Classification of Persian handwritten digits using spiking neural networks

In recent years Spiking Neural Networks (SNNs) have gained in popularity due to their low complexity. They have been used in many processes like learning and classification of data such as images. In this paper we have used the SNN Model, in order to have robust learning and classification of handwritten digits, i.e., to have a learning process which is persistent against changes and high noise levels. Due to the similarities among handwritten digits, the classifications have been erratic but the Deep Belief Network we have used in this paper solves this problem to a great extent. Our model consists of three layers. The first layer, composed of 225 neurons (15∗15 pixels for each image), works as the receptor of input images. The middle layer is used for processes, encoding and network learning, while the last layer, which is composed of 10 neurons (as we have 10 distinct classes), does the job of prediction and classification of images. The model was implemented using MATLAB and we have used Hoda Persian handwritten digits dataset as our input images. The obtained results show that the implemented model can carry out, with good accuracy (95%), the learning and classification of images of handwritten digits with high levels of noise.