Microsoft power point Face recognition
1 like762 views
The document discusses a seminar focused on image quality assessment for fake biometric detection, detailing the definition and types of biometrics, including both physiological and behavioral attributes. It explains various methods for facial recognition and the implementation of techniques like Hidden Markov Models (HMMs) and Singular Value Decomposition (SVD) for image processing and feature extraction. Additionally, it outlines the preprocessing steps and filtering techniques used to enhance image quality for biometric applications.
![REFERENCESREFERENCES
[1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face
Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez
[2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD
coefficients By: - H. Miar-Naimi and P. Davari
[3].Face recognition using SingularValue Decomposition and Hidden Markov Models
PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA
[4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit
Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003.
[5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004.
[6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of
signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375.
[1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face
Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez
[2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD
coefficients By: - H. Miar-Naimi and P. Davari
[3].Face recognition using SingularValue Decomposition and Hidden Markov Models
PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA
[4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit
Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003.
[5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004.
[6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of
signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375.
REFERENCESREFERENCES
[1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face
Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez
[2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD
coefficients By: - H. Miar-Naimi and P. Davari
[3].Face recognition using SingularValue Decomposition and Hidden Markov Models
PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA
[4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit
Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003.
[5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004.
[6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of
signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375.
[1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face
Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez
[2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD
coefficients By: - H. Miar-Naimi and P. Davari
[3].Face recognition using SingularValue Decomposition and Hidden Markov Models
PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA
[4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit
Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003.
[5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004.
[6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of
signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375.](https://image.slidesharecdn.com/microsoftpowerpoint-facerecognition-150402003754-conversion-gate01/85/Microsoft-power-point-Face-recognition-22-320.jpg)
Microsoft power point Face recognition
- 1. WELCOME TO SEMINARWELCOME TO SEMINAR SEMINAR ON IMAGE QUALITY ASSESSMENT FOR FACK BIOMETRIC DETECTION SEMINAR ON IMAGE QUALITY ASSESSMENT FOR FACK BIOMETRIC DETECTION WELCOME TO SEMINARWELCOME TO SEMINAR SEMINAR ON IMAGE QUALITY ASSESSMENT FOR FACK BIOMETRIC DETECTION SEMINAR ON IMAGE QUALITY ASSESSMENT FOR FACK BIOMETRIC DETECTION
- 2. What is a biometric?What is a biometric? A biometric is a unique, measurable characteristic of a human being that can be used to automatically recognize an individual or verify an individual identity . Biometrics can measure both physiological and behavioral characteristics. A biometric is a unique, measurable characteristic of a human being that can be used to automatically recognize an individual or verify an individual identity . Biometrics can measure both physiological and behavioral characteristics. What is a biometric?What is a biometric? A biometric is a unique, measurable characteristic of a human being that can be used to automatically recognize an individual or verify an individual identity . Biometrics can measure both physiological and behavioral characteristics. A biometric is a unique, measurable characteristic of a human being that can be used to automatically recognize an individual or verify an individual identity . Biometrics can measure both physiological and behavioral characteristics.
- 3. BIOMETRICS BEHAVIORAL ATTRIBUTESBEHAVIORAL ATTRIBUTES •Signature •Keystrokes BIOMETRICS PHYSICAL ATTRIBUTESPHYSICAL ATTRIBUTES • Fingerprint • Face • Retina • Iris • Hand and finger geometry
- 4. Fake Biometric Detection:Fake Biometric Detection:-- 1) Hardware-based techniques 2) Software-based techniques Fake Biometric Detection:Fake Biometric Detection:-- 1) Hardware-based techniques 2) Software-based techniques
- 5. General method of fake biometric detectionGeneral method of fake biometric detection based on Image Quality Assessment:based on Image Quality Assessment:-- General method of fake biometric detectionGeneral method of fake biometric detection based on Image Quality Assessment:based on Image Quality Assessment:--
- 7. Biometrics ProcessBiometrics Process Biometric Data Collection Transmission new biometric sample is requested. Biometrics ProcessBiometrics Process Quality Sufficient? Yes Signal Processing, Feature Extraction, Representation new biometric sample is requested. No No Yes Yes Template MatchDatabase GenerateTemplate Decision Confidence?
