Image Binarization Based on Digital Flow Analysis

Abstract

For image recognition and understanding, it is one of the fundamental problems to binarize the gray-scaled image and divide the image into a target and a background. The threshold-based method is now used commonly in the image binarization. However, this method depends on the statistical condition of the image. There is a method to analyze geometrically in one of the methods not to depend on a statistical condition. We can consider the image to be a curved surface when the brightness of a pixel(x, y) is height. The surface generates a flow field based on the its gradient because of the differential geometric property. Therefore, in this article, we define the slope lines that describe the structures of surface very well. These lines emanate from the same maximum, and reach the same minimum. We call this lines dipole flow. We show that a set of the inflection point is classified based on its morphology. And we introduce a binarization method based on this classification.