Binarizing Document Images Acquired with Portable Cameras

Abstract

Although made for "family photos" portable digital cameras, either in standalone models or embedded in cell phones, are often used to take photos of documents today. In general, such photos are sent via networks and either visualized in desktops, printed, or even transcribed via OCR. Binarization may play an important role in such a scheme. This paper follows the idea that "no binarization algorithm is good for all kinds of images". Non-uniform illumination, the possible interference of light sources from the environment, and non-uniform resolution are some of the problems found in photographed document images that are not present in their scanned counterparts. This paper presents a new methodology to assess binarization algorithms in different devices, taking into account the difficulties listed and the particularities of the cameras and documents.