“Biometrics” is a term used to refer to “life measurement.” Biometrics is a technology used for verifying authenticity (Stanoevska-Slabeva, 2011). It uses either behavioral features or biological traits to identify or recognize individuals. In actual sense, Biometric system translates to a system of pattern identification which makes use of various biological characteristics such as facial recognition, voice geometry, fingerprints hand recognition and patterns like retina designs and iris patterns (Saini, 2014). The recognition systems and models of biometrics allow for the verification possibility of people’s identity by only determining who they are and not what they might be remembered of or what they possess. This is made interesting because some security credentials such PIN numbers and passwords could be exchanged among individuals while physical characteristic can never be interchanged.
Biometric recognition methods and performance comparisons
This biometric system entails detailed study and interpretation of facial characteristics. It is a computer-based application system for verifying or determining individuals automatically from digital pictures or images or video frameworks generated from video sources. An easier method of doing this could be evaluating chosen facial characteristics from a pictorial directory. As compared to other biometric systems, facial recognition design is user-friendly, i.e., it has contactless authentication. It also performs robust identification which other biometric systems usually cannot show. Facial recognition systems do not need any direct contact with an individual for their identities to be verified. This capability could be of a great deal in clean environs for tracking or monitoring, and also in automated systems (Bhatia, 2014). Again, this biometric system does not rely on the subject co-operation to perform any activity. Given that this system can identify individuals from an enormous crowd with absolute ease, it can, therefore, be set up in multiplexes, airports and many other open public environments of the likes. Also, facial recognition can perform security incident monitoring with which a photo is captured by a camera, and there is no fingerprint technology evidence for tracking such incidents.
On the other hand, facial recognition possesses certain performance related inefficiencies. This may include instances where images have low resolutions, high associated complexity, and costs when deployed for security incidents. Additionally, it is usually less efficient when facial expressions vary or in weak or poor lighting, or other blockages are partially covering faces of the subjects.
This biometric system provides one of the best-secured strategies as far as recognition and authentication are concerned. The moment the impression of an iris has been captured by a typical digicam, the validation process entails an evaluation of the subject’s iris with saved versions. This system is one of the most concrete methods with little lapse and rejection levels. Iris recognition performance, in comparison to other biometric systems, possesses unique structure shaped by the age of ten months, thus becomes ever staunch throughout life (Soyjaudah, 2013). It captures subtle textures since genetically same people pose entirely independent iris textures. The data collection process is non-instructive, i.e., actual scanner contact is not needed while contact glasses or lenses ensure data capturing. It possesses high and accurate recognition processing speeds with easy identification of false or fake iris. Their security systems are seen as one of the most reliable today as it is an easy and distinctive way of identifying users.
Human fingerprints have numerous valleys and ridges on finger surfaces which are unique to every human being. Fingerprint identification and authentication is a real automated technique of evaluating a match between separate human fingerprints.
Fingerprint systems are easy to install and use with requirements for inexpensive equipment taking little power. Fingerprint patterns have individualistic distinctive characteristics and compositions that stay unchanged with time (Park, 2012). There is no need for password remembrance as swiping of fingers on scanners is all that which is required. The biometric scanners provide a method of recording an identification point that can never be fake. This technique makes this process incredibly safe and accurate. However, fingerprint system may be susceptible to errors since it only scans one part of the human finger.
Voice recognition methods and performance comparisons with mathematical algorithms
This is a technology by which words, phrases, and sounds voiced by people are converted into electrical form of signals that are thereafter transformed into code form. Individuals can be identified by their voice output containing a set of measurable traits in a human voice outcome. This technology emphasizes on the spoken input contents and not the person speaking, and it relies on the algorithms such as vector quantization, frequency estimation, and Markov models. They are used in voice recognition methods as highlighted below:
Text-dependent method: Here the speaker is asked to utter a phrase or word that was earlier saved during the process of enrollment. The measure of correlation is determined by matching initially recorded input vectors with the sequence of appeared vectors of the said information.
