Image Forgery over Social Media Platforms - A Deep Learning Approach for its Detection and Localization

Social media platforms play a significant role in spreading news in the current digital era. However, they have also been spreading fake images. Forged images posted on social media platform such as Twitter create misrepresentation and generate harmful user emotions. Thus, detecting fake images over social media platforms has become a critical need of time. Deep learning convolutional networks can learn the intrinsic feature set of images and can detect forged images. This paper proposes a convolutional neural network to spot fake images shared over social media platforms. High pass filters from image processing are used in the first layer for weight initialization. This helps the neural network converge faster and achieve better accuracy. Interpretability is a common concern in deep learning models. The proposed framework employs Gradient-weighted Class Activation Mapping to generate heatmaps and localize the image's manipulated area. The model is verified against the publicly available CASIA dataset. An accuracy of 92.3% is achieved, which is better than the other previous models. From the social media perspective, the model is verified against the latest Twitter dataset. The experiment proves that convolutional neural networks perform well in detecting forged images over social media platforms, and interpretability can be achieved.