Biometric Recognition from Face-Voice Using Rough-Neuro-Fuzzy Classifiers

Abstract

AbstractReliable and prudent biometric person identification systems achieve authentication of the individual. In this work, eight distinct person identification algorithms are devised based on face recognition (FR), voice recognition (VR) and face-voice recognition (FVR) on video data. The lower face deformation caused by the movable jaw and the lip movement during speech is vectored into local voice-enabled face deformation profile (LVFDP). Feature fusion is employed between principal component analysis (PCA) features or linear discriminant analysis (LDA) with LVFDP and decision level fusion between FR methods, FR and VR methods. The eight algorithms are person identification using (i) VR by MFCC features and rough c-medoid clustering, (ii) VR by MFCC features and fuzzy c-medoid clustering, (iii) FR by PCA over LDA with fuzzy-multi-layer perceptron (fuzzy-MLP) classifier, (iv) FR by PCA over LDA and LVFDP with fuzzy-MLP classifier (v) FVR by PCA over LDA with f-MLP and MFCC with rough c-medoid (vi) FVR by PCA over LDA with f-MLP and MFCC with fuzzy c-medoid (vii) FVR by PCA over LDA and LVFDP with f-MLP and MFCC with rough c-medoid (viii) FVR by PCA over LDA and LVFDP with f-MLP and MFCC with fuzzy c-medoid. It is evident from the experimental results that fusion of LVFDP with facial features achieves increased recognition accuracy.KeywordsFace-voice recognitionVoice-enabled face deformation profileRough-neuro-fuzzy classifiers