Study on Data Compression Coding in Multimedia Communication

In order to solve problems presented in the multimedia data coding in the Nation "9th Five Year’s "Project, researches on fractal image coding、fractal speech coding and video coding are made in this dissertation. There are seven chapters in this dissertation. The chapter 1 is the introduction of the development of image coding、speech coding and video coding. The main research work in this dissertation is also presented in this chapter.In chapter 2, we proposed a fractal image coding method based on genetic algorithms. The searching scheme problem of domain block in fractal image coding is well solved by applying a robust genetic algorithm with well-selected populations and revised genetic operators. Experimental result has show that it can increase fractal image coding speed by this method.In chapter 3, a mathematical model based on short distance self-similarity is suggested and it can be used to predict the searching space of domain block in fractal image coding. Experimental result indicates that it can decrease coding time without degrading image quality and compression rate by applying this model in fractal image coding. The relationship between the prediction model and its parameter is also studied in this chapter.In chapter 4, we proposed a speech coding method based on iterated function system. In order to cut down the computation complexity, a dynamic noise filter is used to erase the non-speech signal. According to the character of speed signal, we realized a two layers fractal speech codec with fractad interpolation method. This codec can work at 8kbits/s (with good speech quality. Since the compressed speech data can be divided into two layers (resample signal layer and error signal layer), therefore, it is suitable to be transmitted on computer networks.In chapter 5, we submit an image distortion criterion based on visual threshold and visual masking effect. In order to increase coding speed, this distortion criterion is applied into H.263 codec. Experimental result indicates that it can cut down H.263 coding time without decreasing compression rate and subjective dquality of reconstructed image by this way. In chapter 6, a H.263 algorithm based on face detection is proposed. H.263 is a video coding standard for low bitrate video conference. The searching range of the motion vector is one of the most important parameters that can influence the coding speed of H.263. considering the character of image sequence in the video conferencing, we design a method to quickly detect the displacement of the face in the sequence and adjust the searching range of the motion vector dynamically. The experiment results indicate that it can greatly increase the coding speed by applying this method in H.263. In chapter 7, the summary is made and the main results are given.

Alternate abstract:

本论文针对国家九五重点攻关项目--"计算机支持的协同工作系统"中的多媒体数据压缩问题，对分形图象编码、分形语音编码、活动图象编码等多媒体数据压缩编码的理论和方法开展了研究。全文共分为七章，第一章为结论，对图象编码、语音编码、视频编码等相关领域的发展概况和现状进行了综述，并于介绍了本文的主要内容。第二章在简介分形图象编码和遗传达室算法的基础上，提出了基于遗传算法的分形图象压缩编码方法；通过有目的选择初始群体、改造遗传算子等手段，较好地解决了分形图象编码中的定义域块搜索策略问题，大大提高了分形编码的速度。第三章在图象近距自相似性的基础上提出一个预测数学模型，对分形图象编码中定义域块的搜索空间作出了定量的描述。在此章中，我们还讨论了该数学模型和有关参数（如图象的自相关性、方差）的关系。实验表明将该数学模型应用到分形图象编码中，可以在不降低压缩比和图象质量的情况下，显著提高编码速度。第四章提出了基于迭代函数系统的分层语音编码方法，该方法首先通过动态地消除背景噪声减少了非读音数据段的计算量，然后根据语音信号和残差信号的特点，利用分形插值的方法实现了一个分层的分形语音编码器，该编码器可以在8K左右的码率下得到较好的读音质量，同时由于这种编码器的码流由亚抽样码流和残差信号码流构成，所以适合于网络上的分层传输。第五章提出了一个符合人眼视觉特性的失真判断准则，该准则能够很好地反映人眼的视觉阀值效应和视觉掩盖效应。将该准则应用到视频H.263编码中，可以降低H.263算法中的运动估值计算时间。实验表明该方法能够在保证图象的主观质量和不明显降低压缩比的情况下，提高H.263算法的速度。第六章提出了一个基于脸部跟踪的H.263算法。H.263是低比特流的桌面视频会议中首选的视频编码标准，而影响H.263算法速度的一个关键参数是运动估值中的矢量搜索范围。本章利用桌面视频会议中人物图象序列的独有特点，提出了一个在图象序列中快速检测面部的算法，并通过对序列中人脸的位移检测，动态地调节不同运动幅度图象的矢量搜索范围。实验结果表明，对于类似于Miss这样的头肩部图象序列，在不明显影响主、客观图象质量和压缩比的情况下，该方法能够有效提高H.263的编码速度。第七章对全文进行了总结，概括了研究成果。