Kernel class-dependence feature analysis (KCFA)


C. Xie,M. Savvides and B.V.K. Vijaya Kumar

Kernel correlation filter based redundant class-dependence feature analysis (KCFA) on FRGC2.0 data

Proceedings of the Second International Conference on Analysis and Modelling of Faces and Gestures (AMFG) 2005, pp.32-43, 2005)


author={Xie, C. and Savvides, M. and Vijaya Kumar, B.V.K.},
booktitle={Analysis and Modelling of Faces and Gestures},
series={Lecture Notes in Computer Science},
title={Kernel Correlation Filter Based Redundant Class-Dependence Feature Analysis (KCFA) on FRGC2.0 Data},



In this paper we propose a nonlinear correlation filter using the kernel trick, which can be used for redundant class-dependence feature analysis (CFA) to perform robust face recognition. This approach is evaluated using the Face Recognition Grand Challenge (FRGC) data set. The FRGC contains a large corpus of data and a set of challenging problems. The dataset is divided into training and validation partitions, with the standard still-image training partition consisting of 12,800 images, and the validation partition consisting of 16,028 controlled still images, 8,014 uncontrolled stills, and 4,007 3D scans. We have tested the proposed linear correlation filter and nonlinear correlation filter based CFA method on this FRGC2.0 data. The results show that the CFA method outperforms the baseline algorithm and the newly proposed kernel-based non-linear correlation filters perform even better than linear CFA filters.



Two Class Minimax Distance Transform Correlation Filter


M. Savvides, B. V. K. Vijaya Kumar, and P. K. Khosla

Two Class Minimax Distance Transform Correlation Filter

Appl. Opt. 41, 6829-6840 (2002))


author={Savvides, M. and Vijaya Kumar, B. V. K. and Khosla, P.},
journal={Applied Optics},
title={Two Class Minimax Distance Transform Correlation Filter,},


A new correlation filter formulation (that we refer to as the minimax distance transform correlation filter (MDTCF) is presented that minimizes the average squared distance from the filtered desired (or true-) class training images to a filtered reference image while maximizing the mean squared distance of the filtered undesired (or false-) class training images to this filtered reference image. This approach increases the separation between the false-class correlation outputs and the true-class correlation outputs. Classification can be performed using the squared distance of a filtered test image to the chosen filtered reference image. We show that the previously introduced distance classifier correlation filter (DCCF) is similar to a special case of MDTCF. We also examine the differences between the DCCF and the MDTCF, and show that MDTCF can offer increased discrimination performance. Also, MDTCF performance is evaluated on two different face databases.


Quad phase minimum average correlation energy filters for reduced memory illumination tolerant face authentication


In this paper we propose reduced memory biometric filters for performing distortion tolerant face authentication. The focus of this research is on implementing authentication algorithms on small factor devices with limited memory and computational resources. We compare the full complexity minimum average correlation energy filters for performing illumination tolerant face authentication with our proposed quad phase minimum average correlation energy filters[1] utilizing a Four-Level correlator. The proposed scheme requires only 2bits/frequency in the frequency domain achieving a compression ratio of up to 32:1 for each biometric filter while still attaining very good verification performance (100% in some cases). The results we show are based on the illumination subsets of the CMU PIE database[2] on 65 people with 21 facial images per person.

Advanced Correlation Filters

Results from NIST Face Recognition Grand Challenge. The correlation filter based approach designed by Professors Marios Savvides and Vjayakumar Bhagavatula outperformed all other algorithms (industry and academia). Image courtesy of