- Date:02 Jan 2020
- Views:42
- Downloads:0
- Pages:37
- Size:1.18 MB

Transcription:

and content based image database management A number of commercial face recognition systems. have been deployed such as Cognitec 5 Eyematic 6 Viisage 7 and Identix 8. Facial scan is an effective biometric attribute indicator Different biometric indicators are suited. for different kinds of identification applications due to their variations in intrusiveness accuracy. cost and ease of sensing 9 see Fig 1 a Among the six biometric indicators considered in 10. facial features scored the highest compatibility shown in Fig 1 b in a machine readable travel. documents MRTD system based on a number of evaluation factors 10. Figure 1 Comparison of various biometric features a based on zephyr analysis 9 b based on. MRTD compatibility 10, Global 2002 industry revenues of 601million are expected to reach 4 04billion by 2007 9 driven. by large scale public sector biometric deployments the emergence of transactional revenue models. and the adoption of standardized biometric infrastructures and data formats Among emerging. biometric technologies facial recognition and middleware are projected to reach 200million and. 215million respectively in annual revenues in 2005. Face recognition scenarios can be classified into two types i face verification or authentication. and ii face identification or recognition In the Face Recognition Vendor Test FRVT 2002. 11 which was conducted by the National Institute of Standards and Technology NIST another. scenario is added called the watch list, Face verification Am I who I say I am is a one to one match that compares a query. Figure 2 Face recognition market 9 a Total biometric revenues 2002 2007 b Comparative. market share by technology, face image against a template face image whose identity is being claimed To evaluate the. verification performance the verification rate the rate at which legitimate users are granted. access vs false accept rate the rate at which imposters are granted access is plotted called. ROC curve A good verification system should balance these two rates based on operational. Face identification Who am I is a one to many matching process that compares a query. face image against all the template images in a face database to determine the identity of the. query face see Fig 3 The identification of the test image is done by locating the image in. the database who has the highest similarity with the test image The identification process is. a closed test which means the sensor takes an observation of an individual that is known to. be in the database The test subject s normalized features are compared to the other features. in the system s database and a similarity score is found for each comparison These similarity. scores are then numerically ranked in a descending order The percentage of times that the. highest similarity score is the correct match for all individuals is referred to as the top match. score If any of the top r similarity scores corresponds to the test subject it is considered. as a correct match in terms of the cumulative match The percentage of times one of those. r similarity scores is the correct match for all individuals is referred to as the Cumulative. Match Score The Cumulative Match Score curve is the rank n versus percentage of correct. identification where rank n is the number of top similarity scores reported. Figure 3 Face identification scenario, The watch list Are you looking for me method is an open universe test The test indi. vidual may or may not be in the system database That person is compared to the others in. the system s database and a similarity score is reported for each comparison These similarity. scores are then numerically ranked so that the highest similarity score is first If a similarity. score is higher than a preset threshold an alarm is raised If an alarm is raised the system. thinks that the individual is located in the system s database There are two main items of. interest for watch list applications The first is the percentage of times the system raises the. alarm and it correctly identifies a person on the watchlist This is called the Detection and. Identification Rate The second item of interest is the percentage of times the system raises. the alarm for an individual that is not on the watchlist database This is called the False. Alarm Rate, In this report all the experiments are conducted in the identification scenario.

Human face image appearance has potentially very large intra subject variations due to. 3D head pose,Illumination including indoor outdoor. Facial expression, Occlusion due to other objects or accessories e g sunglasses scarf etc. Facial hair, On the other hand the inter subject variations are small due to the similarity of individual appear. ances Fig 4 gives examples of appearance variations of one subject And Fig 5 illustrates examples. of appearance variations of different subjects Currently image based face recognition techniques. can be mainly categorized into two groups based on the face representation which they use i. appearance based which uses holistic texture features ii model based which employ shape and. texture of the face along with 3D depth information. Figure 4 Appearance variations of the same subject under different lighting conditions and different. facial expressions 13, A number of face recognition algorithms along with their modifications have been developed. during the past several decades see Fig 6 In section 2 three leading linear subspace analysis. schemes are presented and several non linear manifold analysis approaches for face recognition are. briefly described The model based approaches are introduced in section 3 including Elastic Bunch. Graph matching Active Appearance Model and 3D Morphable Model methods A number of face. databases available in the public domain and several published performance evaluation results are. provided in section 4 Concluding remarks and future research directions are summarized in section 5. Figure 5 Inter subject variations versus intra subject variations a and b are images from. different subjects but their appearance variations represented in the input space can be smaller. than images from the same subject b c and d These images are taken from from Yale database B. 2 Appearance based View based face recognition, Many approaches to object recognition and to computer graphics are based directly on images.

