Fourier fuzzy neural network for clustering of objects based on the gross shape and its application to handwritten character recognition

In this paper an unsupervised feed forward Fourier fuzzy neural network (FFNN) is proposed which is suitable for clustering of object images based on their gross shapes. This 3-layer feed forward neural network is described along with its training. Its performance is tested for synthetic image database containing objects of various shapes and with realistic image database of handwritten Devanagari digits. Performance of FFNN is found superior than the fuzzy min-max neural network (FMN) clustering by P.M. Patil et al. (2002), and it takes less recall time per pattern than FMN.