Image forgery detection using Convolutional Neural Networks

Abstract

Digital forensics vital aspect of picture identity theft has drawn a lot of notice recently. In order to establish the primitive character of images, earlier studies looked at residual pattern noise, wavelet-transformed data and facts, image pixel resolution histograms, and additional characteristics of images. In an attempt to attain high-level picture illustration with the advancement of neural network-based innovations, convolutional neural networks have recently been utilized for recognizing image counterfