Al-Based Approach for Person Identification Using ECG Biometric


The widespread computerisation has tremendously popularised the various digital payment modes, be it online transactions, e-shopping, ATMs, etc. This along with the increased security threat due to menace of terrorism has necessitated the requirement of reliable human identification systems. Biometrics, which utilise physiological (face, fingerprint, iris, etc.) or behavioural (speech, gait, keystroke, etc.) human traits, has provided the much needed solution in this regard. These systems are being extensively employed for various authentication tasks like attendance, passports, citizen registers, etc. The strength of a biometric trait is usually assessed on the basis of seven parameters, namely, uniqueness, universality, permanence, performance, acceptability, collectability, and circumvention. However, none of the presently utilised traits possess all these qualities individually. For example, iris in certain individuals may be damaged due to some eye disease or congenital defect, thus violating the universality trait, while speech tends to vary even in case of simple throat infection. Fingerprints also have some acceptability issues because of their linkage with criminal identification. Moreover, as the popularity of these new age security tools has risen, so has the tendency of the fraudsters to find means in order to deceive or attack these systems. Attempts have been made to fool fingerprint and speech- based biometric systems respectively by use of synthetic template of fingerprints prepared from materials such as latex gelatine etc. and pre-recorded speech utterances. Furthermore, these type of person recognition systems need to be incorporated with a liveness detection mechanism so as to ensure that sample has been obtained from a living individual.

In order to overcome the aforesaid issues, in the past decade, researchers have explored the possibility of developing human recognition system based on bioelectric signals like electrocardiogram (ECG), electroencephalogram (EEG), and photople- thysmogram (PPG). The bioelectric signals have been found to possess characteristics suitable for biometric applications either in unimodal or multimodal configuration. In this chapter, an approach related to ECG-based biometric recognition has been described but before discussing it, an overview of biometrics and its various modes is presented and the same is summarised in Figure 6.1.

Design of a typical biometric system begins with an enrolment phase in which the templates from prospective users of the technology are stored in a database. After the enrolment phase during testing, all biometric systems can operate in two modes - verification or identification. In the verification mode, identity claim made by the user is either accepted or rejected, so while going through the testing phase, the template of the individual who has claimed to be genuine user is compared with his already stored template. Identification on the other hand is more difficult and requires comparison with all templates stored in the database. For the closed set case, i.e., individual to be identified has his template already stored in the database, identification is carried out by finding the most similar template among the enrolled subjects. Whereas for the open set case, individual to be identified may or may not be already enrolled, so in addition to being most similar, the score between the matched template and test sample should be below some pre-decided threshold value. Briefly stated, in verification answer is found to the query “Am I who I claim to be”, while the question which is answered in identification is “Who I am”.

On the basis of permanence property, biometric traits can be categorised as hard and soft. Hard biometrics are those which do not vary drastically over sufficient duration of time and have high uniqueness, while soft biometrics are those which provide some information about a person but lack the uniqueness and permanence property. Hard biometrics can be further divided on the basis of type of trait used. Physiological biometrics are those in which the sample is obtained from a physical characteristic, e.g., fingerprint and face, and are inherent trait of a human being. On the other hand, behavioural biometrics are the ones which are acquired by humans, e.g., walking style and handwriting [1]. As behaviour will always influence interaction of the user with biometric sensor, so all biometric traits have some behavioural

An overview of biometrics

FIGURE 6.1 An overview of biometrics.

aspect associated with them. Moreover, some researchers also consider mixed category for classifying biometric modalities like speech as it is dependent on anatomy of human vocal apparatus besides being strongly influenced by individual’s behaviour. As stated earlier, physiological and behavioural traits may be absent in some humans because of a diseased state or some other reason and as such do not fulfil the universality trait. However, medical biometrics based on ECG, EEG, and PPG provide solution to these issues as they inherently possess universality trait and will ensure entire population coverage. This provides another classification for biometrics where medical biometrics along with other upcoming biometrics like odour, lip-pint, etc. are clubbed under the heading of esoteric biometrics, while biometrics with mature technology like fingerprint, face, etc. can be grouped under traditional biometrics.

In order to explain the approach presented here for ECG-based biometric recognition, first the description of ECG waveform, its generation, and review of the literature related to ECG biometric is given in the following section.

< Prev   CONTENTS   Source   Next >