Face detection using discriminating feature analysis and Support Vector Machine
Introduction
Face detection methods generally learns statistical models of face and nonface images, and then apply two-class classification rules to discriminate between face and nonface patterns [1], [2]. As a face must be located and extracted before it can be verified or identified, face detection is the first step towards building an automated face verification or identification system. Face verification mainly concerns authenticating a claimed identity posed by a person, while face identification focuses on recognizing the identity of a person from a database of known individuals [3], [4]. An automated vision system that performs the functions of face detection, verification, and identification has great potential in a wide spectrum of applications, such as airport security and access control, building (embassy) surveillance and monitoring, human–computer intelligent interaction and perceptual interfaces, smart environments for home, office, and cars [2], [1], [5], [6], [7].
This paper presents a novel face detection method by applying discriminating feature analysis (DFA) and Support Vector Machine (SVM). The novelty of our DFA–SVM method comes from the integration of DFA, face class modeling, and SVM for face detection. Our DFA–SVM method works as follows: First, DFA derives a discriminating feature vector by combining the input image, its 1-D Haar wavelet representation, and its amplitude projections. While the Haar wavelets produce an effective representation for object detection, the amplitude projections capture the vertical symmetric distributions and the horizontal characteristics of human face images. Second, face class modeling statistically estimates the probability density function (PDF) of the face class and defines a distribution-based measure for face and nonface classification. The face class is modeled as a multivariate normal distribution [8], and the distribution-based measure then separates the input patterns into three classes: the face class (patterns close to the face class), the nonface class (patterns far away from the face class), and the undecided class (patterns neither close to nor far away from the face class). Note that the distribution-based measure also derives nonface patterns for SVM training. Finally, the SVM together with the distribution-based measure classifies the patterns in the undecided class into either the face class or the nonface class. Experiments using images from the MIT–CMU test sets show the feasibility of our new face detection method. The DFA–SVM method is trained using 600 FERET facial images [9] and 3813 nonface images that lie close to the face class. When tested using 92 images (containing 282 faces) from the MIT–CMU test sets [10], our DFA–SVM method achieves correct face detection rate with two false detections, a performance comparable to the state-of-the-art face detection methods, such as Schneiderman–Kanade's method [11].
Section snippets
Background
Earlier efforts of face detection research have been focused on correlation or template matching, matched filtering, subspace methods, deformable templates, etc. [12], [13]. For comprehensive surveys of these early methods, see Refs. [14], [6], [7]. Recent face detection approaches, however, emphasize on statistical modeling and machine learning techniques [15], [16]. Some representative methods are the probabilistic visual learning method [8], the example-based learning method [17], the neural
Face detection using discriminating feature analysis and Support Vector Machine
The system architecture of our DFA–SVM face detection method is shown in Fig. 1. An input image is first processed by the DFA, which defines a feature vector by combining the input image, its 1-D Haar wavelet representation, and its amplitude projections. Based on the DFA feature vector, the face class is then modeled using a multivariate normal density and a distribution-based measure is defined for face and nonface classification. The distribution-based measure separates the patterns in the
Experiments
This section details statistical learning, learning the thresholds, face detection performance, and computational complexity of the DFA–SVM method. The training data for the DFA–SVM method comes from the FERET database [9] Batch 15, which contains 600 frontal face images. The testing data comes from the MIT–CMU test sets [10], which include images from diverse sources. Experimental results show that our DFA–SVM method, which is trained on a simple image set yet works on much more complex
Conclusions
This paper presents a novel face detection method by integrating DFA, face class modeling, and SVM. Discriminating feature analysis derives a feature vector by combining the input image, its 1-D Haar wavelet representation, and its amplitude projections. Face class modeling, then, estimates the PDF of the face class and defines a distribution-based measure for face and nonface classification. And finally, SVM together with the distribution-based measure classifies the patterns in an input image
Acknowledgements
This work was partially supported by the TSWG R&D Contract N41756-03-C-4026.
About the Author—PEICHUNG SHIH received his M.S. degree with summa cum laude in Information Systems in 2002 from New Jersey Institute of Technology. He is a Ph.D. candidate in Computer Science at New Jersey Institute of Technology. His research interests lie in the field of image and video processing, computer vision, and pattern recognition with applications toward face recognition. His present work includes developing novel face detection systems by applying computer vision concepts and
References (36)
- et al.
Automatic recognition and analysis of human faces and facial expression: a survey
Pattern Recognition
(1992) - et al.
The FERET database and evaluation procedure for face recognition algorithms
Image Vision Comput.
(1998) - et al.
Face detection: a survey
Comput. Vision Image Understanding
(2001) - et al.
Hierarchical classification and feature reduction for fast face detection with support vector machines
Pattern Recognition
(2003) Looking at people: sensing for ubiquitous and wearable computing
IEEE Trans. Pattern Anal. Mach. Intell.
(2000)- et al.
Statistical pattern recognition: a review
IEEE Trans. Pattern Anal. Mach. Intell.
(2000) Gabor-based kernel PCA with fractional power polynomial models for face recognition
IEEE Trans. Pattern Anal. Mach. Intell.
(2004)Enhanced independent component analysis and its application to content based face image retrieval
IEEE Trans. Systems Man Cybernet., Part B: Cybernetics
(2004)Face and gesture recognition: overview
IEEE Trans. Pattern Anal. Mach. Intell.
(1997)- et al.
Human and machine recognition of faces: a survey
Proc. IEEE
(1995)
Probabilistic visual learning for object representation
IEEE Trans. Pattern Anal. Mach. Intell.
Neural network-based face detection
IEEE Trans. Pattern Anal. Mach. Intell.
Deformable templates for face recognition
J. Cognitive Neurosci.
Detecting faces in images: a survey
IEEE Trans. Pattern Anal. Mach. Intell.
Example-based learning for view-based human face detection
IEEE Trans. Pattern Anal. Mach. Intell.
Cited by (0)
About the Author—PEICHUNG SHIH received his M.S. degree with summa cum laude in Information Systems in 2002 from New Jersey Institute of Technology. He is a Ph.D. candidate in Computer Science at New Jersey Institute of Technology. His research interests lie in the field of image and video processing, computer vision, and pattern recognition with applications toward face recognition. His present work includes developing novel face detection systems by applying computer vision concepts and statistical learning theory, and designing robust face recognition models by utilizing color information and genetic computations.
About the Author—CHENGJUN LIU received the Ph.D. from George Mason University in 1999, and he is presently an Assistant Professor of Computer Science at New Jersey Institute of Technology. His research interests are in computer vision, pattern recognition, image processing, evolutionary computation, and neural computation. His recent research has been concerned with the development of novel and robust methods for image/video retrieval and object detection, tracking and recognition based upon statistical and machine learning concepts.