PATTERN RECOGNITION

Pattern recognition is the research area that studies the operation and design of systems that recognize patterns in data. It encloses sub-disciplines like discrete analysis, feature extraction, error estimation and cluster analysis also known statistical pattern recognition, grammatical inference and parsing, together known as syntactical pattern recognition. Important application areas are image analysis, character recognition, speech analysis, man and machine diagnostics, person identification and industrial inspection. Pattern recognition aims to classify data based on either a priori knowledge or on statistical information extracted from the patterns. The patterns to be classified are usually groups of measurements or observations, defining points in an appropriate multidimensional space.

3.1

Face Recognition

Face recognition in general and the recognition of moving people in natural scenes in particular require a set of visual tasks to be performed robustly. These include: 1. Acquisition: the detection and tracking of face-like image patches in a dynamic scene 2. Normalisation: the segmentation, alignment and normalisation of the face images 3. Recognition: the representation and modelling of face images as identities and the association of novel face images with known models.

3.1.1 Basic Technologies

3.1.1.1 Eigenface Technology A large number of pictures of faces are collected in a database. Then, combining all the pictures as shown below, makes a set of eigenfaces, a two-dimensional face-like
15

arrangements of light and dark areas and looking at what is common to groups of individuals and where they differ most. Eigenfaces work as component faces. Just as a color can be created by mixing primary colors, a facial image can be built by adding together, with different intensities of light, a relatively small number of eigenfaces about 100 is enough to identify most persons but just for one pose and one lighting condition. To identify a face, the program compares its eigenface characteristics which are encoded into numbers called a template with those in the database, selecting the faces whose templates match the target most closely.

3.1.1.2 Local Feature Analysis Local feature analysis is derived from the eigenface method but overcomes some of its problems by not being sensitive to deformations in the face and changes in poses and lighting. Local feature analysis considers individual features instead of relying on only a global representation of the face. The system selects a series of blocks that best define an individual face. These features are the building blocks from which all facial images can be constructed. The procedure starts by collecting a database of photographs and extracting eigenfaces from them. Applying local feature analysis, the system selects the subset of building blocks, or features, in each face that differ most from other faces. Any given face can be identified with as few as 32 to 50 of those blocks. The most characteristic points as shown to the right are the nose, eyebrows, mouth and the areas where the curvature of the
16

bones changes. The patterns have to be elastic to describe possible movements or changes of expression. The computer knows that those points, like the branches of a tree on a windy day, can move slightly across the face in combination with the others without losing the basic structure that defines that face. To determine someone's identity, (a) the computer takes an image of that person and (b) determines the pattern of points that make that individual differ most from other people. Then the system starts creating patterns, (c) either randomly or (d) based on the average eigenface. For each selection, (e) the computer constructs a face image and compares it with the target face to be identified. (f) New patterns are created until (g) a facial image that matches with the target can be constructed. When a match is found, the computer looks in its database for a matching pattern of a real person (h), as shown below:

3.1.1.3 Neural Network As face recognition has evolved from eigenfaces to local feature analysis, three performance issues have remained which is locating a face in images with complex backgrounds, accommodating variable lighting conditions and improving recognition accuracy. A face image or video stream includes one or more faces and a background. The background may
17

contain any number of textures, objects and other people, whether moving or not. Among all the possible objects in an image, the first step, face isolation is to isolate a single face from the image if a face exists. This step is achieved with a neural network that is trained to recognize what a face looks like across different hair styles, lighting conditions, skin color and gender as well as across varying face sizes, locations and orientations. This neural network training comes from hundreds of thousands of example faces. For video streams, a face can be continuously tracked by updating the face isolation process across the video frames. Once a face is isolated and located, features from the entire face are extracted as visual contrast elements such as the eyes, side of the nose, mouth, eyebrows, cheek line and others as shown below.

The features are quantified, normalized and compressed into a template code. Then the program matches the known reference template against the live face template and produces a match score. When a match scores, between the live face template and the stored reference template, is greater than a preset threshold, then the two faces are deemed to come from the same person. Otherwise they are deemed to come from different people. The matching process involves a neural network in which features from both faces vote on whether the faces are similar or different, as shown above. In addition, each vote is weighted by a weight that is adjusted in a learning process. Learning is done by a trial and error method. Initially, all the weights are the same. Then the neural network is exposed to trials, each with a pair of faces. If the neural network produces the right answer in the vote on whether the faces were the same or not, no learning takes place. However, if the neural network produces the wrong answer, the weights for the features are adjusted to correct the answer. By repeating this learning process over many examples of faces, the neural network automatically adjusts all the feature weights to produce the most optimal differentiation between similar and different pairs of faces. Because the neural network learns from experience, it does a better job of
18

accommodating varying lighting conditions and improves accuracy over any other method. After the neural network has been trained, it is ready for use in verification and identification applications.

3.2

Software: FaceIt

Identix, a company based in Minnesota, is one of many developers of facial recognition technology. Its software FaceIt, can pick someone's face out of a crowd, extract the face from the rest of the scene and compare it to a database of stored images. In order for this software to work, it has to know how to differentiate between a basic face and the rest of the background. Every face has numerous, distinguishable landmarks, the different peaks and valleys that make up facial features. FaceIt defines these landmarks as nodal points. Each human face has approximately 80 nodal points.

Some of these measured by the software are:

Distance between the eyes Width of the nose Depth of the eye sockets The shape of the cheekbones The length of the jaw line

19

These nodal points are measured creating a numerical code called a face print representing the face in the database.

The technology for it is based on neural computing and combines the advantages of elastic and neural networks.

