A Hybrid Approach Handwritten Character Recognition for Mizo using Artificial Neural Network

In the past decade we have seen a rapid advancement in object recognition, however Mizo Handwritten Character Recognition (MHCR) remains an untapped field. In this study a handwritten is collected from 20 different writers each consisting of 456 Mizo characters. In total 20 X 456= 9120 characters are used for testing the proposed system. In this process of recognition, the challenging factor is due to the fact that Mizo handwritten consists of vowels character that are made up of multiple isolated blobs (pixel) such as circumflex (^) on top vowel character. This make segmentation of each individual character difficult and challenging. Therefore, to implement MHCR, a hybrid approach character segmentation using bounding box and morphological dilation is combined, which merges the isolated blobs of Mizo character into a single entity. A hybrid approach feature extraction using a combination of zoning and topological feature is implemented. These features are used for classification and recognition. To evaluated the performance of MHCR model an experiment is carried out using 4 different types of Artificial Neural Network Architecture. Each Architecture is compared and analysed. The Back Propagation Neural Network has the highest accuracy with a recognition rates of 98%. This proposed hybrid technique will help in building an automatic MHCR system for practical applications.