Research Of Face Recognition

The automatic recognition of human face is one of the most attractive and challenging problems in fields of computer vision and pattern recognition. It is the aim of many scientists who work in computer science field to make robots have human's intelligence and remember and recognize person just like what human does. With the development of society and the improvement of science and technology, it is urgently needed for convenient and reliable automatic-status-discrimination. So face recognition resumes to be the highlight of machine intelligence research.To the problems of face detection and face recognition, respectively, this dissertation reviews the existing theory and algorithm. Furthermore, it proposes new approaches. The main content of this dissertation is summarized below:Face detection is an important part of automatic face recognition system. We present this question in details in chapter two. In color images, the skin-color is important information of human faces. So the skin-color-liked area can be divided from the context by the skin-color-model in HIS color space. To the noise, singular point, non-face area and the division of face area, this dissertation proposes filter, segmentation by clustering and area combination methods respectively; to the detection of oriental faces in color images and complex context, the author combines the skin-color model based face rough detection and the ellipse based face location methods to one, then verify the candidate face by abstracting face features. Experiment indicates that this algorithm is feasible and highly efficient.In chapter three, the author presents the two main questions in face recognition respectively: feature abstraction and classifier designing. For face feature abstracting, Independent Components Analysis method is more efficient than Principal Components Analysis in making use of high rank statistic information. Based on the formers, this dissertation efficiently selects the face features abstracting using ICA. With no decline of recognition rate, the feature dimension is reduced, so the course of recognition is accelerated.Support vector machine pattern recognition method is based on VC dimension theory, adopting the SRM principle and considering training error and the generalization ability, which has shown many special advantages in dealing with small samples, non-linear and pattern recognition in high dimension. This dissertation proposes a new method. ICA/SVM combined method for face recognition. Compared with other methods of the experiments on ORL face database, this method is prior to them.