Segmentation of text/image documents using texture approaches

The digital computer and computer networks have made it possible to search for and retrieve electronically stored documents in seconds, no matter where in the world they are stored. This is far from the reality for documents stored as paper copies. Therefore there is considerable interest in digitizing paper documents. To digitize existing paper documents, it is of great importance to be able to separate the text from the graphics, in order to make the text searchable and more eeciently stored. In this paper we present an approach to segmentation of text and graphics in scanned documents, based on the assumption that the text in a document may be viewed as one texture, while the graphics is a diierent texture. Using this assumption, we segment the documents with a texture segmentation scheme using lter banks as the feature extractors. While most traditional text-graphics segmentation schemes assumes some a priori knowledge of the input, our approach is independent of document layout, typeface, font size, scanning resolution etc. Another approach to texture segmentation of documents for text-graphics segmentation has been presented by Jain and Bhattacharjee, using the Gabor l-ter as the feature extractor. In this paper we show that equally good results may be obtained using much more computationally eecient critically sampled perfect reconstruction lter banks.