A Novel Approach of Normalization for Online Handwritten Characters Using Geometrical Parameters on a 2D Plane
Gouranga Mandal, Parthasarathi De, Diptendu Bhattachya · SSRN Electronic Journal · 2019
Handwritten text recognition is one of the toughest jobs for a computer system as handwriting does not follow any uniform rules. Recognition of handwriting with greater accuracy is a very difficult task for the researcher. Researchers have done many experiments on the offline handwriting of different script, whereas research work on recognition of online character, word or documents for online handwritten data is quite less. Several offline and few online explorations have been done in recognition of English language but online character and word recognition of Indian language particularly for Bengali is an on-going and exciting research area as the alphabets of Bengali script is comparatively complicated than English in shape. A handwriting recognition system involves many pre-processing steps, after pre-processing step normalization is the most momentous step which is usually done before recognition. As different people write handwritten text in different style, so shapes of every same latter also not same for different people. As a result, it is very hard to distinguish a specific character properly. Here we have introduced a new technique by which every character can be converted into a uniform size and shape so the recognition part will be very easy for system. This normalization process consists of two steps. Step one is resizing, i.e.: height and width correction maintaining the pen movement angle remain same considering only busy zone and the second step is the conversion of pen tip movement of the character into a 3X3X3 3D matrix so that every character can be represented in a common and uniform way and the matching for recognition part will be easier. We have tested this proposed technique on 2655 Bengali handwritten basic characters and have got 96 percent of accuracy which is a great achievement.