A variation measure for handwritten character image data using entropy difference

摘要

Since handwritten characters vary, we need to develop a measure that reflects the degree of variation for a given set of character data. This paper first defines four properties — boundedness, independency, monotonicity and constancy — which such a variation measure is required to have. We show that none of the variation measures previously proposed satisfy all these properties. Then a new variation measure, called Average Entropy Difference, is proposed. We show that the new variation measure satisfies all four properties. Finally, in order to see how different those variation measures including the proposed one are, the values of the measures were calculated for various artificially generated data. The calculated results support well our theoretical analysis.