【作者】 霍炎；

【导师】 冯玉珉；

【作者基本信息】 北京交通大学 ， 通信与信息系统， 2009， 博士

【摘要】 近年来,伴随着我国通信网络基础设施的快速建设和经济的飞速发展,人们已不再满足电话、电视、传真和电子邮件等单一媒体提供的传统语音和文字通信,而是需要数据、图形、图像、音频和视频等多种媒体信息以超越时空限制的方式作为一个整体呈现,其中视频以其丰富的信息量成为多媒体通讯领域中最重要的部分,同时视频压缩编码与传输一直以来也是这一领域的热点。然而如何在满足人类视觉特性的情况下对视频源信号进行高效的压缩,适合无线信道恶劣环境的传输,并且为多媒体网络的设计提供准确反映视频业务流量特性的预测模型,是直接影响到用户对视频业务QoS的满意程度,也是多媒体通信中需要重点研究的部分。论文工作得到了国家自然科学基金课题“无线多媒体网络中基于新业务流模型和模糊进化神经网络/蚂蚁算法的资源管理技术研究(60472034)”与“无线宽带网络联合信道率失真模型相关技术研究(60772042)”,以及航天科工集团的“传感器图像遥测装置”等项目的支持,从多媒体视频信源编码的角度出发,研究视频信源的平滑运动估计、基于人类视觉特性的自适应量化以及无线网络下联合信源信道编码等问题,并针对视频编码业务流量特性进行分析,以得到相应的流量模型,为视频业务的接入控制、带宽分配等提供了模拟流量数据。论文主要创新点如下:在分析了现有运动矢量搜索模型的基础上,给出了一种新型的混合模式搜索模板,并且根据Kalman滤波方法、率失真最优的概念以及通过混合模式搜索模板得到的参考运动矢量,提出了一种平滑的率失真最优运动估计算法。该算法在平滑滤波和代价函数的共同作用下能够给出较为真实的运动矢量场,并从算法计算效率、准确度以及运动矢量分布上对该算法进行仿真分析,表明它能够有效得到目标物体的真实运动。为了使量化参数能够在编码信息量与编码失真之间进行适度平衡,针对目标比特速率与量化参数之间的关系进行分析,给出了比特速率差△R与量化参数调整因子△Q之间的近似二次关系,并由△R～△Q关系根据目标比特速率提出了量化参数调整因子△Q的迭代确定过程,为本文提出的基于人类视觉系统的自适应量化策略中,根据图像误差分布调整量化参数提供了必要的调整因子。针对视频图像归一化活动性值非均匀的特点,分析了归一化函数因子与视频图像的特征之间的关系,提出了一种活动性归一化的改进方法。同时在阐述了人类视觉系统的主要特性以及这些特性对视频编码中量化精度影响的基础上,根据第二章得到的真实运动矢量场,分别分析了视频图像亮度特性与空间活动性的关系,图像运动目标与时间活动性的关系,视频图像帧间相关性以及相应的误差分布,提出了一种基于人类视觉系统特性的误差分布反馈量化策略。实践证明,基于人类视觉特性的自适应量化在牺牲较少复杂度的前提下,在视频图像主客观质量评价上有一定的改善。在视频图像传输的系统模型中,针对BSC信道分别分析了编码端信源失真与解码端残留失真特性,提出了一种基于视频图像DCT变换结果之间的差值之和的信源失真度估计方法,并且分别对Intra和Inter帧的信道失真模型进行描述,同时根据信源编码统计分析描述了现有的信源码率模型从而提出了一种基于DCT特征的码率模型。最后依据上述提出的失真模型和速率模型给出了一种信源信道联合编码策略。通过对解码端重建视频质量的分析,表明联合编码策略较传统独立编码方法在PSNR性能上有一定的改善。分析了MPEG视频流量的长时相关性、自相似性Hurst参数等性能,并以小波模型为基础,针对MPEG流量,提出了一种流量预测模型。相比传统的流量模型来说,基于小波的流量预测模型利用了小波多分辨率分析方法:针对尺度空间中MPEG流量的尺度系数的自相关仍然具有一定的长相关性给出了自适应误差反馈线性预测的方法;同时由于小波变换的“白化”作用使得小波系数只具有短相关性,并给出了线性相关结构进行逼近。最终将尺度系数和小波系数通过小波反演变换得到MPEG流量预测结果。由于分别针对尺度空间和小波空间各自的特点进行不同的预测建模,使得它既能拟合视频流量的概率分布特性、又满足视频特有的相关结构,因此该模型能够比较精确的描述复杂的MPEG流量。更多还原

