Separability measures for error estimation of two normally distributed classes

Joonyong Hong; Chulhee Lee

doi:10.1117/12.323225

1 October 1998 Separability measures for error estimation of two normally distributed classes

Joonyong Hong, Chulhee Lee

Proceedings Volume 3460, Applications of Digital Image Processing XXI; (1998) https://doi.org/10.1117/12.323225
Event: SPIE's International Symposium on Optical Science, Engineering, and Instrumentation, 1998, San Diego, CA, United States

Abstract

In pattern classification and remote sensing, the Gaussian ML classifier is most widely used because of its speed and robustness. In this paper, we propose to use two separability measures, Bhattacharyya distance and divergence to estimate the classification error of the Gaussian ML classifier. In the proposed method, we try to find empirical relationship between the separability measures and the classification error. In order to find such relationship, we generate two classes with normal distribution and compute the separability measures and classification error between the classes. Although there are infinite number of possibilities that two classes can have, we systematically search the whole mean-covariance space. From this exhaustive search, we are able to estimate the classification error accurately using the Bhattacharyya distance and divergence. It is observed that the error estimation using both the Bhattacharyya distance and divergence does not give a significant improvement over the error estimation using the Bhattacharyya distance only.

Citation Download Citation

Joonyong Hong and Chulhee Lee "Separability measures for error estimation of two normally distributed classes", Proc. SPIE 3460, Applications of Digital Image Processing XXI, (1 October 1998); https://doi.org/10.1117/12.323225

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