Image analysis – Learning systems – Trainable classifiers or pattern recognizers
Reexamination Certificate
2002-05-31
2004-11-30
Bali, Vikkram (Department: 2623)
Image analysis
Learning systems
Trainable classifiers or pattern recognizers
C382S118000
Reexamination Certificate
active
06826300
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to field of face recognition systems. More specifically, the present invention utilizes a novel Gabor Feature Classifier for face recognition.
Understanding how people process and recognize each other's face, and the development of corresponding computational models for automated face recognition are among the most challenging tasks for visual form (‘shape’) analysis and object recognition. The enormity of the problem has involved hundreds of scientists in interdisciplinary research, but the ultimate solution is yet to come.
Face recognition is largely motivated by the need for surveillance and security, telecommunication and digital libraries, human-computer intelligent interaction, and smart environments. Some of these security uses may include log in control and physical access control. Further applications may include: law enforcement uses such as mug shot albums, and criminology; and commercial transaction which use credit cards, driver's licenses, passports, or other photo ID identifications. Virtually all applications that depend upon the identification of a person could benefit from this technology.
The solutions suggested so far are synergetic efforts from fields such as signal and image processing, pattern recognition, machine learning, neural networks, statistics, evolutionary computation, psychophysics of human perception and neurosciences, and system engineering. A generic approach often used involves statistical estimation and learning of face class statistics for subsequent face detection and classification. Face detection generally applies a statistical characterization of faces and non-faces to build a classifier, which may then used to search over different locations and scales for image patterns that are likely to be human faces.
Face recognition usually employs various statistical techniques to derive appearance-based models for classification. Some of these techniques include but are not limited to: principal component analysis (hereinafter referred to as PCA); Fisher linear discriminant (hereinafter referred to as FLD) which are also known as linear discriminant analysis (hereinafter referred to as LDA); independent component analysis (hereinafter referred to as ICA); local feature analysis (hereinafter referred to as LFA); and Gabor and bunch graphs. Descriptions of PCA may be found in: [M. Turk and A. Pentland, “Eigenfaces for recognition,” Journal of Cognitive Neuroscience, vol. 13, no. 1, pp. 71-86, 1991], and [B. Moghaddam and A. Pentland, “Probabilistic visual learning for object representation,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696-710, 1997]. Descriptions of FLD and LDA may be found in: [D. L. Swets and J. Weng, “Using discriminant eigenfeatures for image retrieval,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 18, no. 8, pp. 831-836, 1996.]; [P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 711-720, 1997.], and [K. Etemad and R. Chellappa, “Discriminant analysis for recognition of human face images,” J. Opt. Soc. Am. A, vol. 14, pp. 1724-1733, 1997]. A description of ICA may be found in: [G. Donato, M. S. Bartlett, J. C. Hager, P. Ekman, and T. J. Sejnowski, “Classifying facial actions,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 21, no. 10, pp. 974-989, 1999]. LFA is described in [P. S. Penev and J. J. Atick, “Local feature analysis: A general statistical theory for object representation,” Network: Computation in Neural Systems, vol. 7, pp. 477-500, 1996]. A description of Gabor and bunch graphs may be found in [L. Wiskott, J. M. Fellous, N. Kruger, and C. von der Malsburg, “Face recognition by elastic bunch graph matching,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 775-779, 1997].
Face recognition may depend heavily on the particular choice of features used by the classifier. One usually starts with a given set of features and then attempts to derive an optimal subset (under some criteria) of features leading to high classification performance with the expectation that similar performance may be also displayed on future trials using novel (unseen) test data. PCA is a popular technique used to derive a starting set of features for both face representation and recognition. Kirby and Sirovich showed that any particular face may be (i) economically represented along the eigenpictures coordinate space, and (ii) approximately reconstructed using just a small collection of eigenpictures and their corresponding projections (‘coefficients’). [M. Kirby and L. Sirovich, “Application of the Karhunen-Loeve procedure for the characterization of human faces,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 12, no. 1, pp. 103-108, 1990].
Applying PCA technique to face recognition, Turk and Pentland developed a well-known Eigenfaces method that sparked an explosion of interests in applying statistical techniques to face recognition. [M. Turk and A. Pentland, “Eigenfaces for recognition,” Journal of Cognitive Neuroscience, vol. 13, no. 1, pp. 71-86, 1991]. PCA, an optimal representation criterion (in the sense of mean square error), does not consider the classification aspect. One solution for taking into account and improving the classification performance is to combine PCA, the optimal representation criterion, with the Bayes classifier, the optimal classification criterion (when the density functions are given). Toward that end, Moghaddam and Pentland developed a probabilistic visual learning method, which uses the eigenspace decomposition as an integral part of estimating complete density functions in high-dimensional image space. [B. Moghaddam and A. Pentland, “Probabilistic visual learning for object representation,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 696-710, 1997]. While the leading eigenvalues are derived directly by PCA, the remainder of the eigenvalue spectrum is estimated by curve fitting.
Rather than estimating the densities in high-dimensional space, Liu and Wechsler developed a PRM (Probabilistic Reasoning Models) method by first applying PCA for dimensionality reduction and then applying the Bayes classifier and the MAP rule for classification. [C. Liu and H. Wechsler, “Robust coding schemes for indexing and retrieval from large face databases,” IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132-137, 2000]. The rationale of the PRM method is that of lowering the space dimension subject to increased fitness for the discrimination index by estimating the conditional density function of each class using the within-class scatter in the reduced PCA space.
Another important statistical technique widely used in face recognition is the FLD, which models both the within- and the between-class scatters. FLD, which is behind several face recognition methods, induces non-orthogonal projection bases, a characteristic known to have great functional significance in biological sensory systems [J. G. Daugman, “An information-theoretic view of analog representation in striate cortex,” in Computational Neuroscience, E. L. Schwartz, Ed., pp. 403-424. MIT Press, 1990]. As the original image space is highly dimensional, most face recognition methods perform first dimensionality reduction using PCA, as it is the case with the Fisherfaces method suggested by Belhumeur, Hespanha, and Kriegman [P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 711-720, 1997]. Swets and Weng have pointed out that the Eigenfaces method derives only the Most Expressive Features (MEF) and that PCA inspired featu
Liu Chengjun
Wechsler Harry
Bali Vikkram
George Mason University
Grossman David G.
LandOfFree
Feature based classification does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Feature based classification, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Feature based classification will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3363385