FATE:Fingerprints Automatically Targeting and Extracting for image source identification

摘要

The convenience of obtaining perfect photos with mobile devices not only promotes the quality of modern life, but also facilitates privacy hacker for crimes like taking images secretly. Such dilemma causes widely concerns and calls for forensic community to reverse-engineer the evil images, which can be seen as a source identification problem. In recent years, convolutional neural network (CNN) has jumped to the crown jewel due to its high performance in image recognition. In this light, we devise a CNN-based system named FATE (Fingerprints Automatically Targeting and Extracting) which can identify the image source out of various models of mobile devices. To verify our FATE system, we first set up a dataset consisted of images from six different model mobile devices, then we train FATE on our dataset and test it with real data. Besides, we introduce another two stat-of-the-art CNN models for comparison. The results show that our FATE beats down the rivals and achieves an accuracy of 98.5% in recognition.