- 8. FACE RECOGNITIONFACE RECOGNITION Face recognition is recognizing a special face from a set of different face There have been a several face recognition methods, Common face recognition methods are: Geometrical Feature Matching – Based on the extraction of a set of geometrical features having 75% recognition rate. Eigen faces method -Uses the Principal Component Analysis (PCA) to project faces into a low dimensional space having 90.5% recognition rat Bunch Graph Matching- Neural Networks- Uses Probabilistic Decision-Based Neural Network (PDBNN) for face recognition having 96% recognition rate SupportVector Machines Elastic Matching Hidden Markov Models along with SVD coefficient- Face recognition is recognizing a special face from a set of different face There have been a several face recognition methods, Common face recognition methods are: Geometrical Feature Matching – Based on the extraction of a set of geometrical features having 75% recognition rate. Eigen faces method -Uses the Principal Component Analysis (PCA) to project faces into a low dimensional space having 90.5% recognition rat Bunch Graph Matching- Neural Networks- Uses Probabilistic Decision-Based Neural Network (PDBNN) for face recognition having 96% recognition rate SupportVector Machines Elastic Matching Hidden Markov Models along with SVD coefficient- FACE RECOGNITIONFACE RECOGNITION Face recognition is recognizing a special face from a set of different face There have been a several face recognition methods, Common face recognition methods are: Geometrical Feature Matching – Based on the extraction of a set of geometrical features having 75% recognition rate. Eigen faces method -Uses the Principal Component Analysis (PCA) to project faces into a low dimensional space having 90.5% recognition rat Bunch Graph Matching- Neural Networks- Uses Probabilistic Decision-Based Neural Network (PDBNN) for face recognition having 96% recognition rate SupportVector Machines Elastic Matching Hidden Markov Models along with SVD coefficient- Face recognition is recognizing a special face from a set of different face There have been a several face recognition methods, Common face recognition methods are: Geometrical Feature Matching – Based on the extraction of a set of geometrical features having 75% recognition rate. Eigen faces method -Uses the Principal Component Analysis (PCA) to project faces into a low dimensional space having 90.5% recognition rat Bunch Graph Matching- Neural Networks- Uses Probabilistic Decision-Based Neural Network (PDBNN) for face recognition having 96% recognition rate SupportVector Machines Elastic Matching Hidden Markov Models along with SVD coefficient-
- 9. HIDDEN MARKOV MODELHIDDEN MARKOV MODEL HMMs are generally used to model one dimensional data. Every HMM is associated with non-observable (hidden) state and an observable sequence generated by the hidden states individually. The Markov process is determined by the current state with initial state distribution π and the transition probability matrix A. We observe only the Oi (the observation sequence), which is related to the (hidden) states of the Markov process by the emission probability matrix B. Using shorthand notation HMM is defined as following: λ =(A,B,π) WHERE A={aij} is the state transition probability matrix, B={bjk} is the observation symbol probability matrix, π={π1,π2,…,πN} is the initial state distribution. HMMs are generally used to model one dimensional data. Every HMM is associated with non-observable (hidden) state and an observable sequence generated by the hidden states individually. The Markov process is determined by the current state with initial state distribution π and the transition probability matrix A. We observe only the Oi (the observation sequence), which is related to the (hidden) states of the Markov process by the