Identity authentication system using face recognition techniques in human-computer interaction

In human-oriented intelligent human-computer interaction, the key context information can be provided by computer vision techniques, such as user identity, position and action. To sufficiently support user vision information, the first problem is human face detection, localization, and recognition. In this paper a robust identity authentication system using face recognition techniques under uncontrolled lighting condition is implemented. To reduce the effects of illumination variation, three preprocessing methods are compared, and Gamma correction, Difference of Gauss filtering (DoG) and contrast equalization are adopted in our system. The original face image is divided into several non-overlapping regions and their corresponding histograms of Local Ternary Pattern (LTP) are extracted as face features. Then an improved AdaBoost method is used to reduce the dimension of feature vector. Final face identity is obtained by the nearest neighbor classification. Experimental results on FERET and CAS-PEAL databases show that the performance of our system is better than that of the Local Binary method when the illumination is changeable.