Human Face Recognition System

• Main Authors: Chih-Ho Lin, 林志和
• Other Authors: I-Chang Jou
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/04537729754713594129

碩士 === 國立高雄第一科技大學 === 電腦與通訊工程所 === 91 === This study proposes an face recognition system﹒This system has been developed many years﹒Many researcher also proposed many different method﹒Because there are many factors unable to overcome so recognition system have not a good result all the time﹒For example﹐the problem are facial expression﹐varying lighting﹐different quality of capture device﹐face feature extraction etc﹒The issue in this study is using color scenery images to develop recognition system﹒This system can be divided into three parts： First part is face detection﹒Because HSI color system are not sensitive to the intensity variations﹒Hence﹐the RGB values of pixel in the input image are first transform into HSI color space﹒The every pixels in the image will be mapped onto one point in the HSI plane﹒If the corresponding point lies on the specified zone﹐then the pixel will be labeled as a skin pixel. The specified zone was statistic skin color range lies on the HSI space﹒ISO DATA(Iterative Self-Organizing Data Analysis Technique Algorithm)must be applied to separate the skin pixels into several clusters﹒We could exploit organ’s location on the face to decide every clusters whether was human face or not﹒ Second part is feature segment﹒In order to distinguish from different faces﹐we have to find out every unique face’s feature﹒We must segment image before find out feature﹒In order to find out which one are organs that we want﹒The invariable features(eye﹐nose﹐lip) on the face have to be exploited﹒In this thesis﹐“Eigenspace Projection”was applied to project eye﹐nose﹐lip and face’s image on the eigenspace﹐then many feature values are gotten﹒ Third part is verification system﹒This system is implemented based on the “Plastic Perceptron Neural Network”﹒This network is more suitable for classification especially and it can parallel and distributed process different class. Network has not overall retraining when you replace patterns or add new ones.“Plastic Perceptron Neural Network”has more elasticity than conventional“Black-Propagation Neural Network”.