Face Recognition–Oriented Biometric Security System
Geetika Singh, Indu Chhabra · 2017
Analysis of faces has always been appreciated long in delivering effective face biometric devices. For a long time, concentration has been focused on those concerns which really contribute to the best recognition of human beings under various real life practical situations where the face to be identified may vary due to their poses, gestures, and various illumination situations. This chapter explores the well-known statistical techniques of feature extraction to recognize the face pictures for poses and lighting conditions. Due to the exceptional properties of image reconstruction and invariance to rotation and noise, geometric moments have been well analyzed. In the present study, the Zernike moments and Polar Harmonic Transforms techniques are well explored and investigated through Neural Networks and Support Vector machines. Due to wide variations in the numerical values of the traditional moments, these techniques are able to capture only prominent features changes. To portray the fine human features, a mechanism is devised where the traditional methods of moments and Polar Complex Exponentials are evaluated to get only those distinct values which are really contributory to recognize facet characteristics. This chapter elaborates on the devised dimensionality reduction-based optimal Feature Selection. The optimal feature set, provided by the existing Zernike moments and Polar Transforms, is reduced by the devised method of dimensionality reduction to provide better recognition accuracy. The improved performance of the proposed Feature Selection is tested against the faces of different poses and light variations. After validating the proposed technique for standard databases of ORL, Yale, and FERET, it is implemented for a self-created database to deliver a secure face biometric system. The resulting authenticity is justified through the better reconstruction ability at reduced order with reduced feature sets required to recognize the same face. The significance of the optimal selection is further justified by citing the various performances and technical reasons. The chapter is divided into various sections from introduction to gaps in the literature for the existing techniques and how these gaps are filled through the proposed feature selection by its implementation and validation through independent as well as hybrid classifiers for standard and self-created face databases.