Document image analysis for generating syntactic structure description

Abstract

A document image analysis is described which automatically converts an input image into a syntactic document tree structure, while simultaneously representing the elements and their relative relations. Top-down image segmentation, using projection profiles, was greatly improved by systematically using a feedback process. As a result, the tree structure, including the blocks and their relative relations, was generated. Both the elements and their relations in the generated tree were finally determined by the bottom-up strategy, based on the general document layout property. Experimental results showed that this proposed method can be appropriately used to automatically describe an input image as a layout structure.<>