An application of pre-trained CNN for image classification

Abstract

Image Classification is a branch of computer vision where images are classified into categories. This is a very important topic in today's context as large databases of images are becoming very common. Images can be classified as supervised or unsupervised techniques. This paper investigates supervised classification and evaluates performances of two classifiers as well as two feature extraction techniques. The classifiers used are Linear Support Vector Machine (SVM) and Quadratic SVM. The classifiers are trained and tested with features extracted using Bag of Words and pre-trained Convolution Neural Network (CNN), namely AlexNet. It has been observed that the classifiers are able to classify images with very high accuracy when trained with features from CNN. The image categories consisted of Binocular, Motorbikes, Watches, Airplanes, and Faces, which are taken from Caltech 265 image archive.