EFFECTIVE AND EFFICIENT BINARIZATION OF DEGRADED DOCUMENT IMAGES
Extracting information from images of documents is easier when the image is crisp, clear and devoid of noise. Consequently, an algorithm that reliably removes noise from imperfect document images and generates better images could clean input to other image processing algorithms thereby improving their outputs and/or enabling simpler techniques. The importance of this task is evident given the rate at which scanners, copier, and smart phones are producing document images.This dissertation makes three contributions to this problem area. The first contribution is an unsupervised method for converting a document image to a strictly white and black image (i.e., cleaning a document image). This initial contribution is the result of examining the hypothesis that acceptable binarization parameters can be found with an automatic parameter search and was patented in US Patent #8,995,782 "System and Method for Enhancing the Legibility of Degraded Image". The second contribution is an improvement on the prior method that eliminates the need for a computationally expensive parameter search. A patent for this contribution has been allowed, but not yet issued, under patent application number 13/949,799 "System and Method for Enhancing the Legibility of Images". The last contribution of this dissertation is a method that manipulates multiple images of the same document that were each captured with a different mono-chromatic frequency of light to improve image binarization.
MetadataShow full item record
Showing items related by title, author, creator and subject.