A Plagiarism Detection Engine for Images in Docode

Abstract

Plagiarism is turning in someone else's work as your own. Though tools exist for checking the originality of texts, those do not work with images, which can be plagiarized as well. In previous works, we have developed the Docode plagiarism detection system, which works with text, but, as commercial requirements have evolved, it is necessary for it to be able to work with images. In this work we present a plagiarism detection engine for images, which works by fusing texture and color features in a weighted combination, so that it can work as a general-purpose engine. We ran experiments with the system, analyzing the improvement made by fusing color and texture features, and the impact of downsizing images on the performance of the system. We see that we reach a recall at 10 elements of 80% and the system allows an image compression of one half without considerable impact on its performance, and with this we can conclude we can build a plagiarism detection engine for images, able to handle general collections for its integration in the Docode system.