Detection of copy-move forgery with deep CNN features using Machine learning Classifier

An image forgery detection method detects and locates forged components from manipulated images. Identifying whether an image is forged or non-forged requires a sufficient number of features to detect manipulation or tampering. The patch descriptor uses a convolutional neural network (CNN) that has already been trained to extract detailed characteristics from images in an efficient and highly effective manner. An eventual discriminative feature for SVM classification is obtained via a feature fusion technique. In this paper, we combine CNN with the SVM to image forgery detection. CNN is used to achieve high accuracy as they learn to recognize patterns from the large data sets of images while SVM is used to find the hyper plane that separate the classes in feature space. The training and testing phases utilize the same input patches, the SVM, as well as the fully connected layer and SoftMax layer from the images. The experimental results of the proposed method are compared with existing state-of-the-art techniques and then demonstrated that the proposed approach has achieved 98.91% accuracy, which indicates that proposed model is both effective and adaptable.