Efficient segmentation of printed tamil script into characters using projection and structure

摘要

Segmenting text lines and touching characters remain a problem. But this paper segments the printed touching lines and characters of Tamil script into lines, words and characters. Standard horizontal projection and vertical projection methods cannot segment the touched lines and characters. The proposed method solves the problem of touching lines and touching characters of Tamil Script based on the structural properties of the characters and projection. The proposed method is implemented on different set of documents collected from different Tamil literary periodicals with different sizes and fonts. Experimental results are compared with the projection profile based technique and connected component labelling technique. Results shown that the proposed method segment the documents into lines, words and characters with good accuracy for the regular fonts with any size even though the lines and characters are of touching nature. This method can be applied in preparing OCR system and in document analysis and recognition system when the printed documents are with line and character overlapping.