Handwritten digit recognition with kernel-based LVQ classifier in input space

In this paper, a kemel-based LVQ classifier in input space is proposed to recognize handwritten digit. Classical Learning Vector Quantization is performed in the input space through Euclidean distance, but it doesn't work well when the input patterns are highly nonlinear. In our model, the kernel method is used to define a new metric of distance in input space so we can get a direct view of the clustering result. At last, we test our model by handwritten digit recognition using MNIST database and it obtains better recognition performance than traditional LVQ.