Double JPEG compression forensics based on a convolutional neural network

Qing Wang,Rong Zhang

doi:10.1186/s13635-016-0047-y

Abstract

Double JPEG compression detection has received considerable attention in blind image forensics. However, only few techniques can provide automatic localization. To address this challenge, this paper proposes a double JPEG compression detection algorithm based on a convolutional neural network (CNN). The CNN is designed to classify histograms of discrete cosine transform (DCT) coefficients, which differ between single-compressed areas (tampered areas) and double-compressed areas (untampered areas). The localization result is obtained according to the classification results. Experimental results show that the proposed algorithm performs well in double JPEG compression detection and forgery localization, especially when the first compression quality factor is higher than the second.

Highlights

Blind forensics techniques utilize statistical and geometrical features, interpolation effects, or feature inconsistencies to verify the authenticity of image/videos when no prior knowledge of the original sources is available
To demonstrate the superiority of our method, we first compare the accuracy of double JPEG compression detection for different block sizes with the method proposed in [5], which detects double-compressed JPEG images by using mode-based first digit features, and the method proposed in [6], which uses the statistics of lowfrequency discrete cosine transform (DCT) coefficients
For composite JPEG image detection, we compare our localization results with the method proposed in [11], which achieves localization based on the double quantization (DQ) effects combined with Bayes’s theorem, and with the technique proposed in [12] based on Benford’s law using support vector machine (SVM)

Summary

Introduction

Blind forensics techniques utilize statistical and geometrical features, interpolation effects, or feature inconsistencies to verify the authenticity of image/videos when no prior knowledge of the original sources is available. We use the histograms of DCT coefficients as the input to the CNN that is designed to automatically learn the features of these histograms and perform classification for single and double JPEG compression. 3.2 Locating tampered regions To achieve localization, the image of interest, I, with a resolution of M × N, is divided into overlapping blocks with a dimension of W × W, and an overlapping stride of 8 pixels, the size of the blocks on which DCT is performed during JPEG compression. To demonstrate the superiority of our method, we first compare the accuracy of double JPEG compression detection for different block sizes with the method proposed in [5], which detects double-compressed JPEG images by using mode-based first digit features, and the method proposed in [6], which uses the statistics of lowfrequency DCT coefficients. We only select some representative QF1 and QF2 for experiment to show the effectiveness of the CNN classifiers

70 Proposed

Proposed Method

Conclusions