Computational models for historical document image enhancement

Rapid evolution of scanning and computing technologies in recent years has led to the creation of large collections of scanned historical documents. Usually, these scanned documents suffer from some form of degradation. Large degradations make documents hard to read and substantially deteriorate the performance of automated document processing systems. Enhancement of degraded document images is normally performed assuming global degradation models. When the degradation is large, global degradation models do not perform well. In contrast, we propose to learn local degradation computational models for binarization and enhancement.;We approach the task of document enhancement from a machine learning perspective by generating computational models using both unsupervised and supervised learning techniques and applying them in a principled manner. Our novel contributions are in these three areas of document imaging analysis: binarization, enhancement, and evaluation. First, we develop a multi-resolution framework for document image binarization and employ it using separately Expectation Maximization and Linear Discriminant Analysis via the Otsu algorithm.;Second, we propose two enhancement models: a local similarity model based on look-up tables in conjunction with the approximated nearest neighbor algorithm, and a graphical model utilizing a 2-layer Markov Random Field to model dependencies among patterns of degradation and their corrections in a document image. When enhancing document images with the goal of improving readability, it is important to understand human perception of quality. Hence, third, we propose a novel method for learning and estimating human perception of document image quality. Experimental results obtained demonstrate the advantage of our proposed methods to current state of the art techniques.