Scale Detection Based on Maximum Entropy Principle

Abstract

In this study, a method to estimate the scale of the location of an arbitrary image is proposed based on the maximum entropy principle. The scale corresponds to the window covering the most amount of information. Gaussian and error functions are used to reveal entropy-scale relations. This study also introduces the entropy of functions and discusses its connection with differential entropy and Boltzmann's entropy. Experiments showed that the proposed method can effectively estimate the scales of functions and images.