Face Identification from Unconstrained Settings and Occlusions

D. Silambarasan, Anju Mary Thomas

Blur, Different Pose, Non Uniform Illumination, Occlusion, Transformation Spread Function.

The images taken in the ordinary mobile phone cameras are often taken in unconstrained settings. Because of this the identification of face will sometime become very difficult. The paper presented below try to overcome the above conditions of identifying the face from non-uniform illumination, pose, blur and also partial occlusions. Blurred face is modelled as convex combination of transformed focused gallery images .This is extended to those images with illumination variations, which are forming a bi-convex set and finally the variations in pose are also included. For handling the occlusions occurring in the images, the basic characteristics of the human face is taken into consideration. The robustness is increased by the transformation spread function and local binary pattern is added to it that can identify the person even with 70% occlusion in his photograph.