Gaussian mass optimization for kernel PCA parameters

Yong Liu; Zulin Wang

doi:10.1117/12.913296

1 October 2011 Gaussian mass optimization for kernel PCA parameters

Yong Liu, Zulin Wang

Proceedings Volume 8285, International Conference on Graphic and Image Processing (ICGIP 2011); 82854P (2011) https://doi.org/10.1117/12.913296
Event: 2011 International Conference on Graphic and Image Processing, 2011, Cairo, Egypt

Abstract

This paper proposes a novel kernel parameter optimization method based on Gaussian mass, which aims to overcome the current brute force parameter optimization method in a heuristic way. Generally speaking, the choice of kernel parameter should be tightly related to the target objects while the variance between the samples, the most commonly used kernel parameter, doesn't possess much features of the target, which gives birth to Gaussian mass. Gaussian mass defined in this paper has the property of the invariance of rotation and translation and is capable of depicting the edge, topology and shape information. Simulation results show that Gaussian mass leads a promising heuristic optimization boost up for kernel method. In MNIST handwriting database, the recognition rate improves by 1.6% compared with common kernel method without Gaussian mass optimization. Several promising other directions which Gaussian mass might help are also proposed at the end of the paper.

Citation Download Citation

Yong Liu and Zulin Wang "Gaussian mass optimization for kernel PCA parameters", Proc. SPIE 8285, International Conference on Graphic and Image Processing (ICGIP 2011), 82854P (1 October 2011); https://doi.org/10.1117/12.913296

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available