Coding for Multimedia Sources and Sensor Data: New Theory, Algorithms, and Applications

摘要源编码在许多现代数字系统中起着至关重要的作用。事实上，作为一个研究领域，源编码比以往任何时候都更加重要和丰富，随着数字世界中多媒体和传感器数据的日益普及，产生了大量新的机遇和挑战。本研究在这一领域中追求一套特别有趣、新颖和及时的方向，旨在影响实际系统的设计。该研究有三个互补的方面：理论、算法和应用。在理论方面，重点是分布式源编码的问题。人们对这类问题的兴趣日益浓厚，例如，对网络中资源高效量化和压缩技术的需求。该调查追求分布式源编码理论的一个实质性但很大程度上未被探索的方面：如何在这样的问题中利用失真侧信息-即关于要编码的信号组件的相对重要性的信息。在算法领域，重点是涡轮量化-低复杂度迭代源编码技术及其实现。在过去的十年中，在开发用于纠错码的低复杂度迭代信道解码算法方面取得了巨大的成功。通过对偶原理获得的相应源代码具有类似地改变源编码实践的潜力。最后，研究了重要的新兴决策系统应用，其中源编码和失真准则发挥了不可或缺的作用，但以非传统的方式。所追求的具体示例包括用于媒体数据库的散列系统和用于验证多媒体内容的系统。

ABSTRACTSource coding plays a critical role in many modern digital systems.Indeed, as a research area, source coding is more important andfertile than ever, with a wealth of new opportunities and challengesarising from the increasing pervasiveness of multimedia and sensordata in a digital world. This research pursues a set of particularlyinteresting, novel, and timely directions in this realm, and aims toimpact the design of practical systems. There are three complementaryaspects to the research: theory, algorithms, and applications.In the theory area, the emphasis is on problems of distributed sourcecoding. There has been strong growth in interest in such problems,motivated by, for example, the need for resource efficientquantization and compression techniques in networks. Theinvestigation pursues a substantial but largely unexplored aspect ofdistributed source coding theory: how distortion side information--- i.e., information about the relative importance of the signalcomponents to be coded --- can be exploited in such problems.In the algorithms area, the focusses on turbo-quantization ---low-complexity iterative source encoding techniques and theirimplementation. Over the past decade, there has been a tremendoussuccess in developing low-complexity iterative channel decodingalgorithms for use with error-correction codes. The correspondingsource codes obtained via duality principles have the potential tosimilarly transform the practice of source coding.Finally, the research investigates important emergingdecision-system applications in which source coding anddistortion criteria play an integral role, but in nontraditional ways.Specific examples being pursued include hashing systems formultimedia databases and systems for authenticating multimediacontent.