Transition thresholds and transition operators for binarization and edge detection

Abstract

The transition method for image binarization is based on the concept of t-transition pixels, a generalization of edge pixels, and t-transition sets. We introduce a novel unsupervised thresholding for unimodal histograms to estimate the transition sets. We also present dilation and incidence transition operators to refine the transition set. Afterward, we propose the simple edge transition operator for detecting edges. Our experiments show that the new approach increases the effectiveness of OCR applications outperforming several top-ranked binarization algorithms. © 2010 Elsevier Ltd. All rights reserved.