A numeral character recognition using the PCA mixture model

This paper proposes a method for recognizing the numeral characters based o n the PCA (Principal Component Analysis) mixture model. The proposed method is motivated by the idea that the classification accuracy is improved by m odeling each class into a mixture of several components and by performing t he classification in the compact and decorrelated feature space. For realiz ing the idea, each numeral class is partitioned into several clusters and e ach cluster's density is estimated by a Gaussian distribution function in t he PCA transformed space. The parameter estimation is performed by an itera tive EM (Expectation Maximization) algorithm, and model order is selected b y a fast sub-optimal validation scheme. The proposed method is also computa tion-effective because the optimal feature components for a cluster are det ermined by a sequential elimination of insignificant feature due to the ord ering property of the significance among the feature components in the PCA transformed space. Simulation results shows that the proposed recognition m ethod outperforms other methods such as the k-NN (Nearest Neighbor) method, a single PCA model, or the ICA (Independent Component Analysis) mixture mo del in terms of recognition accuracy. (C) 2002 Elsevier Science B.V. All ri ghts reserved.