Text Detection in Sign Images Using Maximally Stable External Regions and Stroke Width Transform

Sign images are widely used in daily life and contain a lot of important text information, but there is few text detection method for its characteristics. In view of different styles, sizes, and colors for its fonts, a text detection method based on Maximally Stable External Regions and Stroke Width Transform is proposed. Firstly, the candidate regions are extracted from Maximally Stable External Regions. Secondly, the candidate regions are optimized by Stroke Width Transform. Thirdly, the non-text areas are filtered according to several heuristic rules and CIELab color model. Finally, text regions are merged to text lines as the detected regions. Experimental results show that the proposed method is efficient on the horizontal text sign images, and achieves the F-measure of 87.43% under the ICDAR2013 evaluation index.