Segmentation of Medieval and Ancient Coins over Complex Backgrounds: A Fast Automatic Method

Coin segmentation, or separating the coin area from its background, is inevitably the first step in any robust classification method. Yet, the fact that almost every research relies exclusively on grayscale images taken in controlled environments, with uniform illumination and backgrounds, wastes a vast asset of images commonly taken by numismatists, sellers and collectors. Admittedly, very often these are not easy images to work with, combining problems ranging from high cast shadows to highly textured backgrounds. In this paper we propose a fast and automatic method to segment ancient coins over complex backgrounds using a Histogram Backprojection approach. Results are compared against a proposed automation of GrabCut algorithm, and its benefits are demonstrated. Although the present article is oriented to ancient coin segmentation, the method can also be used in other contexts presenting thin objects with uniform colors.