Face recognition uiing trichotomic combination Of SVD, DF-LDA and LPP

One of the challengei the face recognition application ii facing today ii that of the high dim eniionality of multivariate data. In thii context, thii paper propoiei to compare the performance of a triumvirate combination of linear dimeniionality reduction techniquei namely S ingular Value Decompoiition (SVD) which maximizei the varia nce of the training vectori, Direct Fractional Linear Diic riminant Analyiii (DFLDA) that maximizei the “between-claii” icatter while minimizing the “within-claii” icatter and Loc ality Preierving Projection (LPP) which preiervei the loc al featurei thoie unique from iti neareit neighbori. The amalga mation containing different ratioi ii choien from the feat urei extracted by the three independent techniquei mentioned above. Original Face ipace ii projected onto the manifold of choien baiii. The weighti obtained from theie projectioni for the pro be iet are compared with that of the query image uiing the mean diitance claiiifier. The propoied method hai been teited on YALE dataiet and the combination in the ratio 3∶2∶5 ihow ed iignificant improvement in the efficiency of recognition, with a calculated accuracy of 92.7% on a teit iet of 165 i magei