Convolutional Neural Network: A boon for deep facial recognition in Biometrics.Vishalakshi Rituraj1, Research Scholar-phD (CS), Magadh University, Bodhgaya.
Email id: [email protected] Krishna Singh2, Associate Prof., Mathematics Dept., A. N. College Patna.Abstract:-Today Biometric recognition systems are gaining much acceptance and lots of popularity due to its wide application area. They are considered to be more secure compared to the traditional password based methods.
Research is being done to improve the biometric security to tackle the risk and challenges from surroundings. Artificial Intelligence has played a significant role in biometric security. Convolutional Neural Network (CNN) belongs to AI family, has been designed to work a little like human brain but not exactly, handles the complexity and variations in facial images very effectively. This paper is going to focus on Artificial Intelligence, Machine Learning, Deep Learning and how a CNN carries out facial detection.Keywords:- Biometrics, Neural network, Learning, convolution, neurons, Pattern Recognition. 1) Introduction:-The increasing demand of technology in each and every field of our lives has raised the risk of data security in parallel. From the very ancient time, man is putting his best effort to get his things secured.
But today in this digital world, we are facing more problems due to impostors and other types of security hacks. Besides these, the curious human nature has always been trying to do something new and to cross the predefined boundaries. Intelligence is a by birth human quality but now a days, technology has made machines to think and behave like us to some extent. This concept of manmade intelligence created by rigorous use of complex mathematical operations and searching algorithms is known as Artificial Intelligence (AI). When we saw the AI used in Hollywood movie TERMINATOR, we didn’t even imagine the concept of such a smart machine that could handle different situations. But now, it seems impossible is going to be possible due to AI as it has opened the door of a completely new world of opportunities. Artificial intelligence is a branch of computer science aiming to make a computer, robot, or a software think intelligently, in the same manner the intelligent humans think and it has been proved very useful where traditional algorithmic solutions don’t work well.
We are using AI based applications everywhere in our day to day life, such as- spam filters in gmail account, plagiarism checker, Google’s intelligent prediction in web searching, suggestions on Facebook and Youtube and many more. The main purpose of designing AI system is to include the following areas:-PlanningLearningProblem SolvingPattern RecognitionSpeech/Facial RecognitionNatural language processingCreativity, and many more.Neural networks and deep learning, a branch of AI currently provide the best methods to solve many problems associated with the Biometric authentication. Biometrics is a noble technique for personal authentication either on the basis of physical attribute (fingerprint, iris, face, palm, hand, DNA etc.) or behavioral (Speech, signature, keystroke etc.). As we all know, our face is one of the wonderful creations of God and the unique diversities among all faces help us to differentiate one another. Facial recognition is the fastest growing field because a large no.
of applications is adopting it. Recently, Apple launched its face recognition system equipped iPhone X on 12 Sept 2017 and it is claimed that it can identify the face in dark or even when owner has different hairstyle or look as well. Apple says that the facial recognition cannot be spoofed by using a photograph or even a mask 1.(2) Application areas of Facial Recognition- Facial biometric recognition is being popular due to its wide range of applications and it can easily be deployed and integrated anywhere if there is modern high definition camera.
Some of the trending applications are-Many electronic devices are integrated with face biometric to eliminate the need of passwords and thus providing enhanced security and accessing method.Facebook’s automatic facial detection feature recognizes our friends’ faces with pretty good accuracy and starts suggestion based on it.Criminal identification has become simpler by better recognition of facial image through CCTV surveillance. It may minimize traffic rule breaking and road accidents.Some universities use facial recognition system as a tool to monitor the attendance of the students so that the management cannot be fooled by letting students to sign in behalf of others. ESG Management School in Paris is using facial recognition software in its online classes to make sure students aren’t slacking off. Using a software called Nestor, the webcam on a student’s computer will analyze eye movements and facial expressions to find out if he or she is paying attention during video lectures.
2In our paper, we will focus on the need of facial recognition and how deep learning and neural networks have been a backbone for this technology. 2) Machine Learning (ML) and Deep Learning (DL):- Machine learning is considered as subset of AI which uses statistical techniques and algorithms which make a machine capable of making decision or prediction by learning from the given data and adapt through experience. The process of learning begins with observations or data, such as examples, direct experience, or instruction, in order to look for patterns in data and make better decisions in the future based on the examples that we provide. The primary aim is to allow the computers learn automatically without human intervention or assistance and adjust actions accordingly 3. Deep learning is a subset of Machine learning where a machine has a higher level of recognition accuracy and aims to solve real world problems like image recognition, sound recognition, space exploration, weather forecasting and so many other automated applications. Here, the word ‘deep’ refers to the no. of layers in the network to accomplish a task.
Deep learning methods use neural network architectures, very much like neurons in human brain, introducing a concept of Artificial Neural Network (ANN). 3) Concept of Artificial Neural Network in problem solving:- Today, automated systems have made our lives too easy and have replaced man in some places. But when we talk about ‘intelligence’, man will always be superior to machines because of their god gifted nervous system which is composed of billions of neurons. These neurons are interconnected together and pass signals to one another which make the entire system to identify, classify and analyze things.
Getting inspiration from biological neural network, the concept of ANN came into existence. The inventor of the first neurocomputer, Dr. Robert Hecht-Nielsen, defines a neural network as – “…a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs.” 4 Figure1: A simple ANN structure.
5 3.1) Types of ANN: – (A) On the basis of topological arrangement, there are two types of ANN-a) A Feed-Forward Network :- In this type of ANN, data flow takes place in only one direction through different layers and none of the layers is fed with signal from background direction. This network does not have feedback loops as output of one layer becomes the input for other layers. Practically, in a Feed forward network, any prediction does not have to be affected with the previous predictions.Figure 2: A Feed-Forward Network 6b) Recurrent Neural Networks (RNN):- This type of neural network allows feedback loop by transmitting signals not only in one direction, instead data flow is carried out from backward direction too, sometimes also known as FeedBack ANN. In RNN, each neuron has its connection with others and how the flow of data is maintained, will be governed by its internal memory.
The decision taken by RNN gets affected by the decision made by the network at previous. It means, the current output of a RNN depends on both the previous output as well as the current input.Figure 3: Recurrent Neural Networks (RNN) 7(B) On the basis of layering, there are two types of ANN-(a) Single Layer Network- In this type of network, neurons on input layers are connected with the neurons present at the output layer and there is no layer in between these two layers.(b) Multi Layer Network- This type of ANN consists of more than one layer in between input and output layer which are called hidden layers. These hidden layers carry out computation by passing data from one layer to another. In this scheme, output from one layer becomes input for next layer and so on; finally output is obtained from output layer.
(4) Convolutional Neural Network (CNN):- A convolutional neural network (CNN) is a subset of deep learning and belongs to the category of multilayer, feed-forward artificial neural networks. One of the most promising areas where this technology is rapidly growing, is security. It has been very helpful in monitoring suspicious banking transactions, as well as in video surveillance systems or CCTV.Figure 4: A typical CNN architecture 8Besides input and output layers, CNN has many hidden layers in between which may be classified as-Convolutional Layer:- This layer performs the core operations of training and forms the basis of CNN.
Each layer has a single set of weights for all neurons and each neuron is responsible for processing a small part of the input space. Thus, the convolutional layer is just an image convolution of the previous layer, where the weights specify the convolution filter 9.Pooling Layer:- This layer also known as downsampling layer, is placed after the convolutional layer. Pooling layer is responsible for reducing the spatial size (Width x Height) of the Input Volume which will be passed to the next convolutional Layer. Fully Connected Layer:- This layer connects each neuron on previous layer with all the neurons present on the next layer.(5) Facial detection/Recognition using CNN:- A human brain sees multiple images in a day and is able to distinguish each one accurately without realizing how the processing is done. But, there is a different case with machines because they have to recognize an image on the basis of learning.
Facial detection is a method to identify a person or object based on their unique features and this process involves the detection and extraction of the face from the original image or video. After this, the face recognition takes place where different complex computer algorithms are used to recognize a face. Here, we will understand the entire process of face detection and recognition. A face detection system involves two phases:-(I) Enrollment Phase- Face Detection- In this phase, several pictures of the same person is captured to whom the system should recognize as “known” with different facial expressions and head positions.Feature Extraction- In this step, different feature measures are applied which can better describe a human face. There are different algorithms such as Principal Component Analysis (PCA), Haar Features, Local Binary Pattern (LBP) etc. available for the facial measurement. On the basis of these measurements, CNN is trained for learning in future.
Storing in Database- All the extracted features are stored in a database so that they can be used further in identification process.Face DetectionPre-processingFeature ExtractionFace RecognitionImageVerification/Identification(II) Recognition Phase-Figure 5: Architecture of Face Recognition System 10Face Detection- When an image is admitted for identification, It is checked that whether it matches with the captured and stored images from the database by using face detection algorithms. Pre-processing- Pre-processing is necessary to make an easier and smooth training phase. The collected face images or video frames need to be passed through Pre-processing phase to eliminate the noise, blur, shadows, lighting and other unwanted factors. The final smooth image obtained so, will be passed to the next feature extraction phase.
Feature Extraction- After Pre-processing phase, feature extraction is carried out by the CNN which was trained during Enrollment phase.Recognition- This is the last step where a suitable classifier such as Nearest Neighbor, Bayesian classifier, Euclidean Distance classifier etc., can be chosen. This classifier compares the feature vector stored in the database with the query feature vector and finally the best matched face image comes as a recognition output. 6) Conclusion:Biometric verification/authentication is going to be deployed everywhere from government to private organizations in coming days. In this paper, we studied the relation among AI, ML, DL, ANN and CNN. We have also demonstrated the way CNN carries facial detection with improved accuracy.
The field of AI has a wide spectrum and open for researchers. So, it aims to provide better result in biometric security in future.References”You can stymie the iPhone X Face ID – but it takes some work”, Anick Jesdanun, https://phys.org/news/2017-10-stymie-iphone-id-.
html”Entrepreneur India”, https://www.entrepreneur.com/slideshow/280493#2″What is Machine Learning? A definition” Luca Scagliarini, Marco Varone, http://www.expertsystem.com/machine-learning-definition/.
“Artificial Intelligence-Neural Networks”, https://www.tutorialspoint.com/artificial_intelligence/artificial_intelligence_neural_networks.htm. “Artificial neural network”, https://en.wikipedia.
org/wiki/Artificial_neural_network.”Artificial Intelligence-Neural Networks”, https://www.tutorialspoint.com/artificial_intelligence/artificial_intelligence_neural_networks.
htm.”Artificial Intelligence-Neural Networks”, https://www.tutorialspoint.com/artificial_intelligence/artificial_intelligence_neural_networks.htm.
“Convolutional neural network”, https://en.wikipedia.org/wiki/Convolutional_neural_network.”Convolutional Neural Networks”, http://andrew.gibiansky.com/blog/machine-learning/convolutional-neural-networks/.
“Face Recognition Using Neural Network: A Review”, Manisha M. Kasar, Debnath Bhattacharyya and Tai-hoon Kim, International Journal of Security and Its Applications, Vol. 10, No. 3 (2016), pp.81-100.