Digital Image Forgery Detection Using Convolutional Neural Network

Abstract

Digital images are a main source of shared information in social media. Digital image forgery has become a growing concern with the advancement of image editing tools, leading to the spread of misleading and manipulated content. Detecting such forgeries is crucial for ensuring the authenticity and reliability of digital images. Various digital image forgery detection techniques are tied to detecting only one type of forgery, such as image splicing or copy-move it is not applied in real life. To enhance digital image forgery detection using deep learning techniques via transfer learning is uncover two types of image forgery at the same time. This study proposes a deep learning-based approach utilizing Convolutional Neural Networks (CNNs) to detect forged images effectively. The proposed model is trained to identify common types of image forgeries, including copy-move, splicing, and retouching. By leveraging CNNs, the model extracts spatial and texture-based features to distinguish forged regions from authentic ones. The dataset used for training and evaluation consists of both real and manipulated images, ensuring robust performance across different types of forgeries. Experimental results demonstrate that the proposed CNN-based model achieves high accuracy in forgery detection compared to traditional image processing methods. This research highlights the potential of deep learning techniques in forensic image analysis and contributes to the advancement of automated forgery detection systems.

Introduction

Overview

Digital image forgery involves manipulating images to deceive viewers. Common methods include:

• Copy-move

• Splicing

• Retouching

Traditional detection techniques are manual, time-consuming, and struggle with complex forgeries. In contrast, Convolutional Neural Networks (CNNs) offer automatic feature extraction, high accuracy, scalability, and adaptability across various forgery styles and datasets.

Why CNNs?

CNNs are a type of deep learning model well-suited for image analysis. They work through layers:

• Input Layer: Receives raw image pixels

• Convolution Layers: Extract features (edges, patterns) using filters

• Activation Functions: Add non-linearity to learn complex patterns

• Pooling Layers: Downsample data to retain essential features and reduce computation

• Fully Connected Layers: Final classification or prediction

• Output Layer: Produces results (e.g., forged or original)

CNNs learn patterns automatically and are trained using:

• Loss functions, backpropagation, and optimization techniques to fine-tune weights.

Literature Review Highlights

Implementation Details

A. Modules

1. Dataset Collection

   • Used the CASIA2 dataset (12,615 images) from Kaggle.

2. Libraries Used

   • Python with Keras, TensorFlow, sklearn, PIL, numpy, etc.

3. Image Preprocessing

   • Resize to 200x200 pixels, normalize pixel values, convert to NumPy arrays.

4. Error Level Analysis (ELA)

   • Detects image manipulations by analyzing differences from JPEG compression.

   • Modified regions show higher ELA values than original areas.

5. Data Splitting

   • 80% for training, 20% for testing.

6. Model Building

   • CNN architecture: Sequential model with Conv2D, MaxPooling, Dropout, Flatten, and Dense layers.

   • Used Softmax activation in the final layer for binary classification (forged vs. original).

   • Achieved 98% training accuracy.

B. CNN Concepts Explained

• Padding: Maintains output size by adding borders. Prevents shrinking and preserves corner features.

• Stride: Controls filter movement. Larger strides reduce image size.

• Volume Convolution: Handles 3D image inputs (e.g., RGB) using filters with matching depth.

• Pooling: Max or average pooling to reduce dimensionality and computation.

• CNN Layers: Multiple convolution and pooling layers followed by fully connected layers make up the network.

C. Model Evaluation & Deployment

1. Performance Evaluation

   • Accuracy on test set: 92%

   • Plotted accuracy and loss over training epochs.

2. Saving the Model

   • Saved the trained model using .pkl or .h5 formats with pickle.

3. Detection Pipeline Summary

   • Data collection → ELA analysis → Preprocessing → CNN Training → Evaluation → Model Saving

Advantages of CNN-Based Forgery Detection

• High accuracy in identifying subtle manipulations

• Automatic feature learning (no manual extraction)

• Fast and scalable for large datasets

• Adaptable to various forgery types and evolving threats

Conclusion

The effectiveness of CNNs in automatically learning intricate patterns and features associated with forged regions, achieving high accuracy and robustness across diverse datasets. It eliminates the need for manual feature extraction, making it more adaptable to various forgery types and resolutions. The potential of CNN-based systems as a reliable and scalable solution for digital image forensics, with significant applications in media authenticity verification and cybersecurity.

References

[1] Bayar, B., & Stamm, M. C. (2016). A deep learning approach to universal image manipulation detection using a new convolutional layer. ACM Workshop on Information Hiding and Multimedia Security, 5-10. [2] Rao, Y., & Ni, J. (2016). A deep learning approach to detection of splicing and copy-move forgeries in images. IEEE International Workshop on Information Forensics and Security (WIFS), 1-6. [3] Zhou, P., Han, X., Morariu, V. I., & Davis, L. S. (2018). Learning rich features for image manipulation detection. IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1053-1061. [4] Bappy, J. H., Roy-Chowdhury, A. K., Bunk, J., Nataraj, L., & Manjunath, B. S. (2017). Exploiting spatial structure for localizing manipulated image regions. IEEE International Conference on Computer Vision (ICCV), 4970-4979. [5] Huh, M., Liu, A., Owens, A., & Efros, A. A. (2018). Fighting fake news: Image splice detection via learned self-consistency. European Conference on Computer Vision (ECCV), 101-117. [6] Verdoliva, L., Poggi, G., & Cozzolino, D. (2020). A feature-based approach for image tampering detection and localization. IEEE Transactions on Information Forensics and Security, 15, 1231-1245. [7] Bi, Q., Liu, X., & Wen, Y. (2019). Video forgery detection based on convolutional neural networks and long short-term memory. Journal of Visual Communication and Image Representation, 58, 567-575. [8] Zhuang, P., Ni, J., & Zhao, Y. (2019). Image forgery detection via convolutional neural networks and multi-view feature learning. IEEE Transactions on Information Forensics and Security, 14(10), 2517-2532.

Copyright

Copyright © 2025 Kandukuri Nirosha, Dr. D. Vivekananda Reddy. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.