This chapter proposes an optimal cost security approach for the current and emerging trends in the Engineering centric IoT applications that offer an optimized infrastructure and human safety through bimodal deep face recognition. Human face determines the person identity that reveals information like age, gender, emotions, attractiveness and others. Face recognition attracted researchers to enhance its performance because of its potential usage in several commercial, law enforcement, government and video surveillance applications in which individuals perceive each other. In this chapter, authors propose a new secured optimal cost approach for deep face recognition based on feature level fusion of bi-features extracted through unsupervised deep learner, Autoencoder and Local Binary Patterns (LBP) respectively. The dimensionality of fused feature map is reduced and protected through Forward Error Correction (FEC) technique. An efficient optimal cost region matcher (OCRM) is accomplished with Canny edge detector to maximize the face recognition accuracy. OCRM uses north-west corner rule of the transportation problem that fulfills the Monge property. The experimental results demonstrate the superiority of the proposed face recognition system over unimodal systems (Autoencoder and LBP alone) when tested on ORL and Real face datasets with OCRM matcher which is interfaced through diverse IoT applications.
Top1. Introduction
A biometric system is fundamentally a pattern recognition system that automatically identifies individuals based on their physiological and behavioral characteristics. (Jain, A. K.et al.2007) demonstrated that physiological biometrics use measurements from the human body, like face, fingerprint, iris and retina, while behavioral biometrics use dynamic measurements based on human action, like signature, voice, keystroke and gait. The striking feature of face biometric considered in this work when compared with diversified biometric traits is that facial images can be acquired from a distance without user cooperation. This feature is particularly beneficial for security and surveillance purposes. Furthermore, data obtained from hand and finger images will be inoperative if the skin is injured. Correspondingly, the data acquisition from retina, iris modalities necessitates costly devices which are easily susceptible to body motion. Voice recognition is vulnerable to environmental conditions and sound-related variations on a telephone line or tape recording while signatures can be easily spurious and alterable. However, facial data can be captured with an ease through low cost rigid cameras. The various approaches to extract face features are grouped into three categories, namely, geometric, holistic and hybrid approaches. A few researchers in the literature like (Xie, J et al. 2012; Boureau, Y. L., and Cun, Y. L. 2008; Glorot, X et al. 2011; Krizhevsky et al. 2012) proposed various machine learning feature extractors such as deep neural networks, deep belief networks, deep convolution neural networks, learning deep face are tested and illustrated by (Taigman et al. 2014; Fan, H.et al. 2014) and Zhu, Z. et al. 2013) explained how to preserve face space with deep learner. The features extracted from these learning algorithms reduce the input dimensionality and reconstruct the feature vector to minimize the intra-class variations in face biometrics is well explained by (Bengio, Y. et al. 2013). (Bronstein, A. M. et al. 2007) demonstrated an expression invariant representation of a face can also be used for recognition through canonical form. (Besides, Abaza, A. et al. 2014) illustrates that the image quality factors like brightness, focus, illumination, contrast and sharpness are significant in face recognition systems. Further, face recognition is absolutely non-invasive and does not convey any health hazards. It has turned out to be progressively vital owing to quick advances in image acquisition gadgets (surveillance cameras, camera in mobile phones), accessibility of extensive face images on the Web, and raise in demand for higher security is explained by (Jain, A. K., and Li, S. Z. 2011).