A Topological Structure Construction Approach and its Application in Off-line Handwritten Digit Recognition

Abstract

In searching for an adequate feature extraction approach for pen-like smart scanners to scan and recognize handwritten digits, the authors propose a topological structure construction approach to help extracting features served for recognition. Unlike most of the feature extraction approaches that simply focus on the pixels constituting the visual images of digits, the proposed approach tries in a different direction, that is, adding some lines of pixels to the images and counting the topological structures in the newly formed images. Experimental results show that for samples collected from students' assignments, 9 features is enough for a 93.5% recognition rate, with the help of a suitable classification tree.