Authors - Sanasam Inunganbi, Robin Singh Katariya Abstract - Deep learning is increasing rapidly and is becoming a leading mechanism for various applications in machine learning and artificial intelligence. Among the many research area where deep learning shines, one crucial area is image classification. Handwritten character recognition is a fundamental research area in image classification. It has the ability to ourish in numerous utilizations such as postal automation, banking, form filling, etc. Still, establishing such a system with high accuracy is a challenging effort with the diversified writing fashions, variable size, or different strokes for the same character and resemblance of diverse characters. Further, shapes and the inconclusive writing style of several individuals complicate the problem. The stated problem can be solved by an intelligent and appropriate extraction of features is required. In this paper, a recognition system is presented using transfer learning. The technique is popular for building accurate designs in computer vision. Transfer learning helps to begin from patterns that have the experience to solve a different problem instead of starting from the basics. The performance of the proposed method has experimented on a self-acquired handwritten Meitei Mayek (Manipuri script) database contributed by diverse people holding different education backgrounds and ages. A total of 14; 700 sample images are used for learning various pre-trained models. The highest average accuracy achieved among the models is 98.41%.
Open Zoom