This paper addresses the use of Linear Discriminant Analysis for recognition of human faces. It presents the effect of various similarity measures and the dimensionality of underlying PCA subspace on the recognition rate of the system. Commonly used 4 similarity measures such as City block, Euclidean, Cosine and Mahalanobis are tested. In order to test the performance of the face recognition using Linear Discriminant Analysis, various experiments are carried out on AT&T face database. AT&T is publicly available database which has 400 images of 40 different persons. It was observed that changing similarity measure caused significant change in the performance of the system. The performance improved with the dimensionality of the final subspace. We achieved the best recognition performance using Cosine distance measure. It is observed that recognition rate depends on the dimensionality of underlying PCA subspace as well as on the similarity measure used.

1. Jain, A. , Ross, A. , and Karthik, N. 2011. Introduction to biometrics.
2. Belhumeur, P. , Hespanha, J. , and Kriegman, D. 1997. Eigenfaces vs. fisherfaces: recognition using class specific linear projection. IEEE Trans. Pattern Anal. Mach. Intell. 19 (1997), 711-720.
3. Belhumeur, P. , Hespanha, J. , and Kriegman, D. 1996. Eigenfaces vs. fisherfaces: recognition using class specific linear projection. In Proceedings of the 4th European Conference on Computer Vision.
4. Borade, S. N. , and Adgaonkar, R. P. 2011. Comparative analysis of PCA and LDA. In Proceedings of the International Conference on Business, Eng. and Industrial Applications.
5. Turk, M. , and Pentland, A. 1991. Eigenfaces for recognition. J. Cogn. Neurosci. 3 (1991), 71-86.
6. Borade, S. N. , Deshmukh, R. R. , and Shrishrimal, P. Effect of distance measures on the performance of face recognition using principal component analysis. In Berretti, S. , Thampi, S. M. , and Srivastava, P. R. (eds. ) Intelligent Systems Technologies and Applications, AISC, 384 (2016), 569-577.
7. Mathwork help for similarity measures, http://www. mathworks. com/help/stats/pdist2. html.
8. Perlibakas, V. Distance measures for PCA-based face recognition. 2004. Pattern Recogn. Letters. 25 (2004), 711-724.
9. ORL face database, AT&T laboratories, Cambridge, U. K. ,http://www. cl. cam. ac. uk/research/dtg/attarchive/facedatabase. html.
10. Martinez, A. M. , and Kak, A. C. PCA versus LDA. 2001. IEEE Trans. Pattern Anal. Mach. Intell. , 23 (2001), 228-233.

Biometrics Face Recognition Principal Component Analysis Linear Discriminant Analysis