场景切换视频自适应帧间码率控制

针对视频帧中可能出现的大量场景切换,提出一种基于非连接点的场景切换检测算法,提高编码性能,该场景检测算法复杂度低,在运动估计的同时,完成视频场景切换检测。场景切换将导致GOP(group of pictures)长度的变化,并可能出现GOP长度太短的情况。提出改进的自适应GOP时域滤波技术,避免由于GOP太短引起的编码性能下降。针对视频场景切换检测分割出的不同长度的GOP,提出一种基于率失真模型的帧间码率控制算法,利用视频的失真与码率及视频帧复杂度的关系,对帧间码率分配进行优化,提高重构视频帧的总质量。实验结果表明,基于场景检测的自适应帧间码率控制算法能够获得较好的编码性能。     

一种H.265视频带宽实时预测模型

针对视频直播过程中带宽抖动导致视频帧丢失造成用户体验降低问题,提出了一种H.265视频带宽实时预测（VBRP）模型。模型基于马尔可夫链,专注于H.265编码的直播视频流,研究了在GOP（group of pictures）编码模式下B帧出现的统计特性,发现B帧大小对视频流传输速率的影响较大。应用此特性可预测出直播流中B帧的大小,并根据网络带宽选择性丢弃B帧,同时引入步长调整因子AF和误差阈值FT,均衡模型在预测B帧时的训练频率和预测帧数。最后基于模型实现VBRP预测算法并在直播系统中验证算法的有效性。     

基于错误传播模型的非均等视频流丢失保护

提出一种适用于丢包网络、面向图像组（GOP）层的非均等视频流丢失保护方案。利用GOP中不同帧之间的非均等显著性,将不同数量前向错误校正包分配到GOP层的不同帧中。采用帧间包交错机制将突发包丢失分散到不同帧上,提高处理突发包丢失时的鲁棒性。仿真结果表明,在不同信道丢失模式下,该方案能提高视频接收质量。     

监控视频聚簇模式挖掘及其应用

在监控视频中,新输入的视频帧与可更新背景的差异可以实时反映监控场景中运动目标的大小和多少等信息.本文以此计算帧运动量,提出一种监控视频场景聚簇模式挖掘方法.先依据帧运动量的变化分割监控视频流,然后计算视频段的平均运动量,并对其进行K均值聚类分组,最后利用获得的聚簇知识,可以对监控视频实现多尺度摘要和相似视频检索.     

一种MPEG-4编码的视频源业务建模

针对MPEG－4的低质量和高质量两种编码质量的视频源，给出了一种混合模型，此模型在多个时间尺度上反映帧大小的变化；场景的变化和同一场景内码流的波动，用几何分布来描述场景的变化，用AR（2）模型来反映同一场景内码流的波动，对I帧采用合成模型建模，对P，B帧采用几何分布建模，最后将此模型与不考虑场景变化的几种情况进行了比较，并且对它们的排队性能作了分析，证明了这种合成模型能够很好地反映实际视频流的特征。     

基于小波变换的MPEG视频流预测模型

提出了一种基于小波变换的MPEG视频流的预测模型,该模型对原始数据做Haar小波变换以后,充分考虑尺度系数和乘子的不同统计特性,对尺度系数和乘子采用不同的预测模型分别预测,最终通过多重分形小波模型建模过程合成MPEG视频流序列。利用仿真和数学分析的方法对预测模型在刻画MPEG视频流的主要统计特性方面进行性能分析,从误差、数据的概率分布特性以及自相似特性等方面,证实了该预测模型的优越性能。     

ATM网络中全动态视频源的二次建模

通过对ATM网络视频业务传输性能及其建模分析,针对高运动性（即全动态）视频源的建模,给出基于块的MPEG运动估计算法,提出一种非线性模型。基于自回归在GOP层对MPEG视频流进行二次建模。仿真表明,该文提出的NAR模型能准确预测具有适度运动性的实际MPEG流,对具有高运动性的流在不同的负荷和缓冲器的条件下提供可接受的预测。     

AVS实时编码的VBR码率控制

针对AVS实时编码应用,提出一种基于帧层的一次编码VBR码率控制算法。利用GOP层复杂度预测来动态分配码率,在GOP内的帧采用CBR策略,结合AVS编码RDO的特点,采用准确且计算简单的率失真模型计算量化参数。实验结果表明,与AVS现用的码率控制方法JVT-H017相比,该方法码率控制准确,算法复杂度低,平均亮度PSNR提高了约0-2dB,亮度PSNR样本方差约为JVT-H017方法的40％,表明视频质量的平缓性也得到较大改善。     

一种快速的压缩域视频流场景分段算法

提出了一种快速的压缩域视频流场景分段算法，对视频流进行两次分析：第一次分析（粗略分析）只分析P-帧中宏块统计信息，检测出可能存在的镜头边界；第二次分析（精确分析）再对粗略分析找出的边界邻近的B-帧和P-帧的宏块类型进行分析。从而对场景变换进行精确分析和定位，实验结果表明，粗略分析可以满足实时检测的速度要求，帧定位误差控制在10帧之内，精确分析可以进一步把帧定位误差控制在2帧之内。     

基于谱聚类的增强神经网络模型

卷积神经网络(CNN)因其容易产生过拟合而困扰着广大研究者。产生过拟合的重要原因在于模型往往局限于局部最优解。针对此问题,提出了基于谱聚类的增强神经网络模型。使用谱聚类算法对验证集进行聚类,以聚类结果为依据将单任务模型扩展为多任务学习模型。用公开的声音场景识别数据集进行实验,实验结果表明,基于谱聚类的增强神经网络模型可以有效降低过拟合风险,提升模型的测试精度。该方法适用于除声音场景识别(ASC)领域的其他领域。     

