Handwritten Digit Recognition using Slope Detail Features

In this paper, new features called Slope Detail (SD) features for handwritten digit recognition have been introduced. These features are based on shape analysis of the digit image and extract slant or slope information. They are effective in obtaining good recognition accuracies. When combined with commonly used features, Slope Detail features enhance the digit recognition accuracy. KNearest Neighbour (k-NN) and Support Vector Machine (SVM) algorithms have been used for classification purposes. The data sets used are the Semeion Data Set and United States Postal Service (USPS) Data Set. For the USPS Data Set an error rate of 1.3% was obtained, which has been found to be better than any reported error rate on the said data set. General Terms Histogram Features, OCR, Pattern Recognition, Artificial Intelligence.