Recognizing objectionable images using convolutional neural nets

摘要

In recent years different methods for detecting objectionable images have proposed. All of the previous systems are based on extracting pre-defined and certain features from the images. In this paper a method is proposed in order to detect objectionable images using convolutional neural networks. In this method first features are learned through a sparse auto-encoder and then training is done by a convolutional neural network. The architecture of the network consists of convolution and sub-sampling layers followed by a fully connected output layer which feeds a softmax classifier with cross entropy cost function. The proposed method is able to effectively detect 90.5% of images correctly employing a rather small training dataset.