An approach for copy-move image multiple forgery detection based on an optimized pre-trained deep learning model

Abstract

Owing to the rapid increase in the availability of editing tools and image capturing, the amount of fake images available online has increased tremendously. Moreover, social media and other networks have become the foremost channel, which is designed for transmitting manipulated images, propagating rumours, spreading fake news, and so on. Thus, developing effective techniques for detecting such forgeries have become highly essential. One of the most widespread kinds of forgery is Copy Move Forgery (CMF), which alters the image by using the patches inside the image. Though deep learning-based techniques offer superior performance, they suffer from generalization issues. Therefore, a transfer learning-based method for identifying CMF utilizing a Deep Convolutional Neural Network (Deep CNN) is proposed to overcome the issue. Deep CNN is trained with the parameters of the pre-trained GoogLeNet, and is employed for detecting multiple forgeries. Furthermore, a novel optimization algorithm, Fractional Leader Harris Hawks Optimization (FLHHO), is created to modify the weights and bias of the Deep CNN. Further, the presented technique is evaluated for its effectiveness on the basis of the parameters like testing accuracy of 0.930, True Negative Rate (TNR) of 0.938, and True Positive Rate (TPR) of 0.941.