Comparison of Multilayer Feed Forward Neural Network and Hopfield Neural Network upon Buddhist Iconographies Classification

Abstract

This paper proposes two types of artificial neural networks to classify the Buddhist iconographies in order to compare the performances of two networks; multilayer feed forward network and Hopfield network. The quality of the image of Buddhist iconography is improved by smooth filtering. The smooth filtered image is transformed into intensity image. The resultant gray image is edge detected to obtain the abstract graph or the edge of the Buddhist iconography. The resultant image is resized into the defined pixel area. This is the final step of the image preprocessing stage. The resized image is applied into the two networks. Multilayer network is composed of three layers. The input layer accepts the preprocessed images and the certain values of the output layer indicate the type of the Buddhist iconography. Hopfield network consists of a single layer which serves as both input and output layers. The training time and accuracy of the two networks are then analyzed and compared in this proposed paper.