An innovative fully-distributed automatic object classification algorithm with a new content-based video indexing research platform

The area of Content-Based Video Indexing and Retrieval (CBVIR) is an exciting and fast-moving field of research. It incorporates a wide variety of disciplines, from image processing and acoustics, over database management and user interface design, to networking and distributed computing. When the Advanced Telecommunications Engineering Laboratory (TEL) decided to expand its Foreign Media Project to incorporate a CBVIR system the focus was soon on how to accommodate all the required processing steps in a fully automated system without impeding the real-time multimedia streams or sacrificing accuracy. My solution addresses these needs by combining algorithmic optimizations with strategies for enabling these algorithms to utilize the vast computing resources provided through distributed processing. This dissertation was aimed at developing a unique research platform, titled Retina, which enables the rapid prototyping of new distributed CBVIR system algorithms, particularly for image recognition and object classification. I also present an innovative fully distributed object classification algorithm that was developed in conjunction with this research platform, called DOC-TM. Additionally, I have analyzed and compared the performance gains that each of the applied techniques provides and have presented recommendations for future improvements. My work shows that the new algorithm achieves performance improvements of several orders of magnitude over other approaches and provides the capabilities required to be included in TEL’s real-time CBVIR system. Furthermore this dissertation provides a primer on important issues related to the algorithm specifically and the field of computer learning and image processing in general, in the hope of providing others with a fast and easy start into this exciting but complex research area