Face Recognition Using The Principal Components Of The Scatter Matrix In The Frequency Domain

Abstract

This paper develops a Linear Discriminant Analysis based face recognition system in the Discrete Cosine Transform (DCT) domain as a departure from the traditional analysis in the spatial domain. In the training mode, the truncated DCT coefficients are used to find discriminating features for all the subjects in the image database. The compact representation of the truncated DCT coefficients leads to notable reductions in data dimensionality and also bypasses the well-known Small Sample Size problem. The input data is projected on the dominant Eigenvectors of the scatter matrix capturing within-class and betweenclass information. To further enhance the system performance, a correction matrix computed in the training mode depending on a penalty function is used to adjust the Euclidean distances between the testing and the training poses. The final classification decision is based on the minimum distance. The ORL, YALE, FERET, and FEI databases are used to evaluate the system performance. The proposed system is shown to achieve higher recognition rates, reduced computational complexity and low storage requirements compared to its existing counterparts.

Publication Date

7-2-2016

Publication Title

Midwest Symposium on Circuits and Systems

Document Type

Article; Proceedings Paper

Personal Identifier

scopus

DOI Link

https://doi.org/10.1109/MWSCAS.2016.7869955

Socpus ID

85015883505 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/85015883505

This document is currently not available here.

Share

COinS