Handwriting character recognition system in documents containing abbreviations using artificial neural networks

Offline Handwriting Character Recognition (HCR) is more challenging than online HCR because of inadequate temporal information such as number and direction of the stroke, ink pressure, unpredictable and high handwriting variations. These difficulties cause low accuracy achieved. The purpose of this study was to perform handwriting pattern recognition on documents containing abbreviations using ANN. The document was written in Indonesian. Some steps are taken to detect abbreviations: Collecting handwritten samples, doing image processing, practicing ANN. From the classification process, there are two indicators of accuracy used. Character accuracy based on classes and unusual abbreviations detected. Character accuracy based on class achieved is 60.47%, and for accuracy of abbreviations detected is 27.89%. Low accuracy results because the accuracy of the introduction of 9 of 26 letters is not more than 50%. This study contributes to providing knowledge about image processing and the application of ANN to pattern recognition problems, namely handwriting. In addition, this research also contributes in providing a model for the development of a system for detecting abbreviations in Indonesian written which can be applied in essay exams in education. In future research, adaptive thresholding can be applied to improve system performance. Besides, to minimize recognition errors caused by imperfect character segmentation, recognition is done in one word.