Applying Image Analysis and Machine Learning to Historical Newspaper Collections

Abstract

Abstract Diving below the surface has its challenges, however. For example, “noise effects” are especially widespread when digital images have been created from earlier microphotographic copies, as is common in historical newspaper collections. Noise effects introduce interference to the primary signals of the pages, both for human vision and computer vision and processing. Various types of noise effects (fig. 1) are common, including unevenly distributed luminosity (i.e., range effects), visible characters from the other side of the page (bleed-through), tilted document scans (skewed orientation), and markings on the newspaper that obscure text (blobs).1 There is a wide range of severity for each of these effects, and images can range from very clean to very noisy within and across datasets.