DETECTION OF IMAGE FORGERY FOR COPYRIGHT APPLICATIONS USING DEEP LEARNING

Abstract

Technical evolution of the world is conquering; the trust on digital imaging technology is grinding down. Image forgery is the process of manipulating/ changing an original photo into different kinds. Now a day, people can easily tamper/forged image from tabloid magazines to the business industry due to the availability of powerful image processing & editing software i.e., Photoshop, ai etc. It could be done by the use of various techniques such as copy move, splicing of images, and image retouching. Image tampering can be a creative work. However, in some cases it is not good, since images appear to be proof for medical reports, crime scenes, etc. The result is death of patients and escape of criminals, respectively. Conventional methods in the literature for image forgery detection are based on traces developed while manipulations are performed on images. Most of such traces are confined to the scope of predefined assumptions about handcrafted features, size, and contrast. This paper proposes a fusion-based decision approach for the detection of forged images. So, the lightweight deep learning models employed in this fusion decision technique are Squeeze Net, MobileNetV2, and Shuffle Net. In this regard, the system of fusion decision will be implemented in two phases. First of all, the pre-trained weights of lightweight deep learning models were considered to evaluate the forgery of the images. Then, the results were compared with the fine-tuned weights of forgery of the images with pre-trained models. Experimental results prove that the proposed fusionbased decision approach ensures better accuracy than state-of-the-art approaches.