Fingerprint Recognition By Using Convoloutional Neurla Network And Support Vector Machine Classification

Biometrics seeks to solve the problems of traditional verification methods by using certain physiological properties associated with an individual. Among all the biometric indicators, fingerprints have been shown to have good levels of reliability. The most widely used local representation is based on the details (minutiae) of the fingerprints. The pattern of the minutiae on a fingerprint forms a valid representation of the fingerprint. The minutiae that are most used for automatic recognition are branches and endings. However, given fingerprint acquisition techniques, it is common for endings and bifurcations to undergo deformations, which is why they are commonly referred to as minutiae. That is why in this document we will simply refer to these characteristics as minutiae. In this work we describe the results obtained using a methodology proposed for the recognition of minutiae using convolutional neural networks CNN, trained with different databases that contain fingerprints then we use the support vector machine classification to classify newly input images of fingerprints based on the features extracted by the CNN and matched with the dataset, our method proves to have better accuracy and lower MSE than the previous linear methods use for fingerprint recognition.

I Introduction

Because biometrics is a relatively new science whose goal is to make computers capable of recognizing each person with certainty based on their unique characteristics, such as fingerprints, features, or the signs of their eyes; A biometric prototype is implemented that is capable of recognizing people through the fingerprint. There are several types of measurement that can be moved for biometrics. The ones most know, such as fingerprint, hand geometry, and eye analysis, are also the best performing today. Thanks to the characteristics of fingerprints; which are unique and invariant over time for each individual, a prototype is implemented in which people are automatically identified, based on the characteristics of the fingerprint of the index finger. –– et.al []. was the first to study the brain as a way of looking at the world of computing. However, the first theorists who conceived the foundations of neural computing were Warren McCulloch, a neurophysiologist, and Walter Pitts, a mathematician, who, in 1943, launched a theory about the way neurons work. They modeled a simple neural network using electrical circuits. –– et.al []. was the first to explain the learning processes (which is the basic element of human intelligence) from a psychological point of view, developing a rule of how learning occurred. Even today, this is the foundation of most of the learning functions that can be found in a neural network. His idea was that learning occurs when certain changes in a neuron are activated. He also tried to find similarities between learning and nervous activity. Hebb’s works formed the foundations of Neural Network Theory. –– et.al [].In his series of essays, he found that information was not centrally stored in the brain but distributed above it. –– et.al []. This Congress is frequently mentioned to indicate the birth of artificial intelligence. –– et.al [].The development of the Perceptron began. This is the oldest neural network; being used today for application as pattern identifier. This model was able to generalize, that is, after having learned a series of patterns, it could recognize other similar ones, even if they had not been presented in training. However, it had a number of limitations, for example, its inability to solve the problem of the exclusive-OR function and, in general, it was unable to classify non-linearly separable classes. Lavanya et.al [1]..