Face Recognition Technology - Research Paper Example

Face Recognition Technology The Face Recognition technology in today’s world of network is getting increasing importance due to the growing security risks and industrialization of hackers. The fundamental flaw in today’s network conventional access control system is that the access is granted on the basis of ‘what we have’ such as passwords, keys, PIN or ID cards, etc. instead of ‘who we are’ like face, eyes, etc. According to news from Reuters in 1999, crooks from cyber world took huge advantage of this flaw and caused losses worth well over US$100 million. (Lin, 2000) Availability of ‘biometrics’ technology provides controls for verifying true identity of an individual. These controls are automated processes that recognize physiological characteristics such as fingerprints, face, eyes, DNA etc. of a living person which are not easy to forge as they are attributes of an individual gifted by nature. There are also automated processes that recognize individual behaviors such as handwriting style, key stroke patterns etc. (Lin, 2000) Physiological controls are more stable when compared with behavioral controls. The main reason is that the features of physiological controls are non-alterable unless some serious injury is inflicted on a living being. On the other hand the patterns of behavior controls fluctuate with the mood and activities of an individual. In real-life, it is found that verification of physiological attributes is although very accurate, yet it is far more intrusive than the behavior attributes (Lin, 2000) One of the few biometrics controls that have the merits of both low intrusiveness and high accuracy is the Face Recognition technology. Researches in the field of image processing, security and psychology were attracted towards the concepts of computer vision which led to the designing of face recognition technology. (Lin, 2000) Digital Images The real-world image has only size in inches or centimeters. The capturing device such as camera or scanners uses digitization process through which it stores the number of pixels that contains in an image. It is called Resolution which is of two types; Spatial Resolution and Colour Resolution (JISC Digital Media, 2006) The capturing device in Spatial Resolution is concerned with the frequency at which samples are taken from the real-world object or art-work. Frequency is mostly expressed as samples per inch (spi) when scanning and pixels per inch (ppi) when processing the digital image. The resolution to use for capturing an image is dependent mostly on its ‘end-use’. The digital size of the picture is dependent on its resolution; higher the resolution, larger is the size of graphic file and lower the resolution smaller is the size of graphic file (JISC Digital Media, 2006) The capturing device in Colour Resolution is concerned with the quantity of colours. If the image is a colour picture it saves bit depth and if the image is gray scale it saves brightness or shades of grey. The digital size of file depends on the size of the binary string used to describe colours for example 16 bits, 32 bits, etc (JISC Digital Media, 2006) The quality and accuracy of an image as well as its digital size depends on the spatial and colour resolutions chosen for capturing and producing the image. (JISC Digital Media, 2006) Face Recognition Technology in Digital Images In computer vision the face recognition problem is defined as identification or verification of one or more faces of humans in a still image or a scene of video by using a stored database of faces. There are a large number of algorithms as shown in Figure 1 developed for Face Recognition technology (Face-rec.org, 2008) Figure 1: Algorithms for Face Recognition in Still Images (Face-rec.org, 2008) The core technique used in face recognition technology is two standard biometric measures; Face Rejection Rate (FRR) and False Acceptance Rate (FAR). FRR is referred as Type I Error and FAR is referred as Type II Error and both measurements are inversely proportional. These two measures are called Performance Evaluation Metrics (Lin, 2000) Face Recognition Technology recognizes face from outdoor and non-frontal images at low false accept/alarm rates. It also recognizes males easier than females and its performance is affected by the demographic factors. To understand performance of the technology better statistical methods are developed with models for identification performance prediction on very large galleries. Face recognition performance is integrated with the sequence of development and changes with effect on covariate performance by system training and algorithm. (Face-rec.org, 2008) The task of face recognition is achieved when a good sensor device with proper pattern matching or feature extraction