Camera model identification based on forensic traces extracted from homogeneous patches

A crucial challenge in digital image forensics is to identify the source camera model used generate given images. This of prime importance, especially for Law Enforcement Agencies their investigations Child Sexual Abuse Material found darknets or seized storage devices. In this work, we address by proposing a solution that characterized two main contributions. It relies on extraction rather small homogeneous regions extract very efficiently from integral image, and hierarchical classification approach with convolutional neural networks as underlying models. We rely they contain traces are less distorted than high-level scene content. The propose important scaling up making minimal modifications when new cameras added. Furthermore, scheme performs better traditional single classifier approach. By means thorough experimentation publicly available Dresden data set, achieve an accuracy 99.01% 5-fold cross-validation ‘natural’ subset set. To best our knowledge, result ever reported • identification using ConvNets based patches. consists (brand, model, device) scheme. Natural images sufficient forensic trace extraction. 18 models With result, outperform state-of-the-art same