Image spam classification based on convolutional neural network

Abstract

Image classification is a fundamental problem in computer vision and pattern recognition. Feature extraction is often regarded as the key for classifying images. Traditional ways rely on handcrafted features heavily, such as SIFT and BoW. In this paper, we concentrate on recognizing some specific categories of images (e.g. adult content and political images) in Email. And most importantly we propose a novel architecture of Convolutional Neural Network (CNN) to apply on image classification. In particular, most of the previous methods often use softmax activation for prediction, while instead of utilizing a softmax layer, we present a linear support vector machine (SVM) to be a part of the learning model, minimizing a margin-based loss to obtain a lower level feature representation. Accordingly, we propose a new dataset which contains 7 categories and over 52934 images totally. Experimental results on this novel dataset validate the effectiveness of our method.