scheme algorithm is used (Lin, 2000). Face Recognition Technology in Videos and CCTV Videos The task of recognizing humans from videos is difficult because images in their frames are of low quality and the images of faces are small. In last two years many research on face recognition from videos was done which brought great improvement in the technology (Lin, 2000) There is a generic framework as illustrated in figure 2 for face recognition algorithm which has functional modules. ‘A face image detector’ is the first functional module that detects location of human faces from a normal size picture against complex or simple background. ‘Face recognizer’ is the second functional module that finds the person from pictures gallery (Lin, 2000) Both functional modules have same framework of feature extractor and pattern recognizer. Feature extractor makes a useful vector representation of the facial image’s pixels through transformation process. Pattern recognizer gets the best match for the image by searching the database (Lin, 2000) The difference in the task of two functional modules is that in the first module ‘face image detector’ the pattern recognizer classifies features of each vector to either ‘face’ images or ‘non-face’images. In the second module ‘face recognizer’ the pattern recognizer describes each ‘face’ images with the name registered on the database such as ‘Jane’s face’ etc (Lin, 2000). Figure 2: Generic framework for Face Recognition System (Lin, 2000) Values collected in Face Recognition Technology Feature Extraction models face image in to a vector space by creating a two dimensional array. Pixels are transformed in to vectors based on some functions. The vector space is created as (Lin, 2000):- Where f1(x), f2(x).... are linear or nonlinear functionals. Pattern Recognition is a much more complicated to model as there are six major variations in facial images(Lin, 2000):- 1. Noise and distortion of camera 2. Background being complex 3. Illuminated surroundings 4. Occlusion, scaling, rotation and translation 5. Expression of face 6. Hair style and makeup There are number of tools and algorithms developed for modeling each one of the above variations of facial images and Figure 1 lists most of them. One of the most popular algorithms is the eigenface representation of facial variations. (Petland, 1994) states the example of implementation of this algorithm in Petland’s Photobook. A facial image of size 128x128 pixels totaling 16,384 pixels is transformed into a vector of only 40 eigenfaces using only 80 bytes memory. The database contains 7562 facial images of 3000 different persons. Test of 200 facial images recognized 95% of the faces from the database (Lin, 2000) Common Applications using Face Recognition Application domain for face recognition is in the development stage but with the changing scenario the opportunities of deployment are enormous. The main area where it is deployed is for access control, identification systems, surveillance and pervasive computing (Senior & Bolle) Access control applications match the face of person physically present with the facial image database and allow access only if the face is recognized. One example is the Visionics’ face-based screen lock bundled with PC cameras. Another example is cheque-cashing kiosk of Mr. Payroll’s which operate without any human supervision. It is also used for control of physical access for example at Cognitec’s FaceVACS, and Miros’ TrueFace (Senior & Bolle) Identification systems use face recognition for welfare benefits at two US states (Massachusetts and Connecticut). Similarly surveillance applications use face recognition from videos although they are still in the development stage. In the field of pervasive computing where many devices equipped with sensors use the technology in equipments and automobiles, as strict security measures for classified and critical places such as nuclear power plants and weapons storages (Senior & Bolle) References Face-rec.org. (2008). Algorithms. In Face Recognition Home Page Retrieved from http://www.face-rec.org/algorithms/ Face-rec.org. (2008). GENERAL INFO. In Face Recognition Home Page Retrieved from http://www.face-rec.org/algorithms/ JISC Digital Media. (2006). The digital still image. In Digitizing analogue media Retrieved from http://www.jiscdigitalmedia.ac.uk/stillimages/advice/the-digital-still-image Lin, S. H. (2000). An introduction to face recognition technology. Informing Science Special Issue on Multimedia Informing Technologies - Part 2, 3(1), Retrieved from http://inform.nu/Articles/Vol3/v3n1p01-07.pdf Senior, A. W., & Bolle, R. M. (n.d.). Face recognition and its applications. In Chapter 4 NY, USA: IBM T.J. Watson Research Center. Retrieved from http://www.andrewsenior.com/papers/SeniorB02FaceChap.pdf Read More