Face Recognition based on Radial Basis Function and Clustering Algorithm Yuanfeng Gao ECE 539 Project Fall 2008Abstract This project consists of two parts. The first part is a general review of the previous and current research on human face recognition, including initial motivation, approaches, major problems and solutions, etc. The second part propose a new method for learning of radial basis function (RBF) neural networks which is based on subtractive clustering algorithm(SCA) and its application to face recognition. Experiments on face recognition using ORL database show feasibility of the method. Results present that RBF neural networks classifier using proposed algorithm is more precise and faster than corresponding one using general K-means clustering algorithm.目录 1. Introduction ....................................................................................................... 4 2. Face localization methods .................................................................................... 5 2.1 Knowledge based methods .......................................................................... 5 2.2 Feature Based Methods ............................................................................... 6 2.2.1 Moment Invariants ............................................................................. 6 2.2.2 Differential invariants ........................................................................ 7 2.2.3 Fourier Descriptor ............................................................................. 8 2.2.4 Summation invariants ........................................................................ 8 2.2.5 Dyadic wavelet invariants ................................................................. 9 3. Two-Dimensional and Three-Dimensional Face Recognition ........................ 10 3.1 Overview of 3D Face Algorithms ............................................................. 10 3.2 Major Challenges and Possible Solutions ................................................. 11 4. Image Resolution and Face Recognition ......................................................... 12 5. Neural network approaches to Face Recognition ............................................ 13 5.1 RBF neural networks model .................................................................. 13 5.2 Subtractive Clustering Algorithm (SCA) .............................................. 14 5.3 Applying SCA to RBF Neural Networks .............................................. 15 5.4 Application in Face Image Recognition and Experiment ..................... 15 5.4.1 Preprocessing and Feature extraction of Face images ................ 15 5.4.2 Networks Parameters ....................................................................... 16 5.4.3 Experiments and Analysis ..... .......................................................... 17 5.4.4 Conclusion ....................................................... ................................ 18 1. Introduction New information technology and media have always attracted the attention of the public and has changed the modern world greatly. Among them, more effective and friendly methods for Human-Computer interaction have been developed, and have distinguished themselves from traditional methods that rely on devices such as keyboards, mice and displays. Furthermore, with the decreasing video image acquisition cost that results in the ever-increasing performance/price ratio, computer vision systems have been deployed in desktop and embedded systems [1]. As a result, research of human face recognition by artificial intelligence that once been limited by restraints of apparatus, has now been rapidly expanding over the last decade. It has attracted much attention though this study has already been worked on for more than twenty years by neuroscientists, engineers, and psychophysicists. Face recognition has found a wide range of applications, and there has been a growing interest in machine recognition of faces due to potential commercial application, such as secure monitoring, person identification, law enforcement, object identification and tracking etc. The newest image processing software, “Iphoto 09 ™” by Apple®, could detect different faces from photos and organize photos by faces in the picture which made sorting photos much easier than before. Over no more than two decades, sales of identity verification products rocketed from $100 million in 1994 [2] to $500 million in 2007 [3]. Face recognition is one of the hottest research topics nowadays and numerous face recognition approaches have been presented so far. The first step of any face recognition system is to detect the locations of different parts of face in the image. For instance, the positions of forehead, nose, cheek, mouth, chin, etc. However, this is a rather challenging task and has drawn a lot of attention. Reasons basically fall within two areas: 1. the environmental reason includes factors such as the variety of scale, location, illumination conditions, image resolutions, and angles. 2. The other reason is related to the human in the image, such as pose (frontal or profile), different expressions, and occlusion. After successfully overcoming the difficulties listed above, the next step usually involves data processing and storage. Depending on the approaches, different “name cards” for each person would be created, and could be used as identification of the same person in the future. Different approaches would result in different recognition rate, ability to reduce noise, and processing time. However, evaluations such as the Face Recognition Vendor Test (FRVT) 2002 [4] also made it clear that the current state of the art in face recognition is not sufficient for the more demanding applications nowadays. The vast majority of face recognition research use normal greyscale intensity images of face, which are referred to as “2-D” images, in contrast to “3-D”. Combinations of “2-D” and “3-D” multi-models are been used in recent research and the term “multi-modal biometrics” refers to these kinds of multiple imaging modalities.2. Face localization methods In this section, we review several existing techniques to localize faces from a single intensity or colour image. 2.1 Knowledge based methods Human could recognize different faces based on the normal knowledge of what constitutes a typical face. These
View Full Document
Unlocking...