without the use of intermediate 3D models Most of these techniques depend on a representation of. images that induces a vector space structure and in principle requires dense correspondence. Appearance based approaches represent an object in terms of several object views raw intensity. images An image is considered as a high dimensional vector i e a point in a high dimensional. vector space Many view based approaches use statistical techniques to analyze the distribution of. the object image vectors in the vector space and derive an efficient and effective representation. feature space according to different applications Given a test image the similarity between the. stored prototypes and the test view is then carried out in the feature space. Figure 6 Face recognition methods covered in this report. This image vector representation allows the use of learning techniques for the analysis and for. the synthesis of images Face recognition can be treated as a space searching problem combined. with a machine learning problem,2 1 Vector representation of images. Image data can be represented as vectors i e as points in a high dimensional vector space For. example a p q 2D image can be mapped to a vector x Rpq by lexicographic ordering of the pixel. elements such as by concatenating each row or column of the image Despite this high dimensional. embedding the natural constraints of the physical world and the imaging process dictate that the. data will in fact lie in a lower dimensional though possibly disjoint manifold The primary goal. of the subspace analysis is to identify represent and parameterize this manifold in accordance with. some optimality criteria, Let X x1 x2 xi xN represent the n N data matrix where each xi is a face vector. of dimension n concatenated from a p q face image where n p q Here n represents the total. number of pixels in the face image and N is the number of different face images in the training set. The mean vector of the training images i 1 xi is subtracted from each image vector. 2 2 Linear subspace Analysis, Three classical linear appearance based classifiers PCA 14 ICA 15 and LDA 16 17 are intro. duced in the following Each classifier has its own representation basis vectors of a high dimensional. face vector space based on different statistical viewpoints By projecting the face vector to the basis. vectors the projection coefficients are used as the feature representation of each face image The. matching score between the test face image and the training prototype is calculated e g as the. cosine value of the angle between their coefficients vectors The larger the matching score the. better the match, All the three representations can be considered as a linear transformation from the original image. vector to a projection feature vector i e, where Y is the d N feature vector matrix d is the dimension of the feature vector and W is the.

transformation matrix Note that d n, The Eigenface algorithm uses the Principal Component Analysis PCA for dimensionality reduction. to find the vectors which best account for the distribution of face images within the entire image. space 14 These vectors define the subspace of face images and the subspace is called face space. All faces in the training set are projected onto the face space to find a set of weights that describes. the contribution of each vector in the face space To identify a test image it requires the projection. of the test image onto the face space to obtain the corresponding set of weights By comparing the. weights of the test image with the set of weights of the faces in the training set the face in the test. image can be identified, The key procedure in PCA is based on Karhumen Loeve transformation 18 If the image. elements are considered to be random variables the image may be seen as a sample of a stochastic. process The Principal Component Analysis basis vectors are defined as the eigenvectors of the. scatter matrix ST,ST xi xi T 2, The transformation matrix WP CA is composed of the eigenvectors corresponding to the d largest. eigenvalues A 2D example of PCA is demonstrated in Fig 7 The eigenvectors a k a eigenface. Figure 7 Principal components PC of a two dimensional set of points The first principal compo. nent provides an optimal linear dimension reduction from 2D to 1D in the sense of the mean square. corresponding to the 7 largest eigenvalues derived from ORL face database 19 are shown in. Fig 9 The corresponding average face is given in Fig 8 ORL face samples are provided in Fig 26. After applying the projection the input vector face in an n dimensional space is reduced to a. feature vector in a d dimensional subspace Also the eigenvectors corresponding to the 7 smallest. eigenvalues are provided in Fig 10 For most applications these eigenvectors corresponding to very. small eigenvalues are considered as noise and not taken into account during identification Several. extensions of PCA are developed such as modular eigenspaces 20 and probabilistic subspaces 21. Figure 8 The average face derived from the ORL face database 19. Independent Component Analysis ICA 22 is similar to PCA except that the distribution of the. components are designed to be non Gaussian Maximizing non Gaussianity promotes statistical. Figure 9 Eigenvectors corresponding to the 7 largest eigenvalues shown as p p images where. p p n derived from the ORL face database 19, Figure 10 Eigenvectors corresponding to the 7 smallest eigenvalues shown as p p images where. p p n derived from the ORL face database 19, independence Figure 11 presents the different feature extraction properties between PCA and ICA.

Bartlett et al 15 provided two architectures based on Independent Component Analysis sta. tistically independent basis images and a factorial code representation for the face recognition task. The ICA separates the high order moments of the input in addition to the second order moments. utilized in PCA Both the architectures lead to a similar performance The obtained basis vectors. based on fast fixed point algorithm 24 for the ICA factorial code representation are illustrated in. Fig 12 There is no special order imposed on the ICA basis vectors. Both PCA and ICA construct the face space without using the face class category information. The whole face training data is taken as a whole In LDA the goal is to find an efficient or interesting. way to represent the face vector space But exploiting the class information can be helpful to the. identification tasks see Fig 13 for an example, The Fisherface algorithm 16 is derived from the Fisher Linear Discriminant FLD which uses. class specific information By defining different classes with different statistics the images in the. learning set are divided into the corresponding classes Then techniques similar to those used. in Eigenface algorithm are applied The Fisherface algorithm results in a higher accuracy rate in. Figure 11 Top Example 3D data distribution and the corresponding principal component and. independent component axes Each axis is a direction found by P. market share by technology face image against a template face image whose identity is being claimed To evaluate the veri cation performance the veri cation rate the rate at which legitimate users are granted access vs false accept rate the rate at which imposters are granted access is plotted called ROC curve A good veri cation

Recent Views:

- Attracting native wildlife into your garden
- We shape air betag com
- Chapitre 3 la gravitation
- Drama lote mltav
- Requirements analysis through viewpoints oriented
- Safety analysis report a comparison of incidents from
- Harvest fatty acid methyl ester and a edible oil
- Microwave multi receivers radars for snowpack monitoring
- Sinopsis u ketegangan berbasis agama terus terjadi
- Chapter 5 property requirements