An Efficient Template Choosing Method of Derived Kernel for Object Recognition
Speaker:Prof. Li Hong
School of Mathematics and Statistics
Huazhong University of Science and Technology
Date & Time:04 Dec 2009 (Friday) 16:00 - 17:00
Venue:JM08

Abstract

With huge amounts of digital images generated by various means, efficient and effective object recognition has become a hot issue. In this talk, a novel method is proposed to improve the object recognition performance of the derived kernel by choosing effective templates. The method we proposed considers not only the label information of the training images, but also the elimination of redundant information. This new algorithm is easy to implement, and it has merit of improving computational efficiency of the derived kernel. Extensive experiments on four standard databases show that the derived kernel based on our method obtains higher accuracy with a small number of templates.