Multilevel segmentation and analysis of facsimile images for document classification

Automatic processing of facsimile images which could classify and sort incoming transmissions and route them accordingly is a necessary component for an office automation system. A multilevel segmentation and classification system is presented. The first level indicates the major `visual´ groupings of the information on the page and the second level explicitly identifies the number and type of internal elements. The medial axis transformation and contour shape parameters are used to analyze irregularly arranged areas and nonmachine generated areas of a document. This approach is efficient because at each step the level of detail used in processing is appropriate for the decisions which are required. Examples of group 2, group 3 standard, and group 3 high resolution images demonstrate the segmentation and classification performance