An algorithm for colour-based natural scene text segmentation

Before the step for text recognition, a text image needs to be segmented into foreground containing only the text area and background. In this paper, a method is proposed for segmenting colour natural scene texts which suffer from a wide range of degradations with complex background. A text image is firstly processed by two 3-means clustering operations with different distance measurements. Then, a modified connected component (CC)-based validation method is used to obtain the text area after clustering. Thirdly, a proposed objective segmentation evaluation method is utilised to choose the final segmentation result from the two segmented text images. The proposed method is compared with other existing methods based on the ICDAR2003 public database. Experimental results show the effectiveness of the proposed method. © 2012 Springer-Verlag Berlin Heidelberg