Authors : R. Mathu Soothana, S. Kumar and K. Muneeswaran
Abstract: An optimal random image component selection algorithm using greedy approach is presented in this study. The proposed algorithm when evaluated with hierarchical ensemble classifier has an enhanced recognition rate with large variations in illumination, pose and facial expression. In the proposed technique, features are extracted from the optimal random image components which are then projected to the multiple discriminant analysis and kernel discriminant analysis subspace for solving linear and non-linear problems. The number of local image components is varied from 1-40 and by means of optimality checking, it is observed that at least 10-20 image components are sufficient for reasonable recognition. The FERET and ORL face datasets were used to generate the results. The method has achieved 99.44 and 100% recognition accuracy and 82.5 and 99.64% recognition accuracy on FERET and ORL datasets for 30% training, respectively. This is a considerably improved performance than one attainable with other standard methodologies described in the literature.
R. Mathu Soothana, S. Kumar and K. Muneeswaran , 2013. ORICS Based Kernel Discriminant Analysis. International Journal of Soft Computing, 8: 223-230.