Early-Vision-Inspired Method to Distinguish between Handwritten and Machine-Printed Character Images Using Hough Transform

Abstract

This paper proposes a method to distinguish handwritten character regions from machine-printed ones using Hough transform (HT). The Gabor filtering in the human early vision realizes a type of Fourier transform (FT). Previously we proposed a FT-based distinction method successfully. However, we noticed simultaneously that the HT, instead of FT, may extract more features when we deal with characters which are regarded as piles of line segments. Experiments show that HT-based method, in combination with real-space features, achieves higher accuracy than the FT-based method. At the same time, the total calculation cost is found lower.