Proposal: Hybrid ViT-CNN Model for Copy-Move Forgery Detection
I propose developing a hybrid Vision Transformer (ViT) and Convolutional Neural Network (CNN) model to detect copy-move forgery in images that have undergone various post-processing transformations, such as scaling, rotation, flipping, blurring, color adjustments, and noise additions.
Approach:
ViT (Vision Transformer): ViT will be used to capture global contextual information and long-range dependencies within the image using self-attention mechanisms.
CNN (ResNet): A ResNet CNN will extract local features, detecting fine details like edges and textures. The CNN will help identify small-scale transformations such as noise or blurring.
Feature Matching: After feature extraction, we will use feature matching techniques to identify copied regions and handle transformations such as rotation and scaling.
Post-processing Detection: The model will be trained to recognize forgeries that have undergone transformations like flipping, blurring, and color adjustments.
The model will be trained on publicly available datasets (e.g., Columbia, CASIA, CE-Forge) and evaluated using accuracy, precision, recall, F1-score, and AUC. The project will be completed in approximately 9 weeks.
With expertise in machine learning and computer vision, I’m confident in delivering a robust, high-performance solution for forgery detection.