An Adaptive Illumination Day And Night Face Recognition System

• Main Authors: Long-Chain Chang, 張容銓
• Other Authors: Yu-Kun Ho
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/16494116497817005024

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 96 === Human face recognition, which is a non-contact mechanism, is an important research area in image vision, and safety monitoring skill by human face identification is becoming a part of high technology living. However, light source is an important factor affecting human face recognition which differs according to the variation of light source. Thus, to correctly obtain human face images in different light environment, for example day or night, without affecting the recognition of the human face image is a valuable subject. Usually images are formed by light reflected by objects. Different images pixels amplitude show different colors or different gray scale distribution, thus non-uniform images are presented. This paper is to use block based local contrast algorithm[9] to solve the problem of non-uniform light source, therefore a human face recognition system can be built as used in a house door monitor system. In human face recognition system proposed in this paper, firstly, for images under different light sources, human face images are processed by Local Contrast Enhancement. Local threshold method is used to reserve important gray scales information lower than threshold, and those gray scales higher than threshold are clipped by Histogram Clipping to acquire important feature information of human face images. Then local contrast extension and enhancement method is used to obtain human face images immune to light sources variation. As a result, in different light source environments, human face feature information can still be clearly obtained. Methods of fetching features of human face make use of Principal Component Analysis (PCA), The PCA method is mainly to get the eigenvalues and eigenvector. PCA dimensions of original image data are reduced and those image data are recognized in eigen space. Finally, these obtained eigenvalues are compared with eigen parameters stored in human images database by Euclid method to recognize human faces. The human face recognition system adaptive to day or night, through experiments of four human face images under different luminance, by local threshold method, Histogram Clipping method, local contrast enhancement method, and Principal Component Analysis features for recognition, It been proven effectively improves the difficult problems of human face recognition in different luminance environment. And has good recognition rate.