Enhancing Image Forgery Detection
A Comprehensive Approach using ELA, CNN, and VGG16
Overview
Our project detects image forgery using advanced techniques such as Error Level Analysis with Convolutional Neural Network (ELA-CNN) and ELA with a pre-trained VGG16 model. The combination of ELA and deep learning models improves the accuracy of forgery detection in digital content.
Features
ELA-CNN Model:
A custom Convolutional Neural Network (CNN) for feature extraction and forgery detection. ELA is used to identify tampering by analysing compression level inconsistencies. It is trained on CASIA 2.0 dataset, a widely used collection for image forensics research.
ELA + VGG16 Model:
We have integrated ELA with a pre-trained VGG16 model for improved performance. VGG16 architecture consists of 16 weight layers that have been fine-tuned for forgery detection. We have leveraged the ImageNet dataset for pre-training and feature extraction.
Usage
Link for dataset https://drive.google.com/file/d/1IDUgcoUeonBxx2rASX-_QwV9fhbtqdY8/view
Link for colab notebooks ELA + CNN : https://colab.research.google.com/drive/1K_mWdGZvkIqO0HeuZq_uTcwJNdc1P_pa?usp=sharing
ELA + VGG16 : https://colab.research.google.com/drive/1h5-EsaKC2Ih5hec1Z8_XkRfrz0n0oIRC?usp=sharing
Upload the dataset to drive, mount on google colab and run the code.
Results
ELA-CNN Accuracy: 100%
ELA + VGG16 Accuracy: 100%
Efforts by:
Harshita Badhan
Ishpreet Kaur
Sandhya Yadav