Quality Evaluation of Character Image Database and Its Application

This paper presents a quantitative evaluation method for hand-written character image database. In fact, this research was done in 1992. However, recently, a large amount of character images has been collected from school children, the quality of these character data have been evaluated by this method. In this paper, firstly, some other evaluation methods are introduced, and their drawbacks are pointed out. Then, a method using entropy are introduced. Our method based on entropy is also presented. We call this variation metric ´variation entropy´. This metric has two kinds of aspects. One is a absolute evaluation of variation, the other is a relative evaluation of variation. The former can be quantified by ´variation entropy for a unit boundary length(VEUB)´, and the latter can be quantified by ´variation entropy for a unit area(VEUA)´. The properties of these two metrics are complementary. Lastly, two variation entropies are applied to the standard kanji character database and a database collected from school children.