|
In our proposed research, we developed a discriminative model for addressing the challenge of face recognition with respect to age variation by utilizing periocular region. Initially, we represent each face by using a densely sampled local feature description technique, in which multi-block local binary pattern (MB-LBP) and weber local descriptor (WLD) are adopted as the local descriptors. In this approach, extract the feature of entire facial image by the aid of both local descriptors. Since the feature space of both the MBLBP and WLD is vast, we develop an advanced multi-feature discriminant analysis to process these two local feature spaces in a unified framework. By random sampling, the training set as well as the feature space, multiple linear discriminant analysis ( LDA) based K-nearest neighbor (K-NN) classifiers are constructed and then combined to generate a robust decision via a fusion rule. In the proposed method, the MORPH and FG-NET datasets are utilized for the experimental evaluation. In the MORPH dataset, our proposed method is compared with existing methods such as SIFT, Multi scale method based on morphology, signed grey level difference (SD), LBP, scale-invariant classification, gabor, hidden factor analysis (HFA), modified HFA (MHFA) and learning discriminant face descriptors (LFD). The accuracy of face and periocular region for our method is 98.83% and 98.32%. The existing method such as sparse null linear discriminate analysis (SNLDA), direct LDA, principle component analysis + LDA, cross-age reference coding (CARC), HFA and sparse representation classification (SRC) are compared with the proposed method and obtained the recognition rate is 96.74% and 97.89%. In parameter evaluation we get following values: recall is 93.34% and 96.56%, precision is 95.49% and 94.76%, F-score is 94.40% and 92.17% for face and periocular region. In the FG-NET dataset, our proposed method is compared with existing methods such as SIFT, LBP, MLBP, MWLD and multi-feature discriminant analysis (MFDA). The average accuracy in face region and the periocular region is 65.32% and 68.92%. Our proposed method produce recognition rate as 98.11% and 97.87% for face region and periocular region when compared with SNLDA, DLDA, SRC, PCA + LDA and Coupled Auto-encoder Networks (CAN). In parameter evaluation we get following values: recall is 90.85% and 91.65%, precision is 92.35% and 94.56%, F-score is 91.59% and 90.84% for face and periocular region. Our Method has error rate as 1.7% and our error rate is less when contrasted with Agarwal et al. (21%), Wallis et al. (11.5%), Wang et al. (4.515%) and Leibe et al. (2.5%) techniques.
|
