Abstract:In this work, we deal with the problem of re compression based image forgery detection, where some regions of an image are modified illegitimately, hence giving rise to presence of dual compression characteristics within a single image. There have been some significant researches in this direction, in the last decade. However, almost all existing techniques fail to detect this form of forgery, when the first compression factor is greater than the second. We address this problem in re compression based forgery detection, here Recently, Machine Learning techniques have started gaining a lot of importance in the domain of digital image forensics. In this work, we propose a Convolution Neural Network based deep learning architecture, which is capable of detecting the presence of re compression based forgery in JPEG images. The proposed architecture works equally efficiently, even in cases where the first compression ratio is greater than the second. In this work, we also aim to localize the regions of image manipulation based on re compression features, using the trained neural network. Our experimental results prove that the proposed method outperforms the state of the art, with respect to forgery detection and localization accuracy.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper is mainly dedicated to solving the problem of JPEG image forgery detection and localization based on recompression. Specifically, when certain areas of an image are illegally modified, these areas will exhibit double - compression characteristics, that is, there are two different compression quality factors (QF1 and QF2) in the image. Existing techniques often fail to effectively detect such forgery when dealing with this situation, especially when the first compression factor is greater than the second compression factor (QF1 > QF2). To overcome this challenge, the author proposes a deep - learning architecture based on convolutional neural network (CNN), which can detect recompression forgeries in JPEG images and accurately locate the tampered areas. Moreover, this method also performs well in the case of QF1 > QF2, which is difficult for many existing techniques to achieve. ### Main contributions of the paper 1. **Proposing a new CNN architecture**: This architecture can detect recompression forgeries in JPEG images and also work effectively in the case of QF1 > QF2. 2. **Improving the accuracy of forgery detection and localization**: Verified by experiments, this method is superior to the existing state - of - the - art methods in terms of detection and localization accuracy. 3. **Introducing 8×8 - pixel localization units**: By using a smaller stride, the localization accuracy of the forged area reaches the 8×8 - pixel level, significantly improving the fineness of detection. ### Summary of mathematical formulas - **Accuracy**: \[ \text{Accuracy}=\frac{|TP| + |TN|}{|TP|+|TN|+|FP|+|FN|} \] - **F - measure**: \[ F\text{-measure}=\frac{2\times\text{Precision}\times\text{Recall}}{\text{Precision}+\text{Recall}} \] where \[ \text{Precision}=\frac{|TP|}{|TP|+|FP|},\quad\text{Recall}=\frac{|TP|}{|TP|+|FN|} \] - **Average F - measure**: \[ F\text{-measure}=\frac{1}{N}\sum_{i = 1}^{N}F\text{-measure}(i) \] Through these formulas, the author evaluates the performance of the proposed CNN model in detecting and localizing recompression forgeries and shows its superiority.

Recompression Based JPEG Tamper Detection and Localization Using Deep Neural Network Eliminating Compression Factor Dependency

Image Tampering Localization Via Estimating the Non-Aligned Double JPEG Compression

A Robust JPEG Image Tampering Detection Method Using GLCM Features

Block-level Double JPEG Compression Detection for Image Forgery Localization

Forgery Detection Using Polynomial Fitting in Recompressed JPEG Images

JPEG Compression-aware Image Forgery Localization

Deep Multi-scale Discriminative Networks for Double JPEG Compression Forensics

Hybrid LSTM and Encoder–Decoder Architecture for Detection of Image Forgeries

Improving the Efficiency of Image and Video Forgery Detection Using Hybrid Convolutional Neural Networks

Detection of Image Tampering Using Deep Learning, Error Levels and Noise Residuals

DCT-domain deep convolutional neural networks for multiple JPEG compression classification

CNN based image resizing forensics for double compressed JPEG images

Detection of Double JPEG Compression Using Modified DenseNet Model

Learning Compressed Artifact for JPEG Manipulation Localization Using Wide-Receptive-Field Network

Dual branch convolutional neural network for copy move forgery detection

Anti-Compression Contrastive Facial Forgery Detection

Two-stream Encoder-Decoder Network for Localizing Image Forgeries

Cloud-Based Fusion of Residual Exploitation-Based Convolutional Neural Network Models for Image Tampering Detection in Bioinformatics

Extraction of image resampling using correlation aware convolution neural networks for image tampering detection

Image Forgery Detection & Localization Using Regularized U-Net

MDCF-Net: Multi-Scale Dual-Branch Network for Compressed Face Forgery Detection