基于HEVC標準的轉碼技術研究
發(fā)布時間：2018-08-28 17:41
【摘要】：新一代視頻編碼標準HEVC作為H.264/AVC的繼承者,在視頻壓縮效率方面取得了巨大的提升,相比于H.264/AVC,HEVC在相似的視頻感知質量下比特率減少了大約50%,因此正在逐漸成為業(yè)界視頻壓縮的主流標準。同時隨著移動互聯(lián)網(wǎng)革命的爆發(fā),人們越來越頻繁地使用移動端設備來觀看視頻,而在移動網(wǎng)絡中視頻的瀏覽受到網(wǎng)絡阻塞的影響十分明顯,暫時的網(wǎng)絡阻塞會大大降低用戶的觀看體驗。因此在視頻服務器端,往往會保存視頻的高比特率版本,并根據(jù)當前網(wǎng)絡情況實時轉碼為不同碼率的視頻流提供給用戶。這種情況就對高清HEVC視頻的轉碼速度提出了新的要求。視頻轉碼實際上是一個先解碼再編碼的過程,其中編碼部分耗時占比達到90%。而HEVC標準因為其特殊性,編碼時需要確定最優(yōu)CU劃分模式,這個過程需要遍歷每一層CU劃分,同時還需要進行復雜的率失真優(yōu)化(RDO)計算,因此這是一個十分耗時的過程。針對這個問題,本文提出了兩種快速確定編碼單元CU劃分模式的算法來降低HEVC標準視頻的轉碼計算復雜度,從而在基本不影響視頻質量的前提下,大大縮短視頻轉碼時間。第一種方法利用了輸入的高比特率視頻流與輸出的低比特率視頻流在深度值上的關聯(lián)性,簡單快速地確定出深度值范圍,從而減少CU劃分模式的遍歷范圍。通過實驗證明,相比于傳統(tǒng)的全解全編轉碼模式,此方法僅僅增加了 0.84%的比特率,而轉碼時間縮短了 54%。第二種方法結合機器學習理論提出了在線訓練在線分類的轉碼框架,利用原始碼流CU劃分信息以及時域前一幀的CU劃分信息,通過樸素貝葉斯分類器預測出編碼端的CU劃分標志,從而確定了 CU劃分模式。實驗表明,通過這種方法視頻實驗幀僅僅增加了 2.74%的比特率,而轉碼時間縮短了 72%左右。
[Abstract]:As the successor of H.264/AVC, HEVC, a new video coding standard, has made great progress in video compression efficiency. Compared with H.264 / AVC HEVC, the bit rate is reduced by about 50% under the similar video perception quality, so it is becoming the mainstream standard of video compression in the industry. At the same time, with the outbreak of the revolution of the mobile Internet, people use mobile devices more and more frequently to watch video. However, in the mobile network, the browsing of video is obviously affected by the blocking of the network. Temporary network congestion can greatly reduce the user's viewing experience. Therefore, in the video server, the high bit-rate version of the video is often saved, and real-time transcoding is provided to the user for different bit-rate video streams according to the current network conditions. This situation puts forward new requirements for high-definition HEVC video transcoding speed. Video transcoding is actually a process of decoding and coding, in which the coding part takes up 90% of the time. Because of its particularity, HEVC standard needs to determine the optimal CU partitioning mode, which needs to traverse every layer of CU partition and perform complex rate-distortion optimization (RDO) computation, so it is a very time-consuming process. To solve this problem, this paper proposes two fast algorithms to determine the CU partition mode of the coding unit to reduce the computational complexity of the transcoding of the HEVC standard video, thus greatly reducing the transcoding time without affecting the video quality. The first method makes use of the correlation between the input high bit rate video stream and the output low bit rate video stream on the depth value, and determines the range of the depth value simply and quickly, thus reducing the traversal range of the CU partition mode. It is proved by experiments that compared with the traditional fully decomposed full-coding mode, the proposed method can only increase the bit rate by 0.84% and shorten the transcoding time by 54%. The second method combines machine learning theory and puts forward the transcoding framework of online training online classification, which uses the CU partition information of the original code stream and the CU partition information of the previous frame in the time domain. By using naive Bayesian classifier to predict the CU partition flag in the coding end, the CU partition mode is determined. The experimental results show that the proposed method can only increase the bit rate by 2.74% and shorten the transcoding time by about 72%.
