Extracting the characters from scanned Telugu document images

Segmentation is an important task of any OCR system. It separates the text document images into lines, words and primitives (characters). The accuracy of OCR system mainly depends on the segmentation algorithm being used. Segmentation Telugu text is difficult when compared with Latin based languages because of its structural complexity and increased character set. It contains vowels, consonants and compound characters.

Therefore, in this paper, an attempt has been made to develop a Segmentation technique for Telugu language. The first major step in the Segmentation phase is demarcation of lines. The problem can be solved by choice of appropriate value for row-threshold so that two text-lines get separated. This can be done by considering the horizontal and vertical profiles of the scanned document image. The technique has been developed for segmenting the scanned documents. Once a line is identified, each word within that line is separated in a similar manner with the proper choice of word-threshold the demarcation of words is done. The segments are identified within each word in the similar manner by scanning each word column wise. In this implementation, a zero column identifies the end of a segment and thereby the segment is extracted from the given word or line. The proposed algorithm is based on projection profiles, connected components and extract primitives. The algorithm extract primitive uses grow region algorithm to identify a connected component. Once the connected component is identified, and is extracted from the segment. The accuracy rate was 100% in case of good scanned Telugu documents and 95% in case of overlapped documents.