In this paper, we present a method to improve the performance of handwritten digit recognition. Our algorithm rotates images and classfies them together with the original image to minimize the effect of undesirable variations in data. For the task, we...
In this paper, we present a method to improve the performance of handwritten digit recognition. Our algorithm rotates images and classfies them together with the original image to minimize the effect of undesirable variations in data. For the task, we build classifiers by noting that some pairs of digits are more likely misclassified than others pairs because of their similarity. We evaluate the proposed algorithm on MNIST hand written digit data set.