Convolutional Neural Network based Digital Image Forensics using Random Forest and SVM Classifier

Recent fast advances in computer graphics and information technology have increased the quality of created digital material. Social media platforms such as Instagram, Facebook, and others, as well as movies and games, have taken use of this phenomenon; nonetheless, these artificial improvements have had severe detrimental consequences for society, such as those produced by phony photographs created with evil purpose. Hardly some strategies attain excellent accuracies for all genre of alterations such as copy-move, compression, splicing, rotation, etc. For ascertaining tampering efficiently, it is utmost of interest to devise a feature extraction mechanism based on deep learning, that ascertain correlation among pixels and evaluate more proficiently. Contrasting with recent divergent surveys, this paper enfolds notable developments in techniques of passive image forensic analysis employing deep learning techniques. Furthermore, a certain feature can only handle one type of forgeries. To overcome these challenges, this research proposes a technique based on Convolution Neural Networks (CNN) as a feature extractor and Random Forest (RF) and Support Vector Machine (SVM) as classifiers. The suggested technique was built and tested using two distinct CNNs, AlexNet and VGG19. CNN is trained on the CASIA dataset, then classify using RF and SVM individually and compare the results. Because the CNN is trained to deal with images of varying sizes, the suggested system is resistant to operations such as noise addition, blurring, and JPEG compression. The CNN-RF and CNN-SVM frameworks are assessed on the CASIA dataset and outperform state-of-the-art approaches.