| dc.description.abstract | In the digital era, we have to identify everyone to protect individual rights and information for bureau of entry and exit, network transactions, entry access control at home or corporate, etc. Traditional identity authentication methods have been insufficient for our needs, for example ID card, seal certification, etc. Currently, biometric technology is the most convenient and secure way on fingerprint authentication, face authentication, retina authentication, etc.
In biometrics, face authentication not only has the highest acceptance of general public, but also the wide range of applications. It has three advantages 1) no need to touch 2) does not require user involvement or cooperation 3) the biometric collection process will not have any discomfort. Therefore, we propose the novel local transform feature: local gradient patterns (LGP) [1] and hybridization feature [1] that combines LBP, LGP by means of the AdaBoost method in face authentication (face verification). It will transform the face images into the LGP, and hybridization feature images. Then face authentication model was trained base on feature images and AdaBoost algorithm.
LGP will calculate the neighboring gradient of a given pixel and its average of neighboring gradient. Then the average of neighboring gradient was set to center pixel. If neighboring gradient is greater than center pixel, LGP assigns one and zero otherwise which makes the local intensity variations along the edge components robust. According to the best local transform feature having the lowest classification error, LBP and LGP feature are fused by AdaBoost for hybridization of local transform features. This hybridization makes face detection performance robust to changes in global illumination by LBP, local intensity changes by LGP.
In the actual experiments, we utilize a different number of positive samples and negative samples training face authentication model and the accuracy under various sample numbers are demonstrated. In addition, CMU PIE database is applied in our experiments. Experimental results show that our LGP and hybridization could improve accuracy and reduce the risk of counterfeit identity in terms of face authentication. | en_US |