Automatic image classification for content filtering

The paper presents a survey of methods for image classification in the context of Web content filtering, as well as results of new experiments using convolutional neural networks and SVM with bag-of-visual-words. Within the experiments, a special difficult dataset was collected that consisting of two hardly distinguishable classes. The quality achieved using convolutional neural network is higher than that of traditional methods in the complicated conditions. Thus, the classifier based on convolutional neural networks proved to be very useful for purposes of Web content filtering.