Study On Face Recognition Based On Improved Eigenface

Human-face recognition is a kind of identity validation technique. It extracts facial features by computer, and compares with images in database using these features. Recently, face recognition technique has made a great progress, accumulated a lot of experience. However, lack of adaptability is a common problem of current algorithms. Many factors such as expression, pose, covering and illumination degrade the recognition performance. So it will be important to excogitate a fast and exact human face recognition method.In this paper, a face recognition method which is based on eigenface and an eyeglasses removal method which is based on image reconstruction are studied.First, the theory of traditional eigenface is studied deeply. Because of the disadvantages of traditional eigenface, some improved methods are introduced, such as image normalization which could decrease the distribution range of image gray parameters like mean and variance, and reduce the influence of illumination, and weighted half-eigenface which could reduce the influence of expression on precision. Weighted half-eigenface is a method that separates the face image into two parts by size averagely: the upper and the lower, and assigns them to different weights, then these two parts are processed respectively with eigenface method.Secondly, in order to reduce the influence of covering such as eyeglasses and overcome the theoretic limitation of weighted half-eigenface, that is the dependence to the character of the upper face image, an eyeglasses removal method which is based on image reconstruction is proposed. It's a method that reconstructs the projection vectors derived from eigenface method, and compensates error for the input image with reconstructed image, then takes the face image without eyeglasses as a new input image.Finally, some experiments validate the practicability of these algorithms, whose samples include Yale face database and some photos collected by digital camera. The results of experiments indicate that the human face recognition algorithm has high accuracy, on condition that face region is detected exactly.