Learning Representative Objects from Images Using Quadratic Optimization

Xiaonan Lu, Jia Li, James Z. Wang
The Pennsylvania State University, University Park, PA 16802

With the development of Content-Based Image Retrieval (CBIR) and ever increasing computing power, there is a notable growing interest in automatic learning from images. In this paper, we introduce a quadratic optimization based learning technique to enable computers to learn visual characteristics of a semantic concept from unlabeled images. In our work, images are represented by regions extracted from segmentation. Given a group of images conveying a semantic concept, we attempt to detect the region corresponding to the concept in every image using quadratic optimization. To characterize the visual properties of the concept, the mean of the feature vectors each describing the concept-associated region of an image is calculated and referred to as the representative feature vector. We apply the proposed learning technique to image classification and object recognition applications and provide experimental results.

Full Paper in Color
(PDF, 0.6MB)

On-line Info

Citation: Xiaonan Lu, Jia Li and James Z. Wang, ``Learning Representative Objects from Images Using Quadratic Optimization,'' Proc. International Conference on Machine Intelligence, invited for a special session, pp. 730-737, Tozeur, Tunisia, ACIDCA, November 2005.

Copyright 2005 ACIDCA. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works, must be obtained from the ACIDCA.

Last Modified: June 30, 2005