Multimodal biometric verification and identification using face and hand

Abstract

The goal of this thesis is to develop user interfaces that identify and verify legitimate users in real-time due to their biometric characteristics. Face and hand geometry were the biometric identifiers that were used in the applications. Application programs using these biometric identifiers were developed in unimodal and multimodal cases. In this study, six Java applications were developed to capture users' hand and face images in real-time, to enrol, to verify and identify them. All of the applications were designed as wizard applications and were built upon the same scenario. In the face recognition, Gabor jets were used to extract feature values of the face images. Matching the person was done according to the Gabor based feature vectors. In the hand recognition task, hand shape-based approach was applied on the hands. Hand segmentation, normalization and feature extraction were the important steps of this approach. In order to test the performance of the applications, data were acquired from 40 people. Recognition performances of the algorithms were analyzed for face, hand and fusion of two modalities. Besides performance, user evaluation was carried out in order to improve the usability of the user interface. A questionnaire regarding the speed, user interface and usability of the applications were collected by the people whose data were acquired. The interface was revised after the recommendations of the users.