Improved Accuracy in Image Tampering Detection using MobileNet

Recent developments in the technology and the exponential increase in accessing the digital networks have increased over the past few years. With this expansion, the data that is being accessed by the remote devices have also increased. This scenario is more pertinent over the images or image-based data which are used by unauthenticated users. It is relatively simple for unscrupulous intruders to subtly alter these digital photographs for illegal transactions. Therefore, preventing the modification of images have become a significant concern. This study mainly concentrates on the development of novel technique for the protection of medical images from tampering. It is proposed to use MobileNet for classification. Like the inception towers, MobileNets use depth-wise separable convolutions. These cut down the latency by decreasing the number of a parameters. Moreover, MobileNets have convenient model-shrinking parameters that can be called before training to achieve the desired precise size. Results show that the accuracy of the proposed technique has increased twice when compared to the existing ResNet based architecture.