Adaptive method for multi colored text binarization

Abstract

This article presents our recent study on multi colored text binarization. In the output image, we represented foreground content as black and background as white regardless the polarity of foreground and background in original image. Here we applied connected component analysis based approach to group the words or characters within bounding or edge box. The main novelty of this reported work includes the calculation of each edge box based local color threshold value from CIELAB color space. This approach makes the proposed system capable of binarizing multi colored texts where a single character has more than one color. The proposed method has been executed on well-known D1BCO2009 and CMATERdb datasets that contain a large set of images to show the efficiency over other existing methods through qualitative comparison study.