QQ客服 官方微博 反馈留言

题名	数字视频快速浏览、无损编码和物体分割算法的研究
作者	夏杰
学位类别	博士
答辩日期	2005
授予单位	中国科学院声学研究所
授予地点	中国科学院声学研究所
关键词	快速DCT部分解码 无损视频编码 误差建模 视频物体分割
其他题名	Research on Digital Video Fast Browsing, Lossless Compression, and Object Segmentation Algorithms
中文摘要	本文的研究工作主要集中在数字视频领域，主要满足低端用户（个人）和高端用户（内容提供商）对视频编码算法在低功耗快速解码、提高无损压缩性能、以及从视频中准确地提取视频对象等方面的需求，主要包括以下三个方面：约在考虑手机屏幕的小尺寸和手机芯片的低功耗因素的前提下，本文提出了一个基于部分IDCT解码的低代价快速浏览算法，仅仅使用3个DCT系数，同时使用整型乘法和移位运算模拟浮点运算进一步减少运算代价，从实验结果可以看出，该算法仅仅需要使用了2个整型乘法运算，8个整型加法运算和5个移位运算和3个内存单元来完成一个8*8大小块的DCT解码，同时在客观评价上和主观评价上，图像画面也基本与Partial3IDCT算法相一致。2）提出了基于有损／无损分级结构和基于混合DPCM／EM模型的无损视频压缩编码算法，并且着重提出了改进型的基于上下文树形结构的EM算法，以及与之相适应的帧间基于宏块的运动估计算法以进一步改进后者的压缩性能。从实验结果可以看出，基于有损／无损分级结构的无损视频压缩编码算法具有可以向下兼容有损视频压缩标准的优越性，从而具有了硬件设计上的简易性，而基于DPCM／BM模型的无损视频压缩编码算法在大部分视频中都胜过了Motion-JPEGLS和Motion-CALIC算法，同时融入了改进型的基于上下文树形结构的EM算法，以及与之相适应的帧间基于宏块的运动估计算法的改进型基于DPCM／EM模型的无损视频压缩编码算法，在压缩率上超过使用JPEG-LS压缩DPCM误差的参照算法最多为18．9％，超过使用CALIC压缩DPCM误差的参照算法最多为16.1％。3）提出了一种新式的基于AnchorNede定位与连接的前处理模式监督式视频物体分割算法，AnchorNode的准确定位是在基于空间域准则、前向时间域准则、反向时间域准则和有效准则的前提下，通过结合边界形状特征图、跟踪模式特征图、边界候选者特征图、空间域特征图和时间域特征图，使用EOSWBMA算法，GMF算法和BC算法来完成的。同时相比于传统的监督式视频物体分割算法的两阶段式框架（帧内分割阶段和帧间分割阶段），我们引入了一个由用户交互进行边界调整的阶段，这样通过局部修改在帧间分割阶段不满足要求的分割结果部分，可以提高算法的效率，减低用户的交互操作量。从实验结果看出，我们提出的一个新式监督式视频物体分割算法在大量视频序列上有不错的分割结果，同时在一些视频序列上也可以取得比其他监督式视频物体分割（例如VSnake算法）更好的分割效果。
英文摘要	The research topic of this thesis paper is mainly foucs on the digital video filed, which is towards the need of low-cost fast browing algoithm, improving lossless compression performance, and extract out the video object, from the low-end user(personal custom) and high-end user (content provding enterprise). The research is including three major points: 1) Considering the small size of LCD and lost-cost designation reqirement of mobile phone, we proposed a partial-IDCT based fast browing algorithm, which only used three DCT coefficients to decode a 8x8 block, and simulated the float multiplication by using integer multiplication and shift operation. From the experiments results, our proposed algorithm only needed 2 integer multiplication, 8 integer addition, and 5 shift operation totally to decode a 8x8 block, and also maintain the similar image quality compared to the Partial 3IDCT algorithm. 2) We also proposed a lossy/lossless scalable structure based lossless video compression algorithm and a hybrid DPCM/Entropy Coding model based lossless video compression algorithm. Also we specially proposed an improved context tree-based entropy coding algorithm and a new block-based motion estimation algorithm, which is used in the hybrid DPCM/Entropy Coding model based lossless video compression algorithm, to improve the coding efficency. From the experiment results, the lossy/lossless scalable structure based lossless video compression algorithm had the advantage of being compatible with lossy video compression system, and as a result, it can be easiler to design on the hardware. The hybrid DPCM/Entropy Coding model based lossless video compression algorithm can achieve better compression performance than Motion-JPEGLS and Motion-CALIC algorithm in most of test sequences, and the compression ratio of the improved hybrid DPCM/Entropy Coding model based lossless video compression algorithm, which incoporated the improved context tree-based entropy coding algorithm and a new block-based motion estimation algorithm, are higher than the results by the reference algorithms which used the JPEG-LS and CALIC to encode the DPCM error respectively, by 18.9% and 16.1% at most. 3) We also proposed a new Anchor Node localization and connection based pre-processing supervised video object segmentation algorithm. The Anchor Node localization is achieved by using EOSWBMA algorithm, GMF algorithm and BC algorithm, according to the Boundary Shape Map, Tracking Mode Map, Boudnary Candidate Map, Spatial Feature Map and Temporal Feature Map, under the Spatial Criteria, Forward Temporal Criteria, Backward Temporal Criteria, and Effectiveness Criteria. And compared to traditional supervised video object segmentation algorithm, we introduced a new local user refinement phase to adjust the unsatisfied segmentation results partially to reduce the user's labor work. From the experiemnt results, our proposed algorithm can achieve good segmentation results in many video sequences, and better segmentation results in some video sequences compared to other supervised video object segmentation algorithm like VSnake.
语种	中文
公开日期	2011-05-07
页码	138
内容类型	学位论文
源URL	[http://159.226.59.140/handle/311008/954]
专题	声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式 GB/T 7714	夏杰. 数字视频快速浏览、无损编码和物体分割算法的研究[D]. 中国科学院声学研究所. 中国科学院声学研究所. 2005.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。