Correlated Run length Algorithm (CURL) for detecting form structure within digitized documents.
Michael D. Garris · 1994
An algorithm designed to detect intrinsic form structure within binary digitized documents has been developed. This Correlated Run Length algorithm automatically locates and extracts line segments, line endings, and combinations of line intersections including corners, crosses, and T's from images. These structures, once detected, can be used as form features to identify the form type in an image, and they can be used to automatically identifying entry fields containing information in need of optical character recognition. This technique has several advantages over more conventional approaches in that form structures are detected without any a priori knowledge of the specific form in the image, and these structures are detected directly from the original image so that any distortions including translation, rotation, and scale are automatically handled. The algorithm performs extremely well on highly cluttered forms and noisy images and is well suited for implementation in a highly para...