METHODS FOR DOCUMENT IMAGE ANALYSIS BASED ON RETINEX THEORY AND ELLIPSE SHAPES

Conversion from gray scale or color document image into binary image is the
main step in most of optical character recognition (OCR) systems and document
analysis. After digitization, document images often suffer from poor contrast, noise,
non-uniform lighting, and shadow. Also when a page of a book is digitized using a
scanner or a camera, a border noise, which is an unwanted text coming from the
adjacent page, may appear. The skew arises in the document due to incorrect
placement or due to shifting of the object when the scanner lid is being closed. These
problems may result in undesired results in OCR and document analysis. Document
image cleanup is an active subject in image processing, which addresses these
problems. In this thesis, we developed techniques of cleaning up and enhancing
document images that suffer from problems resulting from the digitization process.