EFFECT OF FACE AND OCULAR MULTIMODAL BIOMETRIC SYSTEMS ON GENDER CLASSIFICATION

Abstract

This study is concerned with analysing face–ocular multimodal biometric systems for a person gender prediction. Particularly, this is the first study considering fusion of face and ocular biometrics to predict gender of a person via a hybrid multimodal scheme. The authors aim to investigate the effect of multimodal biometric systems at score and feature-level fusion on gender classification. The implementation of uniform local binary pattern (ULBP) feature extractor is taken into account to extract the face and ocular texture information. This paper proposes to select the efficient feature sets of both modalities using a novel evolutionary algorithm called backtracking search algorithm (BSA). On the other hand, support vector machine (SVM) is applied for classification purpose using the fused face and ocular features and scores. The proposed scheme is validated using CASIA-Iris-Distance and MBGC multimodal biometric databases with consideration of a subject-disjoint training and testing evaluation. The achieved gender recognition demonstrates the superiority of the hybrid multimodal face–ocular scheme over unimodal face and ocular schemes implemented in this study for a subject gender prediction.

Existing System

2D wavelet tree, back propagation neural networks (BPN)

Proposed System

The proposed scheme attempts to combine feature and score level fusions to explore the effect of these two fusion levels of face and bio-ocular regions on gender prediction of individuals. The features of all modalities are extracted using ULBP feature extractor under 50% overlapping of each block and 60% selection of computed histogram of each sub-window. The extracted optimised features are then concatenated in order to fuse face and bio-ocular information of the individuals in feature level fusion. The BSA feature selection strategy is applied after concatenation of all modality histograms one time more in this proposed scheme to reduce the effect of high dimensionality and computational time and subsequently improving the classification rate. Indeed, SVM combines the produced matching scores of modalities into a feature vector and then classifies them to get the results of subjects gender decisions.

Architecture

BLOCK DIAGRAM