A Face Recognition System Based on Spatial Distribution of Local Texture Feature and Support Vector Machine

• Main Authors: Kun Lu, 呂昆
• Other Authors: Yu-Keun Ho
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/47088844680040968809

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 97 === Recently, the research of face recognition is more important in biometric authentication field. Though the biometric characteristic of finger print, palm, iris are commonly using in ID authentication, face recognition is non-contact mechanism. We expect that face recognition would be applied to identification extensively. In Multi-Scale Block Local Binary Pattern (MB-LBP), the computation is done based on average values of block sub-regions, instead of individual pixels for extracting the texture feature. MB-LBP not only preserves the advantage of Local Binary pattern (LBP) but also encodes macrostructures of image patterns. Support Vector Machine(SVM) is one of the efficient classifiers and be applied to the classifier of face recognition popularly. From the base of MB-LBP feature extraction. we propose a face recognition system based on spatial distribution of local texture feature and support vector machine. We extract the spatial distribution of feature from Gaussian Mixture Model(GMM). Using the parameters of GMM we get, we can construct good features which can robust against the distortion of face image like shift, resize, and rotation. In the experiment on using ORL facial databases, the proposed system did against the image variation (shifting, resize and rotation).