emission probability matrix B. Using shorthand notation HMM is defined as following: λ =(A,B,π) WHERE A={aij} is the state transition probability matrix, B={bjk} is the observation symbol probability matrix, π={π1,π2,…,πN} is the initial state distribution. A CB HIDDEN MARKOV MODELHIDDEN MARKOV MODEL HMMs are generally used to model one dimensional data. Every HMM is associated with non-observable (hidden) state and an observable sequence generated by the hidden states individually. The Markov process is determined by the current state with initial state distribution π and the transition probability matrix A. We observe only the Oi (the observation sequence), which is related to the (hidden) states of the Markov process by the emission probability matrix B. Using shorthand notation HMM is defined as following: λ =(A,B,π) WHERE A={aij} is the state transition probability matrix, B={bjk} is the observation symbol probability matrix, π={π1,π2,…,πN} is the initial state distribution. HMMs are generally used to model one dimensional data. Every HMM is associated with non-observable (hidden) state and an observable sequence generated by the hidden states individually. The Markov process is determined by the current state with initial state distribution π and the transition probability matrix A. We observe only the Oi (the observation sequence), which is related to the (hidden) states of the Markov process by the emission probability matrix B. Using shorthand notation HMM is defined as following: λ =(A,B,π) WHERE A={aij} is the state transition probability matrix, B={bjk} is the observation symbol probability matrix, π={π1,π2,…,πN} is the initial state distribution. D
- 10. •We divide image faces into seven regions in which each is assigned to a state in a left to right one dimensional HMM. •We divide image faces into seven regions in which each is assigned to a state in a left to right one dimensional HMM.
- 11. SINGULAR VALUE DECOMPOSITIONSINGULAR VALUE DECOMPOSITION A singular value decomposition of a m×n matrix X is any function of the form: X=UWV where U(m×m) and V(n×n) are orthogonal matrix and W is an m×n diagonal matrix of singular values Singular values of given data matrix contain information about the noise level, the energy, the rank of the matrix, etc. SVD provides a new way for extracting algebraic features from an image. A singular value decomposition of a m×n matrix X is any function of the form: X=UWV where U(m×m) and V(n×n) are orthogonal matrix and W is an m×n diagonal matrix of singular values Singular values of given data matrix contain information about the noise level, the energy, the rank of the matrix, etc. SVD provides a new way for extracting algebraic features from an image. SINGULAR VALUE DECOMPOSITIONSINGULAR VALUE DECOMPOSITION A singular value decomposition of a m×n matrix X is any function of the form: X=UWV where U(m×m) and V(n×n) are orthogonal matrix and W is an m×n diagonal matrix of singular values Singular values of given data matrix contain information about the noise level, the energy, the rank of the matrix, etc. SVD provides a new way for extracting algebraic features from an image. A singular value decomposition of a m×n matrix X is any function of the form: X=UWV where U(m×m) and V(n×n) are orthogonal matrix and W is an m×n diagonal matrix of singular values Singular values of given data matrix contain information about the noise level, the energy, the rank of the matrix, etc. SVD provides a new way for extracting algebraic features from an image.
- 12. •The image is resized to around 50% of its size. Originally the images have 112x92 (ORL) and after resizing the images go down to 56x46 or 64x64 pixels. •The image is resized to around 50% of its size. Originally the images have 112x92 (ORL) and after resizing the images go down to 56x46 or 64x64 pixels.
