Handwriting Recognition through Novel Combinational Segmentation Technique and Iterative Blocks Features
Madhuri Maheshwari, Deepesh Namdev, Saurabh Maheshwari · 2018
We propose a novel technique for segmentation of the handwritten answers written in continuous writings by the candidates during theory based examinations. We estimate digital information from the handwritten words. The problem with cursive writings is that the neighboring characters may be overlapping or may be some times away from each other. Also all the characters are not of same size. A single length window cannot segment all the characters while an adaptive length window would either require multiple scans of the image or multiple windows might be needed, increasing the computational complexity. Here, we propose a novel combinational segmentation scheme where variable length cuts are of the word image are generated as tentative segments and the length of cuts depends upon the vertical pixel density threshold. These segments may have some segments with deficient characters or excess parts of neighboring characters. So recombination of these segments with nearest segments is done to generate final segments. The advantage of the method is that we can segment any handwriting. Then, we train the system to identify a character through the proposed iterative blocks scheme. The features extracted are density per block and number of junctions and end points per block. Text written by the same author is used for the testing. We have tested for 10 subjects with character recognition accuracy more than 90%.