An approach to extract text regions from scene image

Abstract

Text data present in scene images may be the important clue for indexing, automatic footnote, and indexing of images. Now-a-days extraction of text from images has become one of the fastest growing research areas in the field of computer vision. In scene images, text data are present with huge variations in font sizes, styles, alignments, and orientations. These variations make the task of detection and extraction of the text regions from scene images challenging as well as difficult. Low image contrast and complex background also affect the task of text detection and extraction from scene images. Extraction of text from images involves detection, localization, extraction, and enhancement. The goal of this paper is to develop a new and efficient method of text extraction by combining some features from edge-based and connected-component based algorithms. A set of images has been used as input to compare the efficiency of the proposed algorithm. We have calculated the precision, recall rates and accuracy of the proposed algorithm and showed that average accuracy for eight test images is higher than the existing methods. Thus, our proposed algorithm is robust in the extraction of text from different types of scene images.