Document Image Segmentation Using Dynamic Thresholds and Identification of Each Region Type

Nowadays, the accumulation of paper in the life of business professional is overwhelming. Digital documents on the other hand being less expensive and more efficient are on the road to a more organized office. Also crucial for many business applications document image analysis is needed before OCR operation. Segmentation of document images into text and non-text regions is essential because OCR recognition engine produces garbage text when it gets non-text components as input. In this paper a segmentation technique is presented to decompose document image into its constituent parts, such as text blocks, pictures and tables. This technique is implemented by combining two approachesrun length smearing algorithm and recursive top down approach. Recursive top down approach works using horizontal and vertical projection profiles. Proposed technique is threshold based but threshold values are automatically calculated depending upon the geometric layout of the document. Binarization and noise removal are done as part of preprocessing. This approach works for documents in any script and in manhattan layout. Keywordsdocument image segmentation, dynamic thresholds, horizontal and vertical projection profiles, OCR, recursive top down segmentation, run