- 13. Image PreprocessingImage Preprocessing The Olivetti Research Laboratory (ORL) face database contains ten differe images of each of the 40 persons. The images are in PGM format. The size of each image is 112x92 pixels with 256 grey levels per pixel. The dataset is divided into two parts – one for training and one for testing. We use 5 images from each folder for training the system and the rest 5 images for testing. Next, SVD is applied to extract features from the images and HMM to buil a recognition model. The model returns probabilities of how likely the unseen face image looks like each one of the images used for training and the face with the highest probability is assigned as the recognized face. The Olivetti Research Laboratory (ORL) face database contains ten differe images of each of the 40 persons. The images are in PGM format. The size of each image is 112x92 pixels with 256 grey levels per pixel. The dataset is divided into two parts – one for training and one for testing. We use 5 images from each folder for training the system and the rest 5 images for testing. Next, SVD is applied to extract features from the images and HMM to buil a recognition model. The model returns probabilities of how likely the unseen face image looks like each one of the images used for training and the face with the highest probability is assigned as the recognized face. Image PreprocessingImage Preprocessing The Olivetti Research Laboratory (ORL) face database contains ten differe images of each of the 40 persons. The images are in PGM format. The size of each image is 112x92 pixels with 256 grey levels per pixel. The dataset is divided into two parts – one for training and one for testing. We use 5 images from each folder for training the system and the rest 5 images for testing. Next, SVD is applied to extract features from the images and HMM to buil a recognition model. The model returns probabilities of how likely the unseen face image looks like each one of the images used for training and the face with the highest probability is assigned as the recognized face. The Olivetti Research Laboratory (ORL) face database contains ten differe images of each of the 40 persons. The images are in PGM format. The size of each image is 112x92 pixels with 256 grey levels per pixel. The dataset is divided into two parts – one for training and one for testing. We use 5 images from each folder for training the system and the rest 5 images for testing. Next, SVD is applied to extract features from the images and HMM to buil a recognition model. The model returns probabilities of how likely the unseen face image looks like each one of the images used for training and the face with the highest probability is assigned as the recognized face.
- 14. FILTERINGFILTERING In order to compensate the flash effect and reduce the salt noise, a nonlinear minimum order-static filter is used. Order-statistic filter is used to improve speed and recognition rate of the system. The filter has a smoothing role and reduces the image information. A sliding window moves from left to right and top to down with steps of size one pixel, at each situation the centered pixel is replaced by one of pixels of the window based on the type of filter. In order to compensate the flash effect and reduce the salt noise, a nonlinear minimum order-static filter is used. Order-statistic filter is used to improve speed and recognition rate of the system. The filter has a smoothing role and reduces the image information. A sliding window moves from left to right and top to down with steps of size one pixel, at each situation the centered pixel is replaced by one of pixels of the window based on the type of filter. Before filtering FILTERINGFILTERING In order to compensate the flash effect and reduce the salt noise, a nonlinear minimum order-static filter is used. Order-statistic filter is used to improve speed and recognition rate of the system. The filter has a smoothing role and reduces the image information. A sliding window moves from left to right and top to down with steps of size one pixel, at each situation the centered pixel is replaced by one of pixels of the window based on the type of filter. In order to compensate the flash effect and reduce the salt noise, a nonlinear minimum order-static filter is used. Order-statistic filter is used to improve speed and recognition rate of the system. The filter has a smoothing role and reduces the image information. A sliding window moves from left to right and top to down with steps of size one pixel, at each situation the centered pixel is replaced by one of pixels of the window based on the type of filter. After filtering
- 15. OBSERVATION SEQUENCEOBSERVATION SEQUENCE The observation sequence is generated by dividing each face image of width W and height H into overlapping blocks of height L and width W A L×W window is slid from top to bottom on the image and creates a sequence of overlapping blocks. The number of blocks extracted from each face image is given by: The observation sequence is generated by dividing each face image of width W and height H into overlapping blocks of height L and width W A L×W window is slid from top to bottom on the image and creates a sequence of overlapping blocks. The number of blocks extracted from each face image is given by: OBSERVATION SEQUENCEOBSERVATION SEQUENCE The observation sequence is generated by dividing each face image of width W and height H into overlapping blocks of height L and width W A L×W window is slid from top to bottom on the image and creates a sequence of overlapping blocks. The number of blocks extracted from each face image is given by: The observation sequence is generated by dividing each face image of width W and height H into overlapping blocks of height L and width W A L×W window is slid from top to bottom on the image and creates a sequence of overlapping blocks. The number of blocks extracted from each face image is given by:
