Discriminant Fused Local Pattern (DFLP) In Face Recognition under Pose variations

Abstract

In [1] Karanwal et al. imposed novel local descriptor Fused Local Pattern (FLP). FLP builds its size by merging the features of MRELBP-NI, RD-LBP and 6x6 MB-LBP. FLP outperforms various individual and the several benchmark methods. After evaluating carefully the descriptor launched in [1], the one major shortcoming which is observed is that FLP builds its size by integrating only 3 descriptors. If one or more descriptors are added then discriminancy of the descriptor is assured. With this note the proposed work makes use of 4 descriptors and develops novel and discriminant descriptor called as DFLP. First three descriptor remains same as used earlier. The additional descriptor which is appended to these 3 is ELBP. Therefore DFLP feature is formed from MRELBP, RD-LBP, 6x6 MB-LBP and ELBP, by merging the features of all. PCA and SVMs are used for feature reduction and matching. For evaluating descriptors ORL used. Results suggest that accuracy enhancement is achieved by using DFLP, which beats the performance of the individually evaluated descriptors and FLP. DFLP also beats the performance of various literature techniques. The accuracy achieved by DFLP is [88.88% 93.75% 98.21%], which is much higher than the compared ones. The matlab environment used for evaluation is R2021a.