High performance quadratic classifier and the application on pendigits recognition

A nonconvex quadratic classifier is proposed for pattern recognition. The classifier is obtained by solving a second-order cone optimization problem on the training data set. Numerical results are presented to compare this classifier with the Gaussian classifier and ĸ-NN classifiers. Regarding to the application of hand written digits recognition, the computational result shows that the proposed quadratic classifier always achieves highest correctness in the testing stage although it takes the longest computational time in the training stage.