20

3.2.1 FaceIts 3D Techniques Capturing a real-time 3D image of a person's facial surface, 3D facial recognition uses distinctive features of the face where rigid tissue and bone is most apparent, such as the curves of the eye socket, nose and chin to identify the subject. These areas are all unique and don't change over time thus increasing the effectiveness and accuracy. Using depth and an axis of measurement that is not affected by lighting, 3D facial recognition can even be used in darkness and has the ability to recognize a subject at different view angles with the potential to recognize up to 90 degrees, a face in profile. Using the 3D software, the system goes through a series of steps to verify the identity of an individual.

21

1. Detection Acquiring an image can be accomplished by digitally scanning an existing photograph (2D) or by using a video image to acquire a live picture of a subject (3D) with a camera. 2. Alignment Once it detects a face, the system determines the head's position, size and pose. As stated earlier, the subject has the potential to be recognized up to 90 degrees, while with 2D, the head must be turned at least 35 degrees toward the camera. 3. Measurement The system then measures the curves of the face on a sub-millimetre or microwave scale and creates a template. 4. Representation The system translates the template into a unique code. This coding gives each template a set of numbers to represent the features on a subject's face. 5. Matching If the image is 3D and the database contains 3D images, then matching will take place without any changes being made to the image. However, there is a challenge currently facing databases that are still in 2D images. 3D provides a live, moving variable subject being compared to a flat, stable image. New technology is addressing this challenge. When a 3D image is taken, different points are identified. For example, the outside of the eye, the inside of the eye and the tip of the nose will be pulled out and measured. Once those measurements are in place, an algorithm will be applied to the image to convert it to a 2D image. After conversion, the software will then compare the image with the 2D images in the database to find a potential match. 6. Verification or Identification In verification, an image is matched to only one image in the database. For example, an image taken of a subject may be matched to an image in the Department of Motor Vehicles database to verify the subject is who he says he is. If identification is the goal, then the image is compared to all images in the database resulting in a score for each potential match. In this

22

instance, that image can be taken and compared to a database of mug shots to identify who the subject is.

3.2.2 FaceIts Surface Texture Analysis The development of FaceIt uses skin biometrics, the uniqueness of skin texture to yield even more accurate results. A picture is taken of a patch of skin, called a skin print. That patch is then broken up into smaller blocks. Using algorithms to turn the patch into a mathematical, measurable space, the system will then distinguish any lines, pores and the actual skin texture. It can identify differences between identical twins, which is not yet possible using facial recognition software alone. According to Identix, by combining facial recognition with surface texture analysis, accurate identification can increase by 20 to 25 percent.

23

FaceIt currently uses three different templates to confirm or identify the subject: vector, local feature analysis and surface texture analysis.

The vector template is very small and is used for rapid searching over the entire database primarily for one-to-many searching.

The local feature analysis (LFA) template performs a secondary search of ordered matches following the vector template.

The surface texture analysis (STA) is the largest of the three. It performs a final pass after the LFA template search, relying on the skin features in the image, which contains the most detailed information.

The surface texture analysis (STA) algorithm operates on the top percentage of results as determined by the local feature analysis. STA creates a skin print and performs either a 1:1 or 1:N match depending on whether for verification or identification.

24

3.3

Pros Ideal for high traffic areas open to the general public such as airports, railway stations, casinos, cash points, stadiums, public transportations, financial institutions and government offices. Hard to fool- It works by comparing facial landmarks which are specific proportions and angles of defined facial features and these cannot easily be concealed by beards, eyeglasses or makeup. No need for a picture ID, bankcard or personal identification number (PIN) to verify a customer's identity. This way business can prevent fraud from occurring. Passive process where the subject doesn't have to be aware of the capture or comparison process. Can use in situations where only facial photos are available. Can use human backup to compare faces as no special training is required compared to fingerprints, hand geometry. Social acceptability is high. Hardware consists of ordinary video camera, inexpensive.

3.4

Cons Face recognition doesnt work properly when: Significant glare on eyeglasses or wearing sunglasses Long hair obscuring the central part of the face Poor lighting that would cause the face to be over- or under-exposed Lack of resolution

25

Reference 1. http://www.sans.org/rr/whitepapers/authentication/132.php 2. http://www.computer.org/itpro/homepage/jan_feb/security3.htm 3. http://www.rycom.ca/solutions/pdfs/iridian/iris_whtppr.pdf 4. http://www.cl.cam.ac.uk/users /jgd1000/history.html 5. http://www.stanford.edu// 6. http://cwis.kub.nl/~frw/people/kraling/content/biomet.htm 7. http://www.biometrics.org 8. http://www.wikipedia/fear 9. http://www.popsci.com 10. http://www.aihorizon.com/essays/generalai/supervised_unsupervised_machine_learning 11. http://www.doc.ic.ac.uk/~gu1/Doc/DT.pdf 12. http://www.moddb.com/tutorials/the-basics-of-ai-in-half-life-2 13. Y. Adini, Y. Moses, and S. Ullman, Face recognition: the problem of compensating for changes in illumination direction, IEEE Trans. Pattern Anal. Machine Intell., vol. 19, pp. 721732, July 1997. 14. Alison Cawsey. (1998). The essence of Artificial Intelligence. Hertfordshire:Prentice Hall Europe. 15. R. Brunelli and T. Poggio, Face Recognition: Features versus Templates. IEEE Trans.Pattern Analysis and Machine Intelligence, Vol. 15, No. 10, October 1993. 16. G. Galicia and A. Zakhor, Recovery of Human Facial Features from Video Sequences.Proc. IEEE Int. Conf. Image Processing, Washington, DC, October 1995.

26