A MULTIMODAL COMBINED MACHINE LEARNING APPROACH FOR FINGERPRINT CLASSIFICATION

Abstract

Fingerprint identification is the most widely used biometric for a multitude of security applications ranging from phone unlocks to bank security. All modern systems use a machine learning approach based on a unique algorithmic - such as Support Vector Machines(SVM) , Convolutional  Neural Networks(CNN) or Residual Convolutional Neural Network(RESCNN).Each and every algorithm is strong in some areas and weak in others. In our work we describe the output yielded using a new proposed algorithm that uses a trifecta combination, i.e three different algorithms are combined to maximise their individual strengths and cover each other's weaknesses. The algorithms we combine are a simple preprocessing algorithm, SVM and a trained convolutional neural network. First, the preprocessing algorithms smoothens, sharpens and filters the image, then an SVM is used to extract the minutiae (fingerprint features) which is finally classified using a trained CNN classifier. This new algorithmic approach will have enhanced accuracy, faster processing time and lower error than the traditional unilateral algorithmic approaches.