Author : Sukant Tiwari and A.K. Shukla

In this paper, we propose a system of face recognition from single sample image, which is robust to pose, illumination, expression and background changes. It is fast to deploy and performs at a good rate. In the method, we extract the normalized face from input image using the Viola-Jones object detection method, eliminating the background effects. The face is divided into regions. and observe the different kinds of patterns observed for each face region. Based upon the patterns observed, a codebook for each region is generated. Illumination variations are overcome using the robust and popular image encoding and compression function, the DCT, is used for encoding the data. Histograms of the feature patterns provide robustness against pose variations of the face features. Performing weighted sum of the differences and having low weight for the mouth region incorporates robustness to expression changes. The distinctiveness of each face is taken into account, thus making the system similar to human beings. The system is robust and fast in a similar manner.

Keywords: Face recognition, Single sample image, Violoa-Jones object detection, DCT

Article published in International Journal of Current Engineering and Technology, Vol.5, No.3 (June-2015)