Real Time Detection and Recognition of Faces for Biometric Application

Abstract

Biometrics refers to the application of analyzing biological data statistically. Phys- iological properties change from person to person and so they are difficult to forge and are thus highly secured. There exist several biometrics system like finger prints, iris, palm, heartbeat, face, voice and ear geometry. Among all the existing biomet- ric systems facial recognition is one of the most accessible and universally accepted systems. The thesis focuses on identifying different approaches to face recognition in or- der to find a robust, effcient and accurate algorithm which can be implemented in real time for the purpose of attendance system. Holistic as well as artificial intel- ligence based approaches are referred in order to get the best possible outcomes in terms of computation time and accuracy. Eigenfaces, Local Binary Patterns, Fish- erfaces based face recognition techniques are tested and implemented in real time. Comparison of state of art algorithm is performed on custom dataset and readily available dataset from online sources to check the recognition accuracy when sub- jected to outdoor enviornmental condition like Illumination variation and Occlusion. An approach towards recognition of occluded faces is also referred and implemented on real faces using a variant of Local Binary Pattern and Local Phase Quantization to improve the accuracy rate in presence of external lightning and occlusion condi- tion. The proposed system is implemented in real time on Raspberry Pi 3 board and the accuracy achieved under normal condition is approximately 94%.