A Grid based Approach for Handwritten Text Segmentation

The hand-written text is subject to inconsistency and variability. To achieve the accuracy and consistency in recognizing handwritten characters, the present work discusses an efficient approach to segment the handwritten text into individual characters. In order to achieve this, the writing space is divided into three different zones and then the zones are subdivided tiny square shape grids, which are distributed evenly. After this, we analyze to find the presence of handwritten character in a selected grid by performing horizontal scanning method. Moreover, we propose an approach of segmenting the character set of a single line text by vertical scanning and curve pattern analysis on the touching zones of the selective grids. The efficacy of the approach has proven through rigorous testing on hand-written texts of both professional as well as non-professional users. The testing set consists of 106 text samples including both single line text phrases and short sentences. The proposed approach achieves around 100% accuracy for word level text segmentation on the handwritten texts of both professional and non-professional users. In addition, the average character level segmentation accuracy on texts written by professional and non-professional users are found to be 98.4% and 96.2% respectively. Thus, the average character level segmentation accuracy is improved by 97.3%.