Text Detection of Two Major Indian Scripts in Natural Scene Images

摘要

In this article, we present a robust scheme for detection of Devanagari or Bangla texts in scene images. These are the two most popular scripts in India. The proposed scheme is primarily based on two major characteristics of such texts - (ⅰ) variations in stroke thickness for text components of a script are low compared to their non-text counterparts and (ⅱ) presence of a headline along with a. few vertical downward strokes originating from this headline. We use the Kuelidean distance transform to verify the general characteristics of texts in (ⅰ). Also, we apply the probabilistic Hough line transform to detect the characteristic headline of Devanagari and Bangla texts. Further, similarity and adjacency measures are applied to identify text regions, which do not satisfy the verification in (ⅱ). The proposed approach has been simulated on a repository of 120 images taken from Indian roads and the results are encouraging. Also, we have discussed the applicability of the proposed scheme for detection of Fnglish texts. Towards this end. we have considered the training and test samples from the image databa.se of ICDAR 2003 Robust Reading Competition.