Handwritten Character Distinction Method Inspired by Human Vision Mechanism

摘要

Human beings are capable of distinguishing a variety of texture almost at a glance. By modelling the mechanism, we will realize a flexible texture analysis. We propose a new technique inspired by such human early-vision ability to distinguish handwritten character regions from machine-printed regions in document images. In the technique, we evaluate the two-dimensional power spectrum to extract feature values that reflects fluctuations unavoidable in handwritten characters. Experiments show that a certain feature value of handwritten characters is often larger than that of machine-printed characters. We generated a map obtained by superimposing the feature value on the document image. The map showed that our proposed method is useful to distinguish handwritten character regions from machine-printed character ones.