Overlapped Character Recognition: An Innovative Approach
Shivadatt D Chame, Anil Kumar · 2016
Immense analysis has been done on optical character recognition (OCR). Numerous works has stated for English, Chinese, Devanagari, Malayalam, Arabic scripts, etc. Segmentation has imp phase in OCR and various articles have been published on different segmentation methods like Thinning, histogram etc for different script during last few years. Generally there is not work done on Overlapped and touching scripts. In this article implemented the latest methods of Overlapped character recognition. Segmentation of Overlapped characters has an extremely strenuous task due to the large variety of characters and their shape, font in the script. It is an important step because inaccurate segmentation will cause errors in recognition. Normalization, binarization and thinning are the pre-processing mechanism used in handwritten character recognition. Proposed Method uses threesholding and Blob Analysis for segmentation and to detect overlapped region using Freeman Chain Code. Finally, we used Support Vector Machine (SVM) for resulting feature vectors and obtain classification performance in the character recognition scheme. An overall performance of 93 % at line and curve set of overlapped images are better than existing methods. Here we have empirically performance of segmentation of overlapped characters with the help of different overlapped images.