Mining multimedia data

Abstract

Data Mining is a young but flourishing field. Many algorithms and applications exist to mine different types of data and extract different types of knowledge. Mining multimedia data is, however, at an experimental stage.We have implemented a prototype for mining high-level multimedia information and knowledge from large multimedia databases. MultiMedia Miner has been designed based on our years of experience in the research and development of a relational data mining system, DBMiner, in the Intelligent Database Systems Research Laboratory, and a Content-Based Image Retrieval system from Digital Libraries, C-BIRD, in the Vision and Media Laboratory.MultiMediaMiner includes the construction of multimedia data cubes which facilitate multiple dimensional analysis of multimedia data, and the mining of multiple kinds of knowledge, including summarization, classification, and association, in image and video databases. The images and video clips used in our experiments are collected by crawling the WWW. Many challenges have yet to be overcome, such as the large number of dimensions, and the existence of multi-valued dimensions.