Research On Face Recognition Based On The Uncorrelated Optimal Discriminant Transformation

The recognition of human face is an active subject in the fields of computer vision and pattern recognition,it has numerous commercial and law enforcement applications,especially in video surveillance.The primary task of face recognition at hand,given still or video images, requires the identification of one or more persons using a database of stored face images.Feature extraction is the main problem of face recognition,the feature should ensure itself has representative attribute, more information, less redundancy and some steady attribute resist interference.Algebra feature extraction is a common method of face feature extraction,the extracted feature not only is satble when it depicts a image sample,but also has many important peculiarity as no alternation when reversed, ratated and moved.This paper analysised and researched some existing algebra feature extraction method such as KL transformation, singular value decomposition, Fisher linear discriminant and optimal discriminant transformation. On the base of this,we expound a method which named uncorrelated optimal discriminant transformation to accomplish the face recognition.Uncorrelated optimal discriminant transformation is a improvement of Fisher linear discriminant and optimal discriminant transformation. For the correlation when using method of Fisher linear discriminant and optimal discriminant transformation,feature vectors have a lot of redundancy and the true useful informations included are not as much.To compare with the two means, uncorrelated optimal discriminant transformation not only has much more efficiency, but also has much more validity because of it could erase correlation of featurevectors.In order to accelerate feature extraction and recognition,we make progress like this: use KL transformation do the frist feature extraction,and use singular value decomposition when calculate matrix to decrease calculation; use uncorrelated optimal discriminant transformation do the second feature extraction on the first extracted feature vectors, so we could derease dimension of vectors and erase correlation.We do laboratory on four methods that involveed before,the result shows,improved system has a considerable progress on correct ratio and speed.