Image Copy Detection Based on Convolutional Neural Networks

In this paper, we present a model that automatically differentiates copied versions of original images. Unlike traditional image copy detection schemes, our system is a Convolutional Neural Networks (CNN) based model which means that it does not need any manually-designed features. In addition, a convolutional network is more applicable to image copy detection whose architecture is designed for robustness to geometric distortions. Our model uses fully connected layers to compute a similarity between CNN features, which are extracted from image pairs by a deep convolutional network. This method is very efficient and scalable to large databases. In order to see the comparison visually, a variety of models are explored. Experimental results demonstrate that our model presents surprising performance on various data sets.