【Abstract】 Recently,with the rapid development of economy and construction of communication network infrastructure,people are not satisfied with the traditional sounds and words transmission based on the single media such as telephone,TV,fax and email.Data,graphics,images,audio and video information are needed to be showed as a whole,in which video with abundant information takes an important part in the field of multimedia communications.However,the factors directly influence QoS of video service and are important parts in multimedia communication,which are how to implement efficient compression of video source in the condition of meeting HVS, transmit video signal in the wireless channel environment,and provide prediction model based on video traffic properties for design of multimedia network.The work of this dissertation is supported by the National Natural Science Foundation of China "Study of New Traffic Stream Models and Fuzzy Evolutionary Neural Network/Ant Algorithm Based Resource Management Technologies in Wireless Multimedia Networks(No.60472034)","Study of Joint Channel Rate-Distortion Model in Wireless Broadband Network(No.60772042)",and the project of China Aerospace Science & Industry Corp.which is "Sensor Images Telemetry Device".The dissertation studies on the problems of video coding,which include smoothing motion estimation, adaptive quantization based on human visual system,and joint source channel coding in wireless networks.And then in order to implement admission control and optimal bandwidth allocation of video traffic,the prediction model and traffic properties are analyzed.The main innovations in the thesis are outlined as following:On the Basis of the existing search methods of motion vector,the novel hybrid pattern search(HPS) method is proposed.And then according to the Kalman filter, Rate-Distortion theory and the candidate motion vectors which are obtained by the HPS method,a smooth rate-distortion optimal motion estimation algorithm is presented.The result of true motion field can be obtained by the smoothing filtering and cost function.Simulations on computational efficiency,accuracy and motion distribution show that this algorithm can capture motion of target effectively.In order to balance the coding information and distortion by adjusting quantization parameter,the relationship between target bit rate and quantization parameter is analyzed.We have studied the approximate quadratic relationship between bit rate differenceΔR and adjusting factors of quantization parameterΔQ,and proposed the iterative process for calculating quantization parameter adjusting factor which is the necessary factor of implementing adaptive quantization based on HVS.With respect to non-uniformity characteristic of the results of video image activity normalization,analyzing the relationship of normalized factors and characteristics of video image,an improved activity normalization method is proposed.Meanwhile, by discussing performances of HVS and according to the true motion field got in chapterⅡ,the following are studied respectively:the relationship of luminance and spatial activity,the relationship of motions and temporal activity,the correlation among the interframes and the corresponding error distribution.A novel method of error distribution feedback quantization scheme based on HVS is proposed. Simulation results show that the objective and subjective quality has been improved at the expense of less complexity by adopting adaptive quantization based on HVS.We have studied the structure of video transmission system,and presented a new method of source distortion estimation based on SDDCT and model of channel distortion for Intra-/Inter-frame based on performances analysis of source distortion and residual distortion in BSC channel.Meanwhile,the source rate model is proposed according to the statistical analysis of source coding.Finally,based on the rate model and distortion model,we have presented a scheme of joint source channel coding.Compared with independent coding,the results of PSNR show that quality of reconstructing video by joint coding is better than by non-joint coding about 2dB.We have analyzed the properties of video traffic and presented a wavelet on-line prediction model based on linear correlation structure of scaling space and wavelet space in MPEG trace.The proposed algorithm of error feedback adaptive linear prediction is given by LRD of scale coefficients of traffic;meanwhile the linear correlation structure is used by SRD of wavelet coefficients of traffic.Finally,the prediction traffic is obtained by inversion wavelet.It is verified by the simulations that this prediction model can capture the probability distribution and LRD of the traffic very well because of modeling scale space and wavelet space respectively.更多还原