The Minimum Description Length Principle and Its Application to Online Learning of Handprinted Characters

Qiong Gao, Ming Li

Abstract: Our objective is to introduce Rissanen's Minimum Description Length (MDL) Principle as a useful tool for character recognition and present a first application. Using MDL principle, a learning system has been implemented which, after simple training, is able to online recognize th trainer's handwriting in English and Chinese (including several thousand characters) with high success rate. The experimental results conform with the theoretical predictions. We will also try to give an elegant explanation of Rissanen's minimum description length principle (MDLP).