Detection and segmentation of text in handwritten Hindi documents

Abstract

Optical Character Recognition(OCR) is one of the most important applications of Computer Science. The accuracy of Character Recognition in the case of machine printed characters has improved, but in case of Handwritten documents researchers are still working on accuracy improvement. The system is complex as different writers have different way of writing and even in the same document a single character or word can be represented differently by the same writer, which leads to less accurate results. In said environment, accurate segmentation plays a very important role. To achieve this, one has to efficiently handle problems like unequal spacing between text lines, overlapped text lines, connected components between the lines, unequal height of characters, separation of upper and lower modifiers. During the segmentation stage, if some error occurs then it will keep on increasing the error rate in the next stages. In this work, new Segmentation algorithm(s) has been proposed for Handwritten Hindi documents. An encouraging segmentation accuracy rate of 96.87% at line, 92.45% at word and 87.13% at character level has been achieved.