An Efficient Feature Set for Handwritten Digit Recognition

Abstract

In this paper, we have discussed the new feature set for handwritten digit recognition. The feature set is very small and simple. The features are extracted using pixel counting technique and contour following techniques. No preprocessing except binarization and thinning is done on the data. The purpose of this paper is two fold. Firstly, we explained by experiments that slant invariant and size invariant features help in developing general software, which is free from some of the pre-processing steps. Secondly, we confirm that pixel counting technique is very useful for deformed images than contour following technique. SVM and Tree classifier are used for classification.