Developing offline handwritten recognition systems for Meitei Mayek

Abstract

Handwritten character recognition plays a prominent role in the digitization of old documents, image restoration, and recognition methods with several applications in academic, banking, and postal system. Developing a potential handwritten character recognition system that could sustain a high accuracy with a large and varied dataset is a complicated task. It can be found from the literature that many researchers have designed various recognition models for handwritten Meitei Mayek character recognition. However, there is still a scope of improvement for the recognition rates. Moreover, most of the existing studies on Meitei Mayek are focused on a single character database. An important fact that is worthy of mentioning is that the quality or efficiency of the system is directly proportional to the input document. Regional language generally makes recognition tasks more complicated to analyze and interpret the characters from the images. Thus, it stands as a challenging area for the researchers. The problem is worth investigating for its two-fold significances. First, designing and developing datasets of isolated handwritten characters and text documents, giving the most critical input for developing a recognition system. Secondly, performing several operations on the collected datasets to complete the recognition process, namely character recognition on isolated character dataset and segmentation on text documents dataset.