Handwritten character recognition by an adaptive fuzzy clustering algorithm

Unconstrained handwritten characters pose a serious challenge to the development of a recognition algorithm. Many approaches have been studied over the years for such a recognition algorithm. We use an adaptive neuro-fuzzy clustering algorithm for classification and recognition of handwritten characters of a variety of styles and investigate the effectiveness of Fourier coefficients as representative features of handwritten characters in the presence of noise. Our results indicate that the adaptive clustering algorithm outperforms k-means clustering in handwritten character recognition for the same data representation. However some misclassifications cannot be avoided due to inherent problems associated with large variability in handwriting styles and the presence of excessive noise in practice.