Text detection based on stroke features

Abstract

A text detection method was presented based on support vector machine（SVM） using the statistics features characterizing character strokes.First,our method extracts stroke edges through a modified edge detector;then,candidate text regions are located by merging the regions that contain stroke edges;finally,a 32-dimensional feature is devised to reflect the unique spatial distribution of stroke edges,and the SVM is utilized to model and verify the candidate text re-gions.Our experiments on Chinese characters demonstrate the proposed stroke texture features have good distinction power,especially for text regions composed of many characters。

Keywords

• text detection;support vector machine