【學位授予單位】：南京理工大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TN919.81
本文編號：2210102
[Abstract]:As the successor of H.264/AVC, HEVC, a new video coding standard, has made great progress in video compression efficiency. Compared with H.264 / AVC HEVC, the bit rate is reduced by about 50% under the similar video perception quality, so it is becoming the mainstream standard of video compression in the industry. At the same time, with the outbreak of the revolution of the mobile Internet, people use mobile devices more and more frequently to watch video. However, in the mobile network, the browsing of video is obviously affected by the blocking of the network. Temporary network congestion can greatly reduce the user's viewing experience. Therefore, in the video server, the high bit-rate version of the video is often saved, and real-time transcoding is provided to the user for different bit-rate video streams according to the current network conditions. This situation puts forward new requirements for high-definition HEVC video transcoding speed. Video transcoding is actually a process of decoding and coding, in which the coding part takes up 90% of the time. Because of its particularity, HEVC standard needs to determine the optimal CU partitioning mode, which needs to traverse every layer of CU partition and perform complex rate-distortion optimization (RDO) computation, so it is a very time-consuming process. To solve this problem, this paper proposes two fast algorithms to determine the CU partition mode of the coding unit to reduce the computational complexity of the transcoding of the HEVC standard video, thus greatly reducing the transcoding time without affecting the video quality. The first method makes use of the correlation between the input high bit rate video stream and the output low bit rate video stream on the depth value, and determines the range of the depth value simply and quickly, thus reducing the traversal range of the CU partition mode. It is proved by experiments that compared with the traditional fully decomposed full-coding mode, the proposed method can only increase the bit rate by 0.84% and shorten the transcoding time by 54%. The second method combines machine learning theory and puts forward the transcoding framework of online training online classification, which uses the CU partition information of the original code stream and the CU partition information of the previous frame in the time domain. By using naive Bayesian classifier to predict the CU partition flag in the coding end, the CU partition mode is determined. The experimental results show that the proposed method can only increase the bit rate by 2.74% and shorten the transcoding time by about 72%.
【學位授予單位】：南京理工大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TN919.81
【參考文獻】
相關期刊論文 前8條
1 李雪晴;何小海;吳小強;林宏偉;;結合模式對應與機器學習的HEVC降分辨率轉碼[J];電視技術;2016年01期
2 程悅;賈克斌;呂卓逸;;一種基于快速模式選擇的H.264跳幀轉碼算法的研究[J];計算機應用與軟件;2015年03期
3 胡乾乾;沈禮權;趙振軍;;基于場景切換的HEVC碼率控制算法[J];電視技術;2015年05期
4 劉昱;胡曉爽;段繼忠;;新一代視頻編碼技術HEVC算法分析及比較[J];電視技術;2012年20期
5 楊高波;夏中超;張兆揚;王會千;;一種基于線性回歸模型的運動矢量重估算法[J];湖南大學學報(自然科學版);2010年07期
6 郭亞寧;馮莎莎;;機器學習理論研究[J];中國科技信息;2010年14期
7 張亞娜;張琦;;碼率控制中的率失真優(yōu)化[J];中國傳媒大學學報(自然科學版);2006年01期
8 程愷英,王宏遠,樊淳標;數(shù)字視頻轉碼技術綜述[J];電視技術;2005年04期
，本文編號：2210102
本文鏈接：http://sikaile.net/kejilunwen/xinxigongchenglunwen/2210102.html
