Provenance Inference for Instagram Photos Through Device Fingerprinting

摘要: Sensor pattern noise (SPN) has been extensively studied in the scientific community and found its applications many practical scenarios law-enforcement sector. However, emergence of photo-sharing social networking sites (SNSs) poses new challenges to SPN-based digital image provenance analysis. One particular issue is that SNSs' built-in editing tools tend inflict distortion on SPNs. well-known example such filters Instagram. We observed some Instagram manipulate high-frequency bands images hence damage SPNs, making source-oriented clustering (SOC) filtered unsatisfactory. To address this issue, we propose first separate processed by different beforehand into two groups, with Group Malignant (M) containing significantly distort SPNs Benign (B) covering other have no significant impact then cluster B calculate centroid each cluster, one representing reference SPN corresponding camera. Finally, use attract M order complete SOC task. identify filter applied so as facilitate clustering, a convolutional neural network based filter-oriented classifier proposed. Tested 19,332 18 filters, delivers very promising accuracy 98.5%. Moreover, compared F1-measure 47.74% directly 1,800 images, our proposed framework achieves much higher 90.33%.