lyphilong / test_adaptive_cfa_forensics

Requirements

python >= 3.6

PIL pytorch numpy tqdm matplotlib scikit-learn

In addition, the following files require pytorch with CUDA enabled: detect_forgeries_interactive.py detect_forgeries_multiple.py train_model.py

Dataset

The dataset can be found here.

Download the dataset to folder input

!wget http://dev.ipol.im/~qbammey/dresden_demosaicing_forgery_detection_dataset_v1/dresden_demosaicing_forgery_detection_dataset_v1.zip
!unzip dresden_demosaicing_forgery_detection_dataset_v1.zip -d input/

Usage

Train or retrain a network

train_model.py [-h] [-m MODEL] [-j JPEG] [-b BLOCK_SIZE] [-o OUT]
                      [-l LEARNING_RATE] [-a EPOCHS_AUXILIARY]
                      [-B EPOCHS_BLOCKWISE] [-s BATCH_SIZE]
                      input [input ...]
Ex: train_model.py 'input/images'

To use a pretrained network and retrain it on data, specify the pretrained model with -m. All images are kept in GPU memory at the same time. As a consequence, training on a large database require more GPU memory.

Detect forgeries on a single image with the proposed method

detect_forgeries.py [-h] [-m MODEL] [-j JPEG]
                                    [-o OUT]
                                    [-b BLOCK_SIZE]
                                    input

The model can be specified with -m. By default, uses the pretrained model (not retrained on the database). If the output image path is not specified, results will be plotted interactively.

Detect forgeries on multiple images with the intermediate values method:

choi_intermediate_values.py [-h] [-j JPEG] [-b BLOCK_SIZE] [-o OUT]
                                   input [input ...]

This is an implementation of the method described in Choi, C., Choi, J., & Lee, H. (2011). CFA pattern identification of digital cameras using intermediate value counting. MM&Sec'11.

Detect forgeries on multiple images with the variance of colour difference method:

shin_variance.py [-h] [-j JPEG] [-b BLOCK_SIZE] [-o OUT]
                        input [input ...]

This is an implementation of the method described in Hyun Jun Shin, Jong Ju Jeon, and Il Kyu Eom "Color filter array pattern identification using variance of color difference image," Journal of Electronic Imaging 26(4), 043015 (7 August 2017). https://doi.org/10.1117/1.JEI.26.4.043015

Citation

Please cite the following if you use our work in your research:

@InProceedings{Bammey_2020_CVPR,
author = {Bammey, Quentin and Gioi, Rafael Grompone von and Morel, Jean-Michel},
title = {An Adaptive Neural Network for Unsupervised Mosaic Consistency Analysis in Image Forensics},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}