Memo No . 1669 Spetember , 1999 Boosting Image Database Retrieval

We present an approach for image database retrieval using a very large number o f highly-selective features and simple on-line learning. Our approach is p redicated on the assumption that each image is generated by a sparse set of visual “causes” and t hat images which are visually similar share causes. We propose a mechanism for g enerating a large number of complex features which capture some aspects of this cau sal structure. Boosting is used to learn simple and efficient classifiers in thi s complex feature space. Finally we will describe a practical implementation of our retri eval system on a database of 3000 images. Copyright c © Massachusetts Institute of Technology, 1998. This publication can be retrieved by anonymous ftp at URL ftp ://publications.ai.mit.edu/ai-publications/ This report describes research done at the Artificial Intell ig nce Laboratory of the Massachusetts Institute of Techno logy. Support for this research was provided in part by Nippon Tele phone and Telegraph under grant number 9807-NTT03. 1http://www.ai.mit.edu/projects/lv