عنوان مقاله
تشخیص چهره (چهره نگاری) با استفاده از شبکه های عصبی RBF
فهرست مطالب
مقدمه
شبکه های عصبی RBF
استخراج ویژگی های چهره
الگوریتم یادگیری هیبریدی
نتایج آزمایش
بحث و نتیجه گیری
بخشی از مقاله
الگوریتم یادگیری هیبریدی
تعدیل پارامترهای واحدRBF فرایندی غیر خطی است در حالیکه شناسایی وزنW(I,j) فرایندی خطی قلمداد می گردد. اگرچه می توان از الگوی گرادیان برای یافتن کل مجموعه پارامترهای بهینه استفاده نمود، اما این الگو عموماً کند عمل کرده واحتمالاً در حد بهینه محلی به تله می افتد.
کلمات کلیدی:
IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 13, NO. 3, MAY 2002 697 Face Recognition With Radial Basis Function (RBF) Neural Networks Meng Joo Er, Member, IEEE, Shiqian Wu, Member, IEEE, Juwei Lu, Student Member, IEEE, and Hock Lye Toh, Member, IEEE Abstract—A general and efficient design approach using a radial basis function (RBF) neural classifier to cope with small training sets of high dimension, which is a problem frequently encountered in face recognition, is presented in this paper. In order to avoid overfitting and reduce the computational burden, face features are first extracted by the principal component analysis (PCA) method. Then, the resulting features are further processed by the Fisher’s linear discriminant (FLD) technique to acquire lower-dimensional discriminant patterns. A novel paradigm is proposed whereby data information is encapsulated in determining the structure and initial parameters of the RBF neural classifier before learning takes place. A hybrid learning algorithm is used to train the RBF neural networks so that the dimension of the search space is drastically reduced in the gradient paradigm. Simulation results conducted on the ORL database show that the system achieves excellent performance both in terms of error rates of classification and learning efficiency. Index Terms—Face recognition, Fisher’s linear discriminant, ORL database, principal component analysis, radial basis function (RBF) neural networks, small training sets of high dimension. I. INTRODUCTION MACHINE recognition of human face from still and video images has become an active research area in the communities of image processing, pattern recognition, neural networks and computer vision. This interest is motivated by wide applications ranging from static matching of controlled format photographs such as passports, credit cards, driving licenses, and mug shots to real-time matching of surveillance video images presenting different constraints in terms of processing requirements [1]. Although researchers in psychology, neural sciences and engineering, image processing and computer vision have investigated a number of issues related to face recognition by human beings and machines, it is still difficult to design an automatic system for this task, especially when real-time identification is required. The reasons for this difficulty are two-fold: 1) Face images are highly variable and 2) Sources of variability include individual appearance, three-dimensional (3-D) pose, facial expression, facial hair, makeup, and so on and these factors change from time to time. Furthermore, the lighting, background, scale, and parameters of the acquisition are all variManuscript received March 29, 1999; revised March 5, 2001 and December 5, 2001. M. J. Er is with the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore. S. Wu and H. L. Toh are with the Centre for Signal Processing, Innovation Centre, Singapore 637722, Singapore. J. Lu is with Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON M5S 3G4, Canada. Publisher Item Identifier S 1045-9227(02)03984-X. ables in facial images acquired under real-world scenarios [1]. As stated by Moses et al. [2], “The variations between the images of the same face due to illumination and viewing direction are almost always larger than image variations due to changes in the face identity.” This makes face recognition a great challenging problem. In our opinion, two issues are central to face recognition: 1) What features can be used to represent a face under environmental changes? 2) How to classify a new face image based on the chosen representation? For 1), many successful face detection and feature extraction paradigms have been developed [3]–[12]. The frequently used approaches are to use geometrical features, where the relative positions and shapes of different features are measured [3], [4]. At the same time, several paradigms have been proposed to use global representation of a face, where all features of a face are automatically extracted from an input facial image [5]–[12]. It has been indicated in [4] that these algorithms with global encoding of a face are fast in face recognition. In [5], singular value decomposition (SVD) of a matrix was used to extract features from the patterns. It has been illustrated that singular values of an image are stable and represent the algebraic attributes of an image, being intrinsic but not necessarily visible. The eigenface approach of describing the features of a face was presented in [6]. The key idea is to calculate the best coordinate system for image compression, in which each coordinate is actually an image that is called an eigenpicture. However, the eigenface paradigm, which uses principal component analysis (PCA), yields projection directions that maximize the total scatter across all classes, i.e., across all face images. In choosing the projection which maximizes the total scatter, the PCA retains unwanted variations caused by lighting, facial expression, and other factors [7]. Accordingly, the features produced are not necessarily good for discrimination among classes. In [7], [8], the face features are acquired by using the fisherface or discriminant eigenfeature paradigm. This paradigm aims at overcoming the drawback of the eigenface paradigm by integrating Fisher’s linear discriminant (FLD) criteria, while retaining the idea of the eigenface paradigm in projecting faces from a high-dimension image space to a significantly lower-dimensional feature space. Instead of using statistical theory, neural-networks-based feature extraction has been reported recently [9]–[12]. The goal of face processing using neural networks is to develop a compact internal representation of faces, which is equivalent to feature extraction. Therefore, the number of hidden neurons is less than that in either input or output layers, which results in the network encoding inputs in a smaller dimen
