Sequential Updating Algorithm For Extracting The Basis Of The Karhunen-Loeve Transformation (TA-P4)

Author(s) : Yanyun Qu (Department of Computer Science of Xiamen University, China) Nanning Zheng (Institute of Artificial Intelligence and Robotics of Xi'an Jiaotiong University, China) Zejian Yuan (Institute of Artificial Intelligence and Robotics of Xi'an Jiaotiong University, China) Cuihua Li (Department of Computer Science of Xiamen University, China)

Abstract : Karhunen-Loeve transformation (KLT) is a popular method for dimensional reduction and feature extraction in image analysis, signal processing and automatic control systems, and so on. The drawback of the KLT is expensive computation. In this paper, we propose a novel updating algorithm for KLT, rank-k updating algorithm, which has advantages especially for image sequences: it is faster than batch algorithm, and can handle the dynamic database, and doesn't save the entire database. Furthernore it make the active learning and recognition possible in computer vision. And finally we analyze the computation complexity and error of the algorithms. We show applications in face analysis. The experiment results demonstrate the efficiency of our algorithms

Menu