A NOVEL METHOD FOR EXTRACTING TEXT FROM A GEOMETRIC REGION

Dedao Wu,∗,∗∗ Peter X. Liu,∗∗∗ and Yanni Zou∗

Keywords

Text detection, edge detection, recursive segmentation, single text line, geometric contour

Abstract

Text detection and text segmentation are two key steps in text information extraction systems. At present, many methods do not perform well on the geometric contour detection of text regions, which directly leads to difficulty in identifying the text in the region of interest in geometric contours. Aiming at this problem, a novel text extraction method based on geometry contours is presented. Specifically, a two-step strategy is developed in this method. First, text interference outside the geometry contour is eliminated by using an edge detection algorithm to detect and locate edges. Second, pixel detection and recursive segmentation algorithms are exploited to extract text information in geometric contours. Experimental results show that the presented method achieves satisfactory results on publicly available image datasets: a self-collected dataset and Google. The presented method also works well on the datasets that we collected, including both Chinese and English.

Important Links:

Go Back