Benchmarking of text segmentation in devnagari handwritten document

摘要

Handwritten Character Recognition is the capability of a computer to receive and interpret handwritten input from paper documents, photographs, touch screens and other devices. In this paper we have introduced a new method for Hindi handwritten character segmentation. It consists of a novel approach segmentation line, word and character using depth first search on the distance metric of connected components identified using 8- connectivity mechanism on the foreground pixels of the image. Next the extracted words are skew normalized using their orthographic projection. The skew corrected words are then segmented into upper modifier, lower modifier and individual letter using average height of consonant in the document. A novel database are developed for recognition. The database for off-line Hindi handwritten character with modifiers consist more than 26000 images of their original size with programmatically segmented consonant and vowels.