research results of this paper are summarized, and the next research directions and tasks in this field are proposed.
【學位授予單位】：浙江大學
【學位級別】：博士
【學位授予年份】：2014
【分類號】：TN919.81

【引證文獻】

相關碩士學位論文 前1條

1 黨允舒;基于能量集中的MVM超平面研究[D];吉林大學;2015年

，

本文編號：2176154