Aggregation of Gabor wavelets and curvelets with PCA for efficient retrieval of face images
S. S. Shylaja, Senthil Kumaran Vijayalakshmi Natarajan, K. N. Balasubramanyamurthy, K. G. Abhijit, Jayashree Diwakar, S. Mohammed Saifulla · 2010
Face Recognition is one of the complex applications the biometric industry has seen. Several algorithms have been used to solve this hard problem. Here a novel method has been proposed based on Gabor wavelets and curvelet transforms. Gabor wavelets exploit vital properties of face such as spatial localization and orientation selectivity where as Curvelets extract edge and directional information from the face. The method extracts features by aggregating Gabor wavelet at single scale and 4 different orientations and Curvelet coefficients at single scale and 8 orientations. These aggregated features cover most of the information of the face subspace. But the coefficients thus obtained still form a larger subspace for classification. Therefore another level of reduction in dimension has been achieved through Principal Component Analysis (PCA) which yields a much reduced basis vector. The vectors are then classified using Mahalanobis distance measure to recognize a face as known or unknown. The performance of the individual methods and their aggregate has been tested on standard YALE database consisting of 217 training images and 400 probe images. The Gabor and Curvelets although individually performed well, the combination outperforms the individual performance with a calculated accuracy of 93.5%.