Face Recognition using Multi-class Support Vector Machines mixture Triangle-based approach

碩士 === 國立臺北大學 === 資訊管理研究所 === 94 === Face recognition has researched for thirty years, in recent ten year, face recognition improved with few kernel technology. Example: Feature-based、Template-based、Neural Network-based. In the past, recognition rate will changed with environment changes, and slow d...

Full description

Bibliographic Details
Main Authors: Jimmy Wu, 吳建樺
Other Authors: Lin C.S.
Format: Others
Language:zh-TW
Published: 2006
Online Access:http://ndltd.ncl.edu.tw/handle/39099930590432240600
Description
Summary:碩士 === 國立臺北大學 === 資訊管理研究所 === 94 === Face recognition has researched for thirty years, in recent ten year, face recognition improved with few kernel technology. Example: Feature-based、Template-based、Neural Network-based. In the past, recognition rate will changed with environment changes, and slow down the recognition speed. Recent researches focus on accuracy and speed of recognition, except that, the researches focus on environment changes too. Example: Light、illuminations、angle of human face and etc. Support Vector Machines is a new generation algorithm based on Statistical Learning Theory (STL). Support Vector Machines is a popular technology and applies in many domains. It applies in texture classification、signature recognition、image classification、bioinformatics and etc. In this paper, we use Triangle-based approach to detect face region in a graphic, retrieve eyes and mouth’s position that is a isosceles triangle, and crop face region based on two eyes’ distance. The crop face region is the training data we need. The training data will transfer to multi-support vector machines, the SVM is our kernel technology for face recognition. We use to database to testing. The test 1, we use database created by ourselves, it’s includes 30 people and each people have 20 images, totally for 600 images. In test 2, we use CMU PIE database, includes 68 people, each people have 60 images, totally for 4080 images. Result in test 1 shows the same data for SVM and PCA, accuracy of SVM is 90.83%, but PCA only have 77.68%. In test 2 shows accuracy of SVM is 94.0441% in CMU PIE database, and can also recognition for facial expression changes.