A New Face Recognition System base on MLP Classifier Using SIFT Keypoint
Tong Liu, Sung‐Hoon Kim, Hyon-Soo Lee, Hyung‐Ho Kim · 한국콘텐츠학회 ICCC 논문집 · 2009
This paper investigates a new face recognition system based on an efficient design of MLP classifier using each SIFT (Scale Invariant Feature Transform) keypoint as feature. This feature based method utilizes SIFT's strong robustness to expression, accessory pose and illumination variations. Instead of using integration of keypoints, we use each keypoint as input of one MLP (Multi Layer Perceptron) based classifier. According to the output of classifier, some confusing key points are rejected by using threshold value. Output assigns ID to each distinctive keypoint to form histograms of class index. Finally an index histogram counting method is applied to identify face image. Also the bootstrapping method is investigated to select training images during training process. The performance of proposed system is improved lot in some challenging databases. Experiment Yale face database shows that the best recognition rate reaches 98. 7%.