Script independent scene text segmentation using fast stroke width transform and GrabCut

Abstract

In this paper, a novel script independent scene text segmentation method is presented. The proposed method introduces an approach for character segmentation which combines the advantages of connected component analysis, stroke width analysis and graph cuts. The method was evaluated on a standard dataset, and also on a custom dataset of scene images of text from various Indian scripts. Additionally, a fast implementation of Stroke Width Transform (SWT) is presented here, which improves speed over other implementations by 79%. From our experiments, we have achieved encouraging results on the usefulness of the combination of text/non-text classification with GrabCut. The technique is robust to the language/script of text. © 2015 IEEE.