A Method of Text Segmentation from Scanned Image with Complex Background

With the development of information technology, the number of scanned images is increasing rapidly. There are many important texts in these images. In order to satisfy the need of images viewing, text identification and text retrieval, this paper presents an efficient method for text segmentation. Firstly, localizes the text blocks in scanned image. Secondly, according to its gray/color distribution, the text block image is decomposed into text sub-layer, background sub-layers and mixed sub-layers which contain both texts and backgrounds. Finally, the backgrounds are filtered out from these sub-layers, and the combination of texts in all remained sub-layers is the text segmentation result. Experimental results show that the proposed method is robust to overlapped complex background.