Simulation-level implementation of face recognition in uncontrolled environment

Abstract

Humans rely on their visual ability to perceive and analyze the visual world. This ability if given to computers can yield scientific and relevant details about the things around us. Recognizing face under uncontrolled environment is still a challenging task, especially under head pose variation. Initially, the gesture containing test input face is estimated; it is then transformed into a reference pose which is already learnt. We have analyzed many existing available techniques for recognizing face for different gestures and proposed a novel system using a combination of support vector machine (SVM) and K-nearest neighbor (K-NN). The system is implemented using MATLAB and simulated using Modelsim. It has been tested on publicly available face databases under varying head poses. The proposed model was trained by randomly selecting 10 images, each of 12 unique individuals, thus 120 images were used for training. The system was then tested by considering 93 different poses of all the 12 unique individuals, thus 1116 images were used for testing. A demo code, along with train images, test images, and results obtained from the proposed system can be downloaded from http://goo.gl/S5ofYR.