Multimodal Biometric Recognition with Optimal Feature Selection and Classification

Abstract

In many areas where personal identification is important security is of great newlineimportance Biometric or multimodal biometric systems which include the newlinephysiological and behavioral features of individuals are more preferred because newlinetraditional methods are insufficient and cannot provide security newlineBiometric recognition systems use certain human characteristics such as facial newlinefeatures fingerprint finger knuckle iris or hand geometry to identify an newlineindividual or verify their identity These systems have been developed newlineindividually for each of these biometric modalities until reaching remarkable newlinelevels of performance newlineMultimodal biometric systems combine various modalities into a single newlinerecognition system The multimodal fusion allows to improve the results obtained newlineby a single biometric characteristic and makes the system more robust to noise newlineand interference and more resistant to possible attacks The fusion can be carried newlineout at the level of the signals acquired by the different sensors of the parameters newlineobtained for each modality of the scores provided by unimodal experts or of the newlinedecision made by said experts newlineThis research tackle several important points concerning multimodal biometrics newlineThis research work is divided into three phases In first phase the exploration of newlinenew technique for fusion of biometric modalities from the face natural and nonintrusive modality and the iris one of the most precise modalities analyzes of newlinesimilarity scores originating from each modality have made it possible to develop newlinean original method of adaptive fusion combining the use of Gabor wavelet newlinetransform. Classification of extracted features is achieved by using Random forest newlineclassifier. newlineThe second phase of research work is devoted to feature selection approach in a newlinemultimodal biometric identification system based on fingerprint, finger knuckle, newlineface, and iris. Feature extraction of these modalities is accomplished by Gabor newlineWavelet Transform (GWT), Histogram of Oriented Gradient (HOG) and Local newlineBinary Patten (LBP). The Fuzzy C-Means me