Hierarchical Segmentation of Falsely Touching Characters from Camera Captured Degraded Document Images

An innovative hierarchical image segmentation scheme is reported in this research communication. Unlike static/ spatially divided sub-images, the current innovation concentrates on object level hierarchy for segmentation of gray scale or color images into constituent component/ sub-parts. As for example, a gray scale document image may be segmented (binarized in case of two-level segmentation) into connected foreground components (text/ graphics) and background component by hierarchically applying a gray level threshold selection algorithm in the object-space. In any hierarchy, constituent objects are identified as connected foreground pixels, as classified by the gray scale threshold selection algorithm. To preserve the global information, thresholds for each object in any hierarchy are estimated as a weighted aggregate of the current and previous thresholds relevant to the object. The developed technique may be customized as a general purpose hierarchical information clustering algorithm in the domain of pattern analysis, data mining, bioinformatics etc.