【作者】 罗智勇；

【导师】 张会汀；

【作者基本信息】 暨南大学 ， 通信与信息系统， 2005， 硕士

【摘要】 作为多媒体通信技术的一项综合应用,视频会议系统正随网络环境的不断完善而飞速发展,并且在将来会拥有更广阔的市场需求。视频编码技术是对视频会议系统的整体效果产生关键影响的核心技术,所以本文针对视频会议应用,以先进的H.264标准为基础,以标准参考代码JM(Joint Model)为蓝本,研究并以软件实现满足实时要求的视频编码器。 本文首先对H.264视频编码标准中运算量最高,对编码效率影响最明显的运动估计技术进行全面的研究。通过分析标准推荐算法,比较大量经典的运动估计快速算法的性能和融合相关的理论假设,改进标准推荐算法,大幅提升运动估计算法的运算耗时。同时利用视频对象分割技术,粗略地检测视频会议视频内容中静止的背景部分,进一步减省编码器中运动估计模块的运算量。 在运动估计模块大幅改进的基础上,利用代码分析工具Intel VTune Analyzer对参考代码JM中各关键功能模块的运算耗时进行定量分析,选取其中运算耗时较高的SAD运算模块,1/4像素插值模块和变换编码模块作为效能优化对象。利用Pentium 4 CPU支持的SSE2指令集的并行运算能力,对上述三个功能模块的算法实现进行适当修改和代码重写,实现效能优化,使编码器整体的编码速度较好地满足实时要求。 最后,本文针对一般视频编码器产品对不同系统资源情况和不同帧速要求,采用相同编码参数设置的缺点,在学习标准中的率失真优化机制和速率控制技术基础上,研究关键的编码参数与运算复杂度间准确的比例关系和编码参数间的相关性特点,提出视频编码器运算复杂度自适应调节算法并给出完整的实现。保证在理想的视频质量下实现编码参数的自适应调节,满足视频会议系统预设帧速的要求,保证视频流畅性。 通过各改进部分和总体的测试数据的实证分析,证明本文研究并实现的视频编码器能很好的满足视频会议应用的实时要求,在带宽要求和编码速度相当的情况下,视频质量水平比通用的MPEG-4编码器(使用针对实时应用的Simple Profile设置)有明显的提升,同时具有较准确的运算复杂度自适应调节功能,对不断发展的视频会议应用有重要的实用价值。更多还原

【Abstract】 Video conference system, a synthetic application of multimedia communication technology, is developing dramatically and will also have wider market demand in the near future. As video coding is the core and critical technology that affects the whole performance of video conference system, researching and software implementation of the real-time video encoder based on the H. 264 standard and reference JM (Joint Model) source code will be provided in this thesis.The Motion Estimation (ME) technology with the highest computing complexity and the most obvious impact on the encoding efficiency is researched firstly. By analyzing the recommended ME algorithm of the H. 264 standard, comparing the performances of a large number of typically fast ME algorithms and merging the relevant theoretical assumptions, the executive speed of the recommended ME algorithm of the standard is greatly improved. More over, the static background of the video contents can be roughly separated from the foreground by adopting the Video Objects Segmentation technology, which also furthers the reduction of the running time of the ME module.Besides of the considerable improvement of ME module, three more time-consuming modules such as, the SAD calculation module, the transform coding module and the 1/4 pixel interpolation module, are selected based on the quantitative analyze of the statistic data from Intel VTune Analyzer and optimized by using the SSE2 instructions to rewrite the source code of them. The post-optimized coding speed of the encoder can basically satisfy the real-time requirement of video conference applications.At last, an adaptive algorithm for computing complexity adjustment is proposed to overcome the deficiency, using the same setting of coding parameters in different frame rate requirements and in different system, in most encoder products. Based on the study of the Rate-Distortion Optimization (RDO) and Rate Control (RC) technology of the H. 264 standard, researching the precise proportional relations between the important coding parameters and the computing complexity and the characteristics of relativity among the coding parameters, the full solution of the adaptive algorithm is designed and realized, which can adaptively adjust the coding parameters to guarantee the frame rate requirement set by the video conference system without visible quality deterioration.The video encoder researched and implemented in this thesis is proved to be efficient and effective by testing data in different stages in the developing process. It can well satisfy the real-time requirement of video conference systems with drastic improvement of video quality comparing to the MPEG-4 encoder (using Simple Profile settings) under the same bandwidth environment. It is also equipped with the adaptive adjustment function of computing complexity and has great practical value for the developing video conference applications.更多还原