【作者】 李榕；

【导师】 余胜生；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2010， 博士

【摘要】 网络技术与多媒体技术的发展,促进了通信技术向综合化、数字化、智能化、个性化的发展。3G技术的迅速普及,在无线信道网络平台上传输语音、数据、图像,成为了新型通信业务的发展动力。由于无线信道相对于有线信道,呈现出完全不同的网络特性；移动终端相对于固定终端,也对应用提出了不一样的要求,因此,研究无线信道环境中的网络视频传输关键技术,具有重要的理论和实践意义。本文分析了无线网络传输环境,指出低码率视频压缩技术、码率控制技术、差错控制技术及节能技术是解决无线传输环境高带宽波动、高误码率和移动终端运算能力弱、待机时间短的关键技术。针对经典的码率控制由于需要计算方差而带来的计算复杂度较高的缺点,提出了一种基于优化比特分配的低复杂度码率控制算法。首先根据图像的复杂度分配帧层的目标比特数；然后在宏块层利用复杂度和运动信息计算编码的权重,确定最优编码器。实验结果表明,该算法能在不降低视频传输质量的情况下,将码率更稳定地控制在目标码率附近。在分析了主流的低码率视频编码技术分类的基础上,选择了近年来兴起的“视觉感兴趣编码算法(ROI算法)”作为研究对象,指出视频分割算法是影响ROI算法效率和质量的关键技术。经典的视频分割算法需要做大量的预处理工作,以提高分割的准确率,运算量大。本文在基于支持向量机分类的基础上,通过合理选取初始类中心,在图像分类的过程中动态地调整分类数量,提出了一种基于运动矢量聚类分析的视频分割算法,对静止背景下的移动目标进行分割,作为感兴趣区域。该算法利用了视频压缩算法的中间处理结果,预处理工作少,算法并行度高,算法复杂度低,易于硬件实现。在此基础上,实现了对运动区域的优先编码,采用基于SPIHT算法的精细编码；对背景区域或视觉不敏感区域的H.264编码。实践证明,该方法提高编码效率和传输效率,符合人眼视觉效果。针对经典视频分割算法对复杂纹理图像分割效果不好,效率低下的问题,使用了离散幅度信号变换理论,建立以信号量化精度作为尺度的离散幅度多分辨率分析为基础的视频分割算法。实验结果表明,该算法能很好对卫星图像等复杂纹理图像进行分割,同时适合于硬件实现高速实时并行处理。最后,针对差错控制算法进行了讨论,提出了一种解码端差错检测及差错隐藏的算法。比较空间相邻宏块、时间相邻宏块,利用大块视觉效应原理,可以检测出错误宏块,并采用空域隐藏、时域隐藏的方法进行错误隐藏。实验表明,该算法能很好的检测出错误宏块,并进行错误隐藏。在各个算法理论及仿真实验的基础上,进一步讨论了硬件实现、算法优化的方法,利用DSP、FPGA和GPU的硬件特性,提高运算速度,降低功耗。更多还原

【Abstract】 The development of network technology and multimedia technology, promote the integration, digitization, intelligentize, individuation of communication technologies. With the rapid popularization of 3G technology, transmission of audio,data and image through wireless network platform are becoming the power of new communication service. Because the network characteristics of wireless channel are very different from wired channel, and mobile terminal differs much from fixed terminal, the research of key network video transmission technology is very important both in theory and in practice.Through analysis of wireless networks transmission environment, low-bit rate video coding technology, rate controlling technology, error controlling technology and energy saving technology are supposed to be the key technology to solve bandwidth fluctuation of wireless transmission environment, high bit-error ratio, low computing power of mobile terminal, low stand-by time.Against the high complexity of classical ratio controlling, a low complexity ratio controlling algorithm based on bit allocation optimization is suggested. First allocate target bit number of frame layer according to the complexity of image, then select the optimal coder according to the complexity and the. Experiment results proved that this arithmetic can made the code rate stable without reduction in video transmission quality.Based on the analysis of low ratio video coding technology, ROI arithmetic is chosen as a research object, image segmentation arithmetic is proposed to be the key technology of ROI arithmetic. In classical image segmentation arithmetic, large amount of preprocessing work are needed to promote the accuracy of segment,which means high operation quantity. A new image segmentation arithmetic based on motion vector cluster analysis are proposed to extract the mobile target as a interested area in a stationary background. This arithmetic applies the intermediate result of video coding arithmetic, need few preprocessing, can be run parallel, and is easy to implement. Furthermore, coding priority for motional area, fine coding based on SPIHT arithmetic, H.264 coding for background and vision insensitive area are realized. It is proved practically that this method can raise the coding and transmission efficiency, improve vision effect.With concerning to the poor effect and low efficiency of segmenting complex texture image under the classical video segmentation algorithm, we adopted the theory of discrete amplitude signal transformation, and established the video segmentation algorithm that based on the analysis of discrete amplitude multi-resolution and the precision of the signal amplitude. The result of the experiment results showed that the algorithm would segment the satellite images and other complex texture images well, and was suitable for achieving high-speed real-time processing hardware. Theory of discrete amplitude signal transformation are employed to establish.Finally, we restudied the error control algorithm, and provided the decoder error detection and error concealment algorithm. It introduced the large visual effect theory, and by Comparing of space adjacent to the macro block, the time the adjacent macro blocks, it could detect the error macro blocks, and use airspace and time-domain hidden method to hide the error. Experiments showed that the algorithm could hide the error and detect error macro block well.Based on various algorithms and simulated experiments, we made further study on hardware accomplishment and algorithm optimization. We employed the hardware function of DSP, FPGA and GPU to improve operation speed, and cut power consumption.更多还原