Multimodal Biometric Authentication

To manage the security issues in places where access to a place, device is restricted, biometrics are used for authen-tication. Palm and signature are the most reliable physiological and behavioral biometric features for managing security issues. These biometrics are preferred as they have high accuracy, low start-up costs, acceptance, and stability. Looking at the texture of the palm we used two types of feature extraction processes, namely the Gabor filters based on the Gray Level Co-occurrence Matrix (GLCM), the mean and variance of the image of the palm and signature. Texture analysis with GLCM and low pixel values are considered. Based on Correlation, Contrast, Angular Second Moment (ASM), Homogeneity, Entropy, Mean and Variance features are extracted and are fused at feature level. Features are classified by using Support Vector Machine (SVM) classifier. In addition to feature level fusion and SVM classification, transfer learning with VGG16 is used for feature extraction of signature and palm print separately. Classification of these features are implemented using Convolutional Neural network (CNN) with score level fusion and decision level fusion. GUI using python is employed for demonstration. The authentication system has given us reasonably good results.