Data mining from large multimedia contents and its applications

点击翻译按钮获取中文摘要

We proposed graph mining techniques for extracting all correlations of different kinds of attributes (e.g., keywords, objects in images, and graph structures) from large amounts of contents including multimedia. We formally defined a tree pattern, which can represent characteristics such as the usage of keywords (e.g., words, objects in image, and the structural relations among nodes in which characteristic words appear) in multimedia contents. Based on computational machine learning, we proposed efficient pattern matching algorithms and efficient graph mining algorithms for tree patterns and graph patterns. Moreover, in order to construct a theoretical base of user-oriented information retrieval system for multimedia contents, we proposed efficient methods of managing all extracted tree patterns.