Korean Text Detection and Binarization in Color Signboards

Text detection and binarization are very important steps in text understanding. Natural scene images bring new challenges to correctly extract interested text regions. In the paper, efficient text detection and binarization methods are used to extract Korean text from color signboards of shops. First, we detect the main Korean text in color signboards by combining the horizontal edge profile in the image and some knowledge of the Korean text specified in our application. Then the detected Korean text is segmented to Korean words by using the vertical edge profile. Finally, each Korean word is fed into binarization module which uses 3-means clustering method in L*a*b* color space to extract text from the background. The experimental results show that our method successfully extracts text with low complexity and therefore can be used in the mobile devices which have limited capability.