Handwriter Identification Using Quantitative Features Extracted from Character Patterns

Abstract

Quantitative freatures extracted from hand-written character pattern were applied to handwriter identification, and the efficiency for hand-written character recognition was evaluated. Three features including local direction contributivity (LDC) feature, directional element (DE) feature and weighted direction index histogram (WDIH) feature of each sample character pattern collected from 20 writers were enforced to learn by the hierarchical neural network (HNN) in order to apply the tool for handwriter identification. Correct identification on DE and WDIH features through one character by HNN ranged from 82.5 to 85.5%, While that by visual inspection of expert examiners was 85.5%. Moreover, this tool performed the perfect identification using the combining sum of HNN's output values for sample characters. These results support that the application of feature extraction on hand-written characters using HNN is significantly effective for handwriter identification.