Structural Approach on Writer Independent Nepalese Natural Handwriting Recognition

Abstract

The writing units vary in writer independent unconstrained handwriting (for example, number of strokes, shape, size, order, and speed etc.). Many algorithms were developed to improve the accuracy of the handwriting recognition system in both statistical and structural approaches on real-time databases, from which researchers still are not satisfied. We propose to use structural properties of the feature vector sequences of strokes of variable writing units by using dynamic programming (DP). This paper focuses on dynamic time warping (DTW) as a global distance calculation along with the use of local distance metric between two real-time feature vector sequences of strokes and is followed by robust agglomerative hierarchical clustering to produce sensible clusters, which have intrinsic characteristics. We are utilizing feature vector sequences of strokes for both training and testing our recognition system. We work with 20 users and experiment on 36 classes of writer independent real-time Nepalese natural handwritten characters onto our dynamic recognition system stroke by stroke basis and achieve considerable performance