- 17. FEATURE SELECTIONFEATURE SELECTIONFEATURE SELECTIONFEATURE SELECTION
- 18. QuantizationQuantization Each element is quantized into Di distinct levels.The difference between two quantized values is: Every element from vector C is replaced with its quantized value Each element is quantized into Di distinct levels.The difference between two quantized values is: Every element from vector C is replaced with its quantized value QuantizationQuantization Each element is quantized into Di distinct levels.The difference between two quantized values is: Every element from vector C is replaced with its quantized value Each element is quantized into Di distinct levels.The difference between two quantized values is: Every element from vector C is replaced with its quantized value
- 20. The Training Process After representing each face image by observation vectors, they are modeled by a seven -state HMM HMM is trained for each person in the database using the baum-welch algorithm After representing each face image by observation vectors, they are modeled by a seven -state HMM HMM is trained for each person in the database using the baum-welch algorithm The Training Process After representing each face image by observation vectors, they are modeled by a seven -state HMM HMM is trained for each person in the database using the baum-welch algorithm After representing each face image by observation vectors, they are modeled by a seven -state HMM HMM is trained for each person in the database using the baum-welch algorithm
- 21. CONCLUSIONCONCLUSION The study of the biometric systems against different types of attacks has been a very active field of research in recent years.Thi interest has lead to big advances in the field of security-enhancing technologies for biometric-based applications. However, in spite o this noticeable improvement, the development of efficient protection methods against known threats has proven to be a challenging task. The study of the biometric systems against different types of attacks has been a very active field of research in recent years.Thi interest has lead to big advances in the field of security-enhancing technologies for biometric-based applications. However, in spite o this noticeable improvement, the development of efficient protection methods against known threats has proven to be a challenging task. CONCLUSIONCONCLUSION The study of the biometric systems against different types of attacks has been a very active field of research in recent years.Thi interest has lead to big advances in the field of security-enhancing technologies for biometric-based applications. However, in spite o this noticeable improvement, the development of efficient protection methods against known threats has proven to be a challenging task. The study of the biometric systems against different types of attacks has been a very active field of research in recent years.Thi interest has lead to big advances in the field of security-enhancing technologies for biometric-based applications. However, in spite o this noticeable improvement, the development of efficient protection methods against known threats has proven to be a challenging task.
- 22. REFERENCESREFERENCES [1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez [2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD coefficients By: - H. Miar-Naimi and P. Davari [3].Face recognition using SingularValue Decomposition and Hidden Markov Models PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA [4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003. [5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004. [6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375. [1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez [2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD coefficients By: - H. Miar-Naimi and P. Davari [3].Face recognition using SingularValue Decomposition and Hidden Markov Models PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA [4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003. [5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004. [6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375. REFERENCESREFERENCES [1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez [2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD coefficients By: - H. Miar-Naimi and P. Davari [3].Face recognition using SingularValue Decomposition and Hidden Markov Models PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA [4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003. [5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004. [6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375. [1]. Image Quality Assessment for Fake Biometric Detection:Application to Iris, Fingerprint,and Face Recognition By:- Javier Galbally, Sébastien Marcel, Member, IEEE, and Julian Fierrez [2].A new, fast and efficient HMM-based face recognition system using a 7-state HMM along with SVD coefficients By: - H. Miar-Naimi and P. Davari [3].Face recognition using SingularValue Decomposition and Hidden Markov Models PETYA DINKOVA1, PETIA GEORGIEVA2, MARIOFANNA MILANOVA [4] S. Prabhakar, S. Pankanti, and A. K. Jain,“Biometric recognition: Security and privacy concerns,” IEEE Securit Privacy, vol. 1, no. 2, pp. 33–42, Mar./Apr. 2003. [5]T. Matsumoto,“Artificial irises: Importance of vulnerability analysis,” in Proc.AWB, 2004. [6]J. Hennebert, R. Loeffel,A. Humm, and R. Ingold,“A new forgery scenario based on regaining dynamics of signature,” in Proc. IAPR ICB, vol. Springer LNCS-4642. 2007, pp. 366–375.
- 23. THANKTHANK YOUYOUTHANKTHANK YOUYOUTHANKTHANK YOUYOUTHANKTHANK YOUYOU