OHCS: A Database for Handwritten Atomic Odia Character Recognition

Abstract

In this paper, a complete database of handwritten atomic Odia characters is suggested. The first version of the database has been modeled and named OHCSv1.0 (Odia handwritten character set). The database comprises of 17,100 transcribed characters, each collected twice from 150 unique people at different point of time. Each character has 300 number of occurrences. The character images are standardized to a size of 64 × 64 pixels. A novel framework for perceiving transcribed Odia characters from this database has also been proposed. The character images are gathered into various groups in view of their shape components utilizing an incremental spectral clustering algorithm. During testing, affinity of probe character to a cluster is first decided. Subsequently, the trained classifier recognizes the character inside the cluster. Suitable simulation has been carried out to validate the scheme.