Boxing code for stroke-order free handprinted Chinese character recognition

Abstract

Many Chinese input methods rely heavily on the dictionary stroke orders of Chinese characters. Any deviation from the standard stroke orders in these methods would result in incorrect inputs. We propose a new coding scheme for handprinted Chinese characters, which allows stroke-order free input methods to be derived. In this scheme, each handprinted Chinese character is assigned an n-or-less-digit boxing code, which is derived from the box-like structure of Chinese characters. Each digit of the boxing code represents a unique stroke type of a given character stroke. Subsequent character recognition is then based on the matching of these boxing codes. Experiments were conducted based on 6 basic stroke types (t=6), and a boxing code length of 10 (n=10). Of the 300 characters we have tested, the boxing codes generated from both the dictionary and random stroke orders were the same for each character. The duplication rate was 5.6%. The results of the test indicate that boxing code is a feasible representation of Chinese characters. Although duplications exist, these could be reduced, if not totally removed, by taking in more stroke types. Further investigation is worthwhile in the fine-tuning of the two parameters t and n so as to obtain a more accurate and wider representation, and in the application of the new coding scheme.