A robust novel method for face recognition from 2D depth images using DWT and DFT score fusion

Face recognition has been an area of interest among researchers in pattern recognition for the past few decades. Researches in face recognition are basically concentrated on texture based and geometry based features. The main advantage of Face recognition systems utilizing depth information is the availability of geometrical information of the face structure which is more or less unique for a subject. This paper focuses on the problems of person identification using 2D Face Depth data. Most of the face recognition systems are based on reconstruction of 3D shapes from the 2D depth data. But the reconstruction requires much more computation time. Further the use of unregistered 2D Face depth data significantly increases the operational speed of the system with huge database enrollment. In this work, the unregistered. 2D Face Depth data is fed to a classifier in multiple spectral representations. Discrete Wavelet Transform (DWT) and Discrete Fourier Transform (DFT) are used for the spectral representations. The face recognition accuracy obtained when the feature extractors are used individually, is evaluated. Fusion of the matching scores proves that the recognition accuracy can be improved significantly by fusion of scores of multiple representations. Robustness analysis which covers the FAR (False Acceptance Rate) and TRR (True Rejection Rate) is also done. FRAV3D database provided by Face recognition and artificial vision group of Universidad Rey Juan Carlos, Madrid Spain is used for testing the algorithm.