MFCC and VQ voice recognition based ATM security for the visually disabled

Abstract

A biometrie based automatic teller machine which works with a two-tier security using voice and fingerprint recognition can help users which are visually challenged, allowing them to use the machine using only their biometric characteristics. An automated teller machine (ATM) requires a user to pass an identity test using their PIN before doing any financial transactions. The current method available for access control in ATM is based on cards and pins which increases the issues of unauthorized access on accounts via card skimming. It is eminently difficult to avoid another person from attaining and using someone else's card also regular smartcards can be lost, duplicated, stolen or falsified with accuracy. Another concern is the accessibility of ATM to differently abled people. These concerns can be overcome by using fingerprint recognition for authentication, as discussed from the researchers' previous study, and by adding up an additional voice recognition system feature as discussed on this paper. The four fingerprint sample patterns of an individual are completely separate and uncorrelated. The action of fingerprint recognition involves pre-processing, feature extraction and minutiae matching. Matching is done by comparing the user's fingerprint with the existing fingerprint database, images which were acquired at the time of opening an account in the bank account. Once the fingerprint of the user passes the authentication procedures of the system the user is now able to carry out further transactions using voice-based commands by speaking through a microphone. This model not only provides security but also accessibility to certain sections of the population like people with visual impairment and eye disabilities. The system uses biometric based user recognition. The authenticity of the account will be checked by the input of the user's fingerprint, this will then allow further transaction via voice recognition that implements a MFCC, DWT and VQ to continue. ATM users are identified by the lowest VQ distortion of each voice input. Out of 50 different legitimate user trials, 42 tries were identified while 8 legitimate user tries were denied of access in the system producing 84% accuracy.