A Method for Handwritten Digits Classification

Kernel PCA, as an unsupervised learning method, is a nonlinear extension of PCA for finding projections that give useful nonlinear descriptors of the data. In the application of handwritten digits classification, kernel based algorithms are indeed highly competitive on a variety of problems with different characteristics. In most real-world pattern analysis tasks, kernel-based can cut the correlative features and prefer discriminable, reliable, independent and optimal features to reduce the complexity of the classifier.