Lightweight Deep Learning Model for Detection of Copy-Move Image Forgery with Post-Processed Attacks

As digital image forgery can be alarmingly detrimental, therefore, an insight into detection and classification of tampered digital images is of paramount importance. Without undermining the significance of other image forgery types, copy-move can be regarded as one of the most commonly used forgeries due to its ease of implementation. To counter the rapidly complicating forgery methods due to easily accessible technologically advanced tools, passive image forensic methods have also undergone massive evolution. Presently, deep learning based techniques are regarded as state-of-the-art for image processing/image forgery detection and classification due to their enhanced accuracy and automatic feature extraction capabilities. But the existing deep learning based techniques are time and resource-intensive as well. To cater for these solutions with complexities as stated, this research focuses on experimentation using two state-of-the-art deep learning models; SmallerVGGNet (inspired from VGGNet) and MobileNetV2. These two models are time and resource friendly deep learning frameworks for digital image forgery detection on embedded devices. After rigorous analysis, the study considers a suitably modified version of MobileNetV2 to be more effective on copy-move forgery detection which also caters for inconsistencies executed post-forgery including visual-appearance related such as brightness change, blurring and noise adding and geometric transformations such as cropping and rotation. The experimental results demonstrate that the proposed MobileNetV2 based model shows 84% True Positive Rate (TPR) and 14.35% False Positive Rate (FPR) for the detection of digital image forgery post-processed with the said multiple attacks.