A Perturbation Technique for Solving the Handwriting Variation Problem in On-Line Handwritten Chinese Character Preclassification

Abstract

In this paper we propose a method based on the perturbation technique to handle the variations in handwritten Chinese characters. The basic concept of the perturbation technique is to recover the possible erroneous stroke features (including stroke types, stroke spatial relation, stroke number, and stroke merging ) with a hope that the resultant character features can find a correct match in the database. The perturbation technique can also be applies to the reference samples in the database. Experiments have been conducted on a test set containing 914 different characters. Each character is written by 61 different persons with no rigid writing constraints. The correct preclassification rate is 81.42% without perturbastion and 96.49% with perturbastions applied to both the input samples and the database.