Text-prompted method: Here the speaker is prompted to read again or repeat a phrase or word from pre-recording terms displayed from the system.
Text-independent method: Here there are no prior records but the system needs to be coached by the speaker to do official recognitions. During the coaching stage, templates for reference are produced for separate phonetic sounds of people’s voices and not samples for particular words. During the operation phase, acquired phonetic references are matched by the system with sounds from a random input. It is hard to design this system even though it provides superior protection against fraudulent and imposing practices (Bhatia, 2014).
Voice Recognition Algorithm:
Voice recognition system manages to work due to the algorithm that it does in digitally verifying the voice matching capabilities. The initial phase occurs when the digital signal is transmitted within a low-order filter to flatten spectrally and also to enable it less affected by tiny accuracy effects. This screen gives a transfer function of the order (Doyle, 2010):
Value for a typically varies from 0.9 to 1.0. Commonly assumed values, a = 0.9375. The signal is after that inserted into frames that are 256 samples long. A correspondence is of about 23 ms of sound in each frame. All frames are then passed to a Hamming window which is used to reduce the discontinuities at the start and finish of every frame. This Hamming window poses the equation (Doyle, 2010):
N is the counts of samples in each frame. Commonly used, N = 256. In the next phase, all windowed frames are auto-correlated to lower the average square projected error established during the LPC phase. Autocorrelation equation is:
P is, therefore, the order of the LPC coefficients, Hence p = 13 when the above values of a and N are assumed. The features in every frame are generated by linear predictive coding (LPC) represented by equation:
Sn is the nth sample speech, and the ai the coefficient of the predictor, and sn imply the forecast of the nth figure of the speech signal.
Cepstral coefficients are then calculated from the LPC coefficients. They are used because they are regarded to be more comprehensive and dependable than PARCOR coefficients, log area ratio coefficients or the LPC coefficients. Recursion is applied to this equation to obtain the Cepstral coefficients (Doyle, 2010):
a k are the LPC coefficients, p = 13, and c(i) equal the Cepstral Coefficients. Cepstral coefficients tend to be very sensitive to noise related variability. Weighted Cepstral coefficients are used, and in the form:
w m is found from:
Still, p = 13
Results for Comparison
For comparison purposes we can use the Blind Equalization technique for normalization as it the most effective in reducing distortions of the channel caused by variances in the frequency features of the input device.
w is the weighing parameter, InE shows current frame’s energy log, bias[i] is always initialized on 0.0 i.e. (0<i<13) and CRef will be the reference cepstrum. Mean Cepstral coefficients are done over its period thereafter those values are deducted from each frame’s Cepstral coefficient.
The final result is an i x j matrix, where i is LPC order and j frame counts. Given the essential features have been extracted; the final results can, therefore, be used by DTW and VQ to correlate the word and speaker respectively. Hence, if DTW and VQ match, then the system should allow the user for being the authenticated voice owner (Doyle, 2010). Alternatively, the Cepstral coefficient generated by the transfer equation (first algorithm) can be compared to the Cepstral coefficient obtained through Blind Equalization technique (second algorithm).
Saini, R. (2014). Comparison of Various Biometric Methods. IJAST International Journal of Advanced Science and Technology, 2(1), 24-30.
Stanoevska-Slabeva, K. (2011). Cloud Basics – An Introduction to Cloud Computing. Grid and Cloud Computing, 47-61.
Biometric Authentication for Cloud Computing. Handbook of Research on Securing Cloud-Based Databases with Biometric Applications, 209-235.
Roberts, C. (2013). The Evolution of Cloud Computing. Implementation, Management, and Security Bhatia, T. (2014). Cloud Computing, 1-28.
Soyjaudah, K. M. (2013). Cloud computing authentication using cancellable biometrics. 2013 Africon, 25-39.
Park, J. J. (2012). Future information technology, application, and service FutureTech 2012. Volume 2. Dordrecht: Springer.
Doyle, S. (2010). Determining voice recognition accuracy in a voice recognition system. The Journal of the Acoustical Society of America J. Acoust. Soc. Am., 128(2), 64-89.