Title:	A new extension to kernel entropy component analysis for image-based authentication systems
Authors:	Varadharajan, Vijay
Citation:	Journal of WSCG. 2015, vol. 23, no. 1, p. 1-7.
Issue Date:	2015
Publisher:	Václav Skala - UNION Agency
Document type:	článek article
URI:	http://wscg.zcu.cz/WSCG2015/!_2015_Journal_WSCG-No-1.pdf http://hdl.handle.net/11025/17144
ISSN:	1213–6972 (hardcopy) 1213–6980 (CD-ROM) 1213–6964 (online)
Keywords:	FDKECA;KECA;rozpoznávání obrazu;mapování prostoru;PCA
Keywords in different language:	FDKECA;KECA;image recognition;mapping space;PCA
Abstract in different language:	We introduce Feature Dependent Kernel Entropy Component Analysis (FDKECA) as a new extension to Kernel Entropy Component Analysis (KECA) for data transformation and dimensionality reduction in Image-based recognition systems such as face and finger vein recognition. FDKECA reveals structure related to a new mapping space, where the most optimized feature vectors are obtained and used for feature extraction and dimensionality reduction. Indeed, the proposed method uses a new space, which is feature wisely dependent and related to the input data space, to obtain significant PCA axes. We show that FDKECA produces strikingly different transformed data sets compared to KECA and PCA. Furthermore a new spectral clustering algorithm utilizing FDKECA is developed which has positive results compared to the previously used ones. More precisely, FDKECA clustering algorithm has both more time efficiency and higher accuracy rate than previously used methods. Finally, we compared our method with three well-known data transformation methods, namely Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), and Kernel Entropy Component Analysis (KECA) confirming that it outperforms all these direct competitors and as a result, it is revealed that FDKECA can be considered a useful alternative for PCA-based recognition algorithms.
Rights:	© Václav Skala - UNION Agency
Appears in Collections:	Volume 23, Number 1 (2015)

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.