It is the Early 21st Century, Science has Deployed all sorts of Machinery and Technologies to make this world more productive and efficient and overall a better place to live. Science has several domains and one such is Machine Learning. Machine learning can be found almost everywhere and will be found in more things as the time passes, from your Youtube Video Recommendations, to the Ads you get while scrolling through your IG.
but less talked about application is the Biometric Devices. thus, which includes the attendance Biometrics install in many places, to your laptops face unlock and even sitting very next to you, i.e your Phone’s Fingerprint Scanner.!!
Let’s first learn what is a biometric, a very basic definition from Wikipedia suggests, “Biometrics are body measurements and calculations related to human characteristics. Biometrics authentication is in use in computer science as a form of identification and access control”.
It provides high security which is need in this electronically connected world and less chance of alteration in this electronically connect world. Biometrics traits can be iris pattern, retinal scans, fingerprints, voice and signature.
but fingerprint comes out on top in terms of uniqueness and acceptability and low cost. in fact in 2009, 28% of all the Biometrics deploy upon the fingerprints.
ML Techniques such as Artificial Neural Networks, Support Vector Machine and Genetic Algorithms play a vital role in fingerprint identification. The basic concept behind the solution is to train the machine to build the feature vector and train the machine how to process the vector according to some particular constraints.
Fingerprint Identification
Let’s first understand the structure of the fingerprint and how the machine divides the overall fingerprint structure in different layers to identify and learn the basic structure of it.
It is generally trained using a basic structural division of a fingerprint, (i) Global Structure, (ii) Low Level Structure, and (iii) Low Level Structure.
The first one represents the overall shape of the finger. thus, the second one represents the valleys and ridges format at local intersecting region, and the later. i.e the low level structure represents the sweat pores on the fingerprint skin.
Now let’s talk about the whole system which is done for Biometric scanning. the most widely used system is Automatic Fingerprint Identification System (AFIS) which has replaced human experts in fingerprint recognition as well as classification.
It begins with the (i) Enrollment Phase, which basically involves the registration phase where the individual identity( the fingerprint structure) is fed to the machine for it to learn and later identify, The second phase is called the identification phase, however, responsible for extracting the individual identity from the database according to the user claimed identity.
Machine Learning Techniques
these systems connect along with building fixable algorithms or techniques that their performance is automatically improved with experience (training). Machine learning system is first trained with training data. it is use to perform required operations according to its acquired experience.
The problem of machine learning techniques is related to their sensitivity to the training data and the training parameters as they may produce different results by changing the training data set. However machine learning includes many techniques such as ANN, SVM, GA, Bayesian Training and Probabilistic Models, we will stress only on the implementation of the first three techniques on fingerprint identification.
Artificial Neural Network
This is the most widely use algorithm of the Machine Learning System. The quality of the acquired fingerprint is assure before its feature extraction. Xie and Qi designed a supervised back propagation neural network that uses grayscale fingerprint image for continuous image quality estimation. But this process turns out to be expensive as the images are to be divide into blocks.
But Zhu and others used the Neural Networks for estimation of the quality of the fingerprint images using fingerprint ridge orientation. Further work found a set of new features for contactless images and designe neural networks to extract the complex features for future fingerprint matching. but this feature had a drawback than the comparatively larger processing time.
Fingerprint classification is of utmost importance when it comes to identification, Prof. Sarbadhikari proposed a two stage fingerprint classifier. In the second stage, the Multi-Layer Perceptron feed forward neural network was use to classify the directional Fourier Image. This classification achieved an accuracy of 84%.
the work was further ahead by Mohamed and Nyongesa by the usage of Neural Network in classification. as it has the ability to work as an adaptive filter and produce reliable results. They used five features mainly, core points, number of delta points, directional image, core point direction and the position of the delta point and was able to classify with an accuracy of 85.0%.
Kumar and Vikram used multidimensional ANN (MDANN) for fingerprint matching using minutiae points. The algorithm achieved more than 97.37% of the recognition rate. In general, Artificial Neural Network is effective when it comes to classification, extraction and matching of the fingerprints.
Support Vector Machine
Support Vector Machine is a training algorithm for linear classification, regression, principal component analysis and for non-linear classifications as well. The idea behind the support vector machine is to maximise the margin between the training patterns and the decision boundary.
Prof. Liu and others used the support vector machine technique with five features to determine the quality of the image and it was successful in classifying the image quality into low, medium and high quality with a very great accuracy of 96.03%, but due to a very long processing during feature extraction his technique was a little inefficient.
But later Prof. Zhao et al stepped in a similar manner and introduce a technique which consider to be robust, this algorithm divides the image into small pixel blocks, and five features been use for constructing the feature vector. The features are gray mean, gray variance, contrast, coherence and the main energy ratio and the accuracy achieved by this technique close to 94.5%. in classification, but the processing speed is much higher than the former technique.
Genetic Algorithm | Fingerprint Identification
Genetic Algorithms are very promising machine learning techniques for solving fingerprint and biometric related problems. Mao and others were the first few who succeeded in using genetic algorithms for singular point extraction. They present a new definition for core point location and orientation which is use like a fitness function for the genetic algorithm. The challenging part with the implementation is that the processing time becomes higher with increased accuracy.
Prof. Tan later implemented his technique but in this case the accuracy was promising but it took 8 seconds to perfectly match the imposter, therefore an optimisation required. On further development, Tan implemented a classification algorithm based on some new features. In the proposed approach, they tried to find unconventional primitives from the orientation images with the help of genetic programming technique.
Then, a Bayesian Classifier in use for conducting the actual classification process and on testing this technique perform well where 2000 images in use as a training set and 2000 more in use for the evaluation process. at the end it successfully classified the 2000 fingerprint images in 5 classes with an accuracy of 93.6%.
Conclusion and Future Work
This article introduced a precise survey on the usage of machine learning techniques for biometric problems and fingerprint identification problems. The paper has focused on three important techniques which are ANN, SVM and GA, and how they have been implemented by different experts,
The review confirms the superiority of using ML for tackling such problems on different fingerprint identification problems. The future work will target towards the development of one of machine learning techniques that will be helpful in tackling some pending fingerprint challenges such as processing time reduction and identification accuracy enhancement. Moreover, other biometrics traits like palm prints and iris patterns can be looked upon for Biometric identification.
written by: Anurag Mukherjee
reviewed by: shivani yadav
If you are Interested In Machine Learning You Can Check Machine Learning Internship Program
Also Check Other Technical And Non Technical Internship Programs
