Convolutional neural networks for the detection of nearly identical high quality recompression

Supervisor(s)Dr. Cecilia Pasquini


Recompression detection is a research field in image forensics, as the number of recompressions might indicate a forged image. Recompressing an image with (nearly) the same high quality factor is thought of as one of the toughest cases in recompression detection. The only method known to provide good detection accuracy in that case is the one based on block convergence.

Convolutional neural networks (CNNs) become more and more popular in areas such as image forensics, as they are supposed to identify high-order correlations in a dataset without being explicitly told where to look.

The goal of this thesis is to implement and train a CNN for the detection of nearly identical high quality JPEG recompression and compare its performance to that of the conventional method based on block convergence.


  • Lai, S.-Y. and Böhme, R. Block Convergence in Repeated Transform Coding: JPEG-100 Forensics, Carbon Dating, and Tamper Detection. In IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP). Vancouver, Canada, 2013, pp. 3028–3032. [PDF]
  • Pasquini, C., Schöttle, P., Böhme, R., Boato, G., and Pèrez-Gonzàlez, F. Forensics of High Quality and Nearly Identical JPEG Image Recompression. In ACM Information Hiding and Multimedia Security Workshop. Vigo, Galicia, Spain, 2016, pp. 11–21. [Publisher]