Palm and palm finger joint surfaces based multibiometric approach

Abstract

This paper presents a novel multibiometric approach based on palm and palm finger joint surfaces which are obtained from same image. In this approach, the patterns (Region-Of-Interests, ROI) were extracted by using Active Appearance Model (AAM) based hand modeling. Then preprocessing steps which include image normalization and Discrete Wavelet Transform (DWT) are sequentially applied to the both palm and palm finger joint surfaces (ROI). Afterwards these two biometric traits are joined to be long vector. Most discriminative features of long vector are extracted by using Kernel Fisher Discriminant (KFD). Finally, Support Vector Machines (SVM) are implemented for classification. Furthermore, a feature level fusion is also used in order to comparately show the performans of the multibiometric approach's success. The proposed multibiometric approach was tested on 1614 palm and palm finger joint surface images which were captured from 132 different people. The recognition results were obtained by utilizing both long vector structure and feature level fusion strategies. Moreover the palm based unibiometric approach and the palm finger joint surfaces based unibiometric approach were separately tested to make a comparison with the proposed multibiometic approach. The achieved results have demonstrated the proposed multibiometric approach's success.