A High-Speed Document Image Classifier

In this paper, a high-speed document image classification algorithm is presented. The algorithm is based on the bottom-up strategy which can successfully segment and classify any type of sophisticated layout documents without the limitation of Manhattan rules. Special techniques are employed to overcome the slow speed problem facing most of the bottomup algorithms. During segmentation process, the algorithm used a byte-based operation to establish a list of connected components within a scan line, and to merge touching connected components in adjacent scan lines. The texture features of each connected component are also obtained during the process. This greatly reduced the computation time spent on classification stage. For a typical A4 size document image with 2OODPI resolution, the average processing time on a Pentium based IBM-PC computer is 1.2 second.