Markov random field model based text segmentation and image post processing of complex scanned documents

Eri Haneda, Purdue University

Abstract

In this dissertation, two image processing studies will be presented. The first study is segmentation for MRC document compression using an MRF model, and the second study is an automatic contrast enhancement scheme for a digital image capture device. In the first study, we developed a new document segmentation scheme for Mixed Raster Content (MRC) standard. The Mixed Raster Content standard (ITU-T T.44) specifies a framework for document compression which can dramatically improve the compression/quality tradeoff as compared to traditional lossy image compression algorithms. The key to MRC compression is the separation of the document into foreground and background layers, represented as a binary mask. Therefore, the resulting quality and compression ratio of a MRC document encoder is highly dependent on the segmentation algorithm used to compute the binary mask. In this study, we propose a novel multiscale segmentation scheme for MRC document encoding based on the sequential application of two algorithms. The first algorithm, cost optimized segmentation (COS), is a blockwise segmentation algorithm formulated in a global cost optimization framework. The second algorithm, connected component classification (CCC), refines the initial segmentation by classifying feature vectors of connected components using an Markov random field (MRF) model. The combined COS/CCC segmentation algorithms are then incorporated into a multiscale framework in order to improve the segmentation accuracy of text with varying size. In comparisons to state-of-the-art commercial MRC products and selected segmentation algorithms in the literature, we show that the new algorithm achieves greater accuracy of text detection but with a lower false detection rate of non-text features. We also demonstrate that the proposed segmentation algorithm can improve the quality of decoded documents while simultaneously lowering the bit rate. In the second study, we developed a robust algorithm to perform automatic contrast enhancement. The motivation for this study is that the background of scanned images by digital scanners sometimes appear too dark. This is particularly true for scans of paper materials such as newspapers, magazines and phone books. In such cases, a lighter and more uniform background color is typically preferred because it has the advantages that the contrast of the image is more distinct, the background noise is reduced, and the background color is more consistent across paper materials. Our objective of this study is to snap the background color of the paper material to display-white and the full black colorant of the printer to display black without large color shift. In addition, we want to reduce show-through effects of thin paper materials. Our algorithm described in this document consists of four components: auto cropping, paper-white estimation, paper-black estimation, and linear contrast stretch. First, the algorithm performs auto cropping of an input image to extract the region which contains only paper. Next, paper-white color and black colorant color are estimated using the maximum and minimum values for each RGB channel. Finally, a linear contrast stretch is performed by snapping the estimated paper-white and estimated black colorant to the largest and smallest encoded value to increase the dynamic range. We show quantitative evaluation for our paper-white estimation algorithm, and show qualitative results of overall contrast stretch.

Degree

Ph.D.

Advisors

Bouman, Purdue University.

Subject Area

Electrical engineering

COinS