Face Recognition Research Based On Independent Component Analysis

Face recognition research is the study about Pattern Recognition, Digital Image Process, Computer Vision and the other relative subjects. It has been a popular research subject today for it's value of applications. It has been widely used in the fields of safety department, ID recognition, television meeting, digital inspection, and so on.Face recognition is the technology of computer analysis about the facial image study which gets information from the effective recognition to "identification" .Face recognition technology is contacted with pattern recognition, image processing, computer vision, neural networks and so on ,as well as the psychology, physiology and many other disciplines, is a typical multi-disciplinary cross study field. Face recognition system can handle the changes in facial images. As we all know the face image will be interfered by age, mood, light, hairstyles and many other factors.So the automatic face recognition is a challenging work of the future.The face recognition system is mainly composed of five modules: Face positioning module, pre-module, training module , feature extraction module and the identification module. The positioning module is to get the location from the original image of the face and make the location data into a function. In order to complete the task of training and recognition, we used pre-module which is consisted of the size of uniform, gray uniform, the elimination of features. Face training module refers to the face image database with training parameters and the algorithm in the identification module to get the informations. And it is the core of the face recognition research. Feature extraction module is to complete extraction of facial features. How to get the stable and effective characteristics by face recognition system is the key to success. Identification Module finished the last task of recognition.Face recognition is a kind of technology that it can complete to recognize face by processing and analyzing the face image in the computer, and extracting the representation of face image from the processed image. Currently, there are many different face recognition technologies, and they have themselves advantage and disadvantage. Comparing those popular face recognition methods, we obtain the face representation by Independent Component Analysis (ICA), and recognize sort those face representation using Euclidean distance method. ICA is the digital signal process that comes from cocktail party question and applies to solve blind signal separate, and it is a generalization of Principal Component Analysis (PCA) who is classical linear transform technology. There are two steps to complete face recognition based on ICA. Firstly, we need to complete the face image's preprocessing of meaning, K-L transform, and whitening. Secondly, in order to construct a new mul-dimension for obtaining facial projection coefficients of facial representation information, we can extract the face representation and optimize and select them, and complete recognition work through comparing coefficients Euclidean distances. The first steps propose is that reducing the image dimensions and calculation quantity of extracting the face representation information for the next step. And the second is that extracting face representation, then comparing to the tested face image representation information, the recognition work is finished.The paper has tested the face image set of ORL library and Cohn-Kanade library that the international usually use, the set has 80 people, and each people have 10 different facial images. The right recognition rate is 94.5% for ORL face library, but 99% for Cohn-Kanade library, the experience data result shows that the way of ICA combined with Euclidean distance method has the better purpose for face recognition, and it has rather stronger self-adaptability, much faster recognition speed, much higher recognition rate and much better robustness.