Abstract
This paper proposes and analyses the pyramidal Gabor eigenface (PGE) algorithm for face recognition. It can be realised directly in the spatial domain by using one-dimensional (1D) filter masks to extract the Gabor facial features. This is in contrast to the two-dimensional (2D) Fourier implementation that must be applied in both frequency and spatial domains in the general 2D Gabor-based facial feature extraction methods. Eigenface decomposition is then used to further reduce the redundancy of these face features. Because, Gabor features are characterised by strong spatial locality with scale and orientation selectivity, they can cope with variation problems due to illumination, facial expression change and orientation. Eigenface cannot handle such variations. The analysis of the algorithm and the experimental results using AT&T (formally Olivetti) face database show that the cost of the algorithm and the number of Gabor features in the PGE algorithm are significantly lower than the ones in the classic 2D Gabor wavelet-based methods and have better recognition results.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call