An Approach Towards Devanagari Character Segmentation Using Adaptive Thresholding And Polynomial

Abstract

Recognition of handwritten characters from the documents is one of the very challenging zones of research in the area of Character Recognition. The recognition of manually handwritten content is a very difficult task because of distinction in the composing styles of various individuals. In the very beginning, a great deal of work has just been done on recognition of written Hindi characters (isolated characters) and numerals, however much work is not accounted for the segmentation and recognition of manually written content in general. The recognition rate and accuracy highly depend upon preprocessing techniques including segmentation. This paper presents development of such techniques for segmentation of the Devanagari characters which ultimately improves the recognition rate. Two approaches are used for segmentation namely adaptive thresholding and thresholding using polynomial. In the first approach adaptive thresholding is used for which the histogram of gray scaled image is considered. This gray image is partitioned into blocks of different size for which standard deviation is calculated. Depending upon standard deviation, threshold value is calculated. Second approach is fitting a polynomial curve using least square distance on the histogram of gray scaled image. Using this polynomial curve, minimum value at which the curve fits is found and used for thresholding of an image.