ABSTRACT
Abstract A new thresholding technique based on transition region determination and using combined grey-level and two-dimensional context information has been developed. The transition region of images is geometrically located between objects and background, and is composed of pixels which have intermediate grey levels between those of objects and background. Threshold selection from the transition region makes the thresholding procedure highly reliable. The process can be expressed with rigorous mathematic tools, and also has significant physical meaning. As contrasted with normal thresholding techniques, no histogram calculation is needed. This technique is fully automatic and no empirical parameters have to be introduced. Moreover, it is quite robust in the presence of noise and unexpected structures as a global clipping transformation has been employed. Finally, no limitations on shape and size of objects are imposed. A comparison with histogram-based threshold selection methods using real images is also presented.
