Text Location in Camera-Captured Guidepost Images

A method is proposed to locate text in camera- captured guidepost images. Firstly, due to its advantage of smoothing the low contrast information, mean shift method is applied to remove some complex background. In order to improve the time efficiency, we modify the traditional mean shift method. Secondly, two stage features are used for the edge map of image to classify it into candidate text blocks and background by K-means algorithm. Finally, the text blocks are identified according to some empirical rules and project profile analysis. Experimental results demonstrate that the proposed method can efficiently locate the text region contained in camera-captured guidepost images and be robust for font, size, color, language, alignment and complexity of background.