MPEG压缩域多尺度镜头切变检测算法

如何在压缩域进行镜头的切变检测一直是视频自动索引和检索中的难点。提出了一种MPEG压缩域多尺度镜头切变检测算法,在GOP、slot和B帧三个尺度上对MPEG视频流进行分析。通过对相邻I帧的检测,确定一个GOP中是否存在镜头切变;通过对slot的分析,确定镜头切变在GOP中所处的区域;通过对B帧的检测,确定镜头切变发生的确切位置。     

The periodic Markov modulated batch Bernoulli process and its application to MPEG video traffic

The Markov modulated batch Bernoulli process (MMBBP) is widely used for modeling a superposition of multiple voices and data at multiplexers in asynchronous transfer mode (ATM) networks, because the MMBBP has the time correlation property and its autocorrelation function is exponentially decaying. However, experimental studies on the MPEG (Motion Pictures Expert Group) video traffic show that its autocorrelation function is periodic as well as exponentially decaying. So, the MMBBP is not adequate to describe the MPEG video traffic.

In this paper, we propose a new stochastic process which we call the periodic Markov modulated batch Bernoulli process (P-MMBBP) to describe the MPEG video traffic adequately. We derive the autocorrelation function of the P-MMBBP and show that the autocorrelation functions of the P-MMBBP and the MPEG video traffic are of the same pattern. We also propose a simple way to match parameters of the P-MMBBP from real MPEG video traffic. We consider the P-MMBBP/D/1 queue and derive the queue length distribution analytically. From numerical results, we show that periodicity gives a significant effect on the tail behavior of the queue length distribution. We also compare tail probabilities obtained by using the P-MMBBP with those obtained by using the 2-state MMBBP. Our comparison shows that the use of the 2-state MMBBP underestimates tail probabilities.     

基于模糊控制的ATM网络VBR视频传输平滑策略

VBR视频传输的突发性是影响ATM网络服务质量的关键因素,文中通过模糊控制方法对视频传输流量的阈值进行监控,实现了接入的平滑并可动态调整传输速度。文中以传输流量及这发级作为评价指标进行分析,结果表明VBR视频传输自适应平滑策略的实效性。     

An error-resilient GOP structure for robust video transmission

Recent advances in technology have resulted in a significant growth in wireless communications and widespread access to information via the Internet, which have resulted in a strong demand for reliable transmission of video data. The challenge of robust video transmission is to protect the compressed data against hostile channel conditions while bringing little impact on bandwidth efficiency. In motion-compensated video-coding schemes, such as MPEG-1 or MPEG-2, an I frame normally is followed by several P frames and possibly B frames in a group-of-pictures (GOP). In error-prone environments, error happening in the previous frames in a GOP may propagate to all the following frames until the next I frame, which is the beginning of the next GOP. In this paper, we propose a novel GOP structure for robust transmission of MPEG video bitstream. By selecting the optimal position of the I frame in a GOP, robustness can be achieved without reducing any coding efficiency. Another advantage of the proposed GOP structure is also analyzed: compared with the conventional GOP structure, it provides reverse-play operation for MPEG video streaming with much less requirement on the network bandwidth. Experimental results demonstrate both the robustness of the proposed GOP structure and the efficient reverse-play functionality it leads to.     

一种新的基于运动矢量的MPEG VIDEO码流cut检出算法

介绍了一种对ＭＰＥＧ ＶＩＤＥＯ码流中场景切换点进行高速检出的新方法,在计算各帧相对于预测帧的类似度之后进行了类似度的归一化,降低了ｃｕｔ检出算法的复杂度,提高了检出的速度。     

An adaptive control for video transmission over bluetooth

This paper is concerned with transmission of a moving picture expert group (MPEG) video stream over a Bluetooth channel, using an adaptive neuro-fuzzy technique. MPEG variable bit rate (VBR) video sources over a network generally experience long delay and unacceptable data loss, due to high variations in bit rate. Furthermore, transmission rate could be unpredictable in a Bluetooth network due to interferences by other wireless devices or general Bluetooth channel noises. Subsequently, it is almost impossible to transmit VBR data sources over Bluetooth without excessive delay or data loss. In this work, an adaptive scheme is introduced so that the controller may adjust itself to the current state of the system under control. This paper utilizes a traffic-shaping buffer to prevent excessive back-to-back transmissions of MPEG VBR data sources. A novel adaptive neuro-fuzzy scheme regulates the output rate of the buffer to ensure that the video stream from the host conforms to the traffic conditions of the Bluetooth channel during the transmission period. The computer simulation results show that the use of the neuro-fuzzy controller reduces excessive delay and data loss at the host-controller-interface, as compared with a conventional VBR video transmission and a rule-based fuzzy controller (RBF1) in Bluetooth.     

Modeling video sources for real-time scheduling

What is the impact of the autocorrelation of variable-bit-rate (VBR) sources on real-time scheduling algorithms? Our results show that the impact of long term, or interframe, autocorrelation is negligible, while the impact of short term, or intraframe, autocorrelation can be significant. Such results are essentially independent of the video coding scheme employed. To derive these results, video sequences are modeled as a collection of stationary subsequences called scenes. Within a scene, a statistical model is derived for both the sequence of frames and of slices. The model captures the distribution and the autocorrelation function of real-time video data. In previous work, the pseudoperiodicity of the slice level auto-correlation function made it difficult to develop a simple yet accurate model. We present a generalization of previous methods that can easily capture this pseudoperiodicity and is suited for modeling a greater variety of autocorrelation functions. By simply tuning a few parameters, the model reproduces the statistic behavior of sources with different types and levels of correlation on both the frame and the slice level.