You are on page 1of 7

Face Recognition Systems

Prateek Murli

Face Recognition Systems


Prateek Murli November 18, 2012
Abstract Face recognition systems are determental in various spans of organisational work and is being continually developed for various platforms to strengthen the current security environment.These technologies are gaining more importance because of the potential it holds in enhancing the current security system capabilities.Only in recent years reliable automatic face recognition technology has become a realistic target of biometrics.In this paper, we discuss about technology, various applications, standards and challenges that are faced in this area.

Introduction

In current scenario the need to maintain the security of information or physical property is becoming both increasingly important and increasingly dicult.The security threats and challenges posed in the current world demands a reliable method to identify and authenticate people. Biometrics is being widely used and adopted these days as a reliable method of authentication as compared to traditional methods of passwords and tokens.Biometric authentication relies on specic biological characteristics such as ngerprints, iris, retina etc. Facial recognition oers many advantage over other biometric identication and authentication techniques as facial recognition systems have the following advantages - Non intrusive - Hygenic as no physical contact required

ric methods such as ngerprint recognition, palm prints, iris recognition etc. In a way it can improve the interaction between machines and humans in an eective manner. Besides there are a large number of commercial, security and forensic applications which require use of face recognition technologies. These applications include automated crowd surveillance, access control, face reconstruction, design of human computer interface, multimedia communication etc. Face recognition systems are distingushed by the technology used by input medium as it determines the quality of image being produced for e.g 3-Dimensional data is invariant to head pose while infra red face imaging is practically invariant to lighting.

Face As a Biometric

- Easy to use as only need to face a camera Face recognition has the advantage of major acceptance in the fact that everyone has a face - Useful for surveillance and everyone displays the face. Face which contains many distinctive features serves as paFace recognition systems provides a rameters for nodal points which is determental friendly way of recognising persons by captur- in recognising and identifying patterns. Each ing the persons face using a camera attached face has approximately 80 nodal points.Face to a computer system. It is less cumbersome recognition as a biometric has many advanfor the end users as compared to other biomet- tages in certain circumstances and can be inapPage 1 of 7

Face Recognition Systems

Prateek Murli

propriate choice in many application depending on the domain of application. Face recognition also has an advantage that devices are cheap and are becoming a commodity(though this is not true for non-visible wavelength devices and some of the more sophisticated face recognition technologies based on 3-D data.

2. Detect: It involves identifying and highlighting the captured image from rest of the image background. Once it detects a face,system determines the heads position,size and pose. 3.Align:Alignment is necessary in case when face may not be perpendicular to the camera. Image alignment is required to be determined before recognition 4.Extract: It involves capturing various facial pattern and features so as to create a facial template or face print for the purpose of matching and identication.The parameters that can be used for creating template can be: 1. Distance between the eyes. 2. Length of jaw line. 3. Skin tone 4. Facial geometrical mesh 5. Width of nose.etc 5. Match: It involves comparing the 2D or 3D facial template acquired with the facial databse. This is often carried out by scoring process in which various algorithms are run to determine the probablity of match being accurate. 6. Report: Reporting involves returning one or more facial matches obtained on basis of user preference and score. In real time situations need may arise to to recognise faces from live video les in case of surveillance etc. In such cases system captures the individual static frames from the video and does comparison by creating facial templates obtained from those frames.

Face Recognition Process

Face recognition process is carried out by means of comparison of features of face which dont change over a period of time. The parameters of comparison can be skin tone against the information that are stored in facial database. The comparison are performed by various algorithms but basic steps remain the same. It involves the following: recognition.png

Generic Framework

Face Image Detector: It locates and isolates 1.Acquire: It involves capturing the image human faces or probe image from a complex containing the face.In case of 2D facial recogni- picture back ground. tion digital camera is reqiuired. For 3D facial recognition additional range camera would be Face Recogniser: It determines the idenrequired for obtaining the depth coordinates. tity of the person which along with Face image detector using feature extractor.It classies the Page 2 of 7

Face Recognition Systems

Prateek Murli

feature vector assuming to be a face as a per- or RGB images.Algorithms tend to calculate sons face already registered in the database. depth details from multiple 2D images. Since it involves calculations, resultant data is an Feature Extractor: It transforms the interpolation and prone to errors. pixel of facial image into useful vector representation 2.Actual 3D facial surface: In order to capture actual 3D facial surface data we require devices such as depth sensors, range cameras, time of ight cameras, or infrared cameras. These camera capture depth image details by emitting IR rays and capturing the reected rays. The time taken by the ray to travel back helps in calculating the distance of a point on the surface from the cameras.The pixels in these images contain the distance information, that is, the distance of the camera from the surface that reected the emitted ray.The camera may need calibration and verication for accuracy. Once the raw data is captured, it must be converted into 3D point cloud with help of algorithms and constants depending upon the time of ight camera used. Various open sources are also available to compute the 3D point cloud. The number of points in the point cloud varies from device to device. Pattern Recognizer: It performs the How It Works: process of searching and matching the best incoming face image.It categorizes images into face images and non face images.

3D Face recognition: Following are ways to get 3D facial surface data: 1.Deriving from RGB images: It is pos-The device module acquire initial 3D fasible to generate 3D facial surface from 2D cial data and transfers it to the processor. Page 3 of 7

Face Recognition Systems

Prateek Murli

-The Data Processing Module Reconstructs 2.From the given training set M Eigen faces the 3D surface for further recognition. are calculated known as face space. -The feature Extraction Module builds the 3.Projecting each individual image on to feature vector(biometric template) based on face space, M dimensional weight space for 3D surface reconstructed for matching process. each individual is obtained The Matching Engine Module provides a 4 Closeness of input image to the face imcomparison of acquired and previously enrolled age is observed to determine whether input biometric template. image is a face or not. If it is suciently close to face space, it is recognised as a face.

5.Depending on the weight pattern obtained from projection, image is classied as a Facial recognition is basically comparison and known face or unknown face. images under comparison are of two types. The rst type is RGB images such as JPEG and PNG, such type is used in 2-Dimensional Face recognition systems. The second type is the one consisting of depth information captured with the means of range cameras. RGB images are obtained from normal digital cameras. Some of the algorithms that work with RGB images are: 1.Principal Component Analysis(PCA):PCA is also known as use of eigenfaces.It was developed by kirby and Sivrich in 1988. PCA requires the probe and gallery images to be of same size and be normalized to line up the eyes and mouth of subjects in the image. In PCA approach the dimension of data is reduced by data compression methods which helps in highlighting the most eective dimension of facial characteristics. The dimension reduction helps in removing unwanted details and accurately restructures the facial patterns into uncorrelated components known as eigen faces. Each face image is stored in 1 dimensional array. Distance between various facial features are measured by comparing probe and gallery images. PCA approach requires frontal face to be presented each time else the image results in poor performance. The main advantage of this technique is that it can reduce the data needed to identify the individual to 1/1000th of the data presented. The key implementation steps are: 1. Initial set of test face images are obtained,known as training set.

Algorithms

Linear Discriminate Analaysis: LDA is an approach for identifying unknown classes by comparing it with known classes.This approach helps in maximizing between-class variance and minimize within-class variance.It considers the discriminating capability of facial features. For example facial region having the eyes and the region having nose and

Page 4 of 7

Face Recognition Systems

Prateek Murli

lips have greater discrimination capapblity.The main dierence between LDA and PCA is that PCA does more of feature classication and LDA does data classication. In PCA, the shape and location of the original data sets changes when transformed to a dierent space whereas LDA doesnt change the location but only tries to provide more class separability and draw a decision region between the given classes. LDA approach faces sample size problem when there are limited number of available samples compared to dimensionality of the sample space. EBGM: Elastic Bunch Graph Matching. EBGM approach is based on many factors such as illuminination(outdoor or indoor lightongs), pose (standing straight or leaning over), expression (smile or frown) etc. The face is projected onto an elastic grid. The elastic grid architecture is depicted with help of Gabor jet node structure which helps in detecting shapes and extract features using image processing. This approach faces challenge of accurate landmark localization,which can sometimes be achieved by combining PCA and LDA methods.

is called as type 2 error rate. FAR=NFA/NIIA Where, FAR = false acceptance rate NFA = number of false acceptance NIIA = number of imposter identication attempts. 2-Dimensional and 3-D sytems have found to have FAR of around 0.001 False Rejection Rates(FRR) The probability that a system will fail to identify a genuine user. FRR=NFR/NEIA Where, FRR= False rejection rate NFR= Number of false rejection rates NEIA=Number of genuine registered entry identication attempt. 2D Systems have FRR in range betweem 0.010-0.017 3D sysytems have FRR in range between 0.005-0.015 Response time:The time period required by a biometric system to return a decision on identication of a sample Threshold: The acceptance or rejection of a data is dependent on the match score falling above or below the threshold. The threshold is adjustable so tha the system can be made more or less strict; depending on the requirements of any given application. Enrollment time:The time period a person must spend to have his/her facial reference template successfully created

2D Vs 3D Facial recognition System

1.Facial expression Variation : In 2D facial recognition becomes dicult if the subject is 6 Performance Evaluation non cooperative whereas 3D recognition works more reliably as dierent facial expressions of False Acceptance Rate(FAR): The proba- a person are treated as isometric surfaces. bility that a system will incorrectly identify an individual or will fail to reject an imposter.It 2.Facial Features: 2D recognition is Page 5 of 7

Face Recognition Systems

Prateek Murli

based on measurements such as distance between eyes, width of nose etc, Whereas 3D recognition is based on contours of the nose, chin, eye sockets etc for which mesh of the facial surface is used.

4.Banking: -Customer Identication: Account number, details of cheques and other credentials can be easily veried when the person approaches the counter. -ATM Access: Facial recognition along with secret pass key can be used for accessing an 3.Face Orientation: Minor variations account. upto 20 degrees can be accomodated in 2D whereas faces upto 90 degrees of orientation 5. Public Surveillance: can be recognised in 3D. -Locate Missing people -Identify Criminals and Terrorists. 4.Lighting Variations: In 2D, face needs -Identify Changes, how the person can possito be well illuminated else poor lightin can lead bly look after years. to low performance. In 3D, Infrared lights with range camera can be used in low light condi6.Time and Attendance System: It tions. can be implemented as abiometric time clock sytem to keep track of attendance and working 5.Reliability: 3D Recognition system are hours of employees.This can prevent buddy more reliable than 2D technologoes which are punching where colleagues register the attenonly reliable in controlled conditions. dance of the employees.

Operational Scope

Challenges

This section discusses futuristic and present scenario of face recognition. 1Access Control:It can be used to provide access to: - A computer or a smartphone. - At Physical security check point of a secure facility or for a territory. - Verication at an examination center. 2.Retail Industry: -Identify Customers: To provide personalized shopping experience. -Loyalty Program: As an alternative for loyalty cards -Store Demographics: Identifying the number of people,gender and age can help the store manager their merchandising and setup eciently.

2D Face Recognition Systems: Even minor variations can lead to low resolution results of degraded quality. The inclination of the persons face with respect to the camera or the distance from the camera usually has some variations, which must be considered during recognition.In addition dierences in lighting conditions and background images along with face can pose diculties in recognizing the face. Variation due to make up, moustaches, hair style etc can also aect the performance of system. 3D Face recognition System: While using 3D algorithms, the processing capability is a challenge as lot of computation power is needed to generate geometrical mesh of images.The cost of devices is also determental in adoption of 3D face technology.

3.Travel: -Immigration Checkpoint: Can be used for verication and security measures. 10 Conclusion -Boarding Pass Issue: Frequent Fliers can be automatically identied at the counter and be Face recognition system and its subsequent issued the boarding pass. technology are an area of huge scope and promises to enhance the current capability Page 6 of 7

Face Recognition Systems

Prateek Murli

of security systems.However, it also has serious shortcomings and key challenges considering which it cannot be completely reliable to be used as an authentication technique.So it might need to be combined with other technologies such as Iris or retina scan to improve reliability.With improvements in technology it is only a matter of time before the incidence of false identication of subjects is reduced to an acceptable level to make facial recognition systems more useful and helpful.

FaceRecognitionLiteratureSurvey.pdf

2 Face Recognition and its applications by Andrew W. Senior and Ruud M.Bollehttp://www.andrewsenior.com/papers/SeniorB02

3.Biometrics: A look at Facial Recognition by JOhn D Woodward http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA41452 4.Wikipedia,http://en.wikipedia.org

5.Face Recognition Algorithms by Proyecto Fin de Carrera sumbitted June 16,2010 11 References ,http://www.ehu.es/ccwintco/uploads/e/eb/PFC1.Face Recognition: A literature Sur- IonMarques.pdf vey by W.Zhao, R.Chellappa,P.J Phillips. http://www.cs.ucf.edu/ dcm/Teaching/COT4810- 6.FRVT 2006 and ICE 2006 Large-Scale Spring2011/Literature/DiegoVelasquezResulthttp://www.face-rec.org/vendors/frvt2006andice20

Page 7 of 7

You might also like