A fast multifunctional approach for document image analysis
Rakesh Kumar, M. Mukerji, Hande Gür · 2003
Collinear arrangement of objects (such as, text elementsor continuous lines) is integral part of any officedocument image, whether structured or unstructured. Theability to analyze such an organization of objects thusprovides the basic and important building block for aplethora of image analysis applications. Most HoughTransform-based line detection approaches do not furnishline widths and other measurements, and are computationallyexpensive for large-sized images. Other approachesoften deploy a filter or morphological operationas a pre-processing step, which introduces reverse noisepattern while attempting to solve the cleaning problem ingeneralized manner. We propose an algorithm for fast,accurate, efficient and customizable detection of lines,which returns complete description of lines withouthaving to apply an image pre-processing or conditioningstep. Our approach, furthermore, allows simultaneousremoval/ reproduction of lines, which is invariably usedin the later phases of image analysis for higher-levelinterpretation and matching. The speed and flexibility ofthe approach presented here makes it serve as a multi-functionalbuilding block for a variety of document imageanalyses. The integration of this approach as a buildingblock for diverse application areas have been implementedand explained.