A Fast Multifunctional Approach for Document Image Analysis

Collinear arrangement of objects (such as, text elements or continuous lines) is integral part of any office document image, whether structured or unstructured. The ability to analyze such an organization of objects thus provides the basic and important building block for a plethora of image analysis applications. Most Hough Transform-based line detection approaches do not furnish line widths and other measurements, and are computationally expensive for large-sized images. Other approaches often deploy a filter or morphological operation as a pre-processing step, which introduces reverse noise pattern while attempting to solve the cleaning problem in generalized manner. We propose an algorithm for fast, accurate, efficient and customizable detection of lines, which returns complete description of lines without having to apply an image pre-processing or conditioning step. Our approach, furthermore, allows simultaneous removal/ reproduction of lines, which is invariably used in the later phases of image analysis for higher-level interpretation and matching. The speed and flexibility of the approach presented here makes it serve as a multifunctional building block for a variety of document image analyses. The integration of this approach as a building block for diverse application areas have been implemented and explained.