Image Classification based on Convolution Neural Network

Abstract

Apart from image processing techniques, the problem of object recognition can also be solved by using machine learning techniques. The main concern of this thesis is to classify images using machine learning techniques. To tackle with such problems, artificial neural network i.e. Convolutional Neural Network has been developed. In order to design the Convolutional Neural Network, different parameters like filter size, number of convolution layers, drop out layers etc were taken. Careful choosing and study of these parameters shows that efficient architecture can be designed. Different neural network architectures for CIFAR-10 and MNIST dataset have been developed. Being different in terms of number of hidden layers, filter size and other measures, they have been trained on Central Processing Unit. Drop out techniques has been used to reduce over-fitting issues. Accuracy of these architectures has been calculated by feeding the networks with the test data. Lastly, results are compared and analyzed to find out best architecture. Thus, this study gives a way to design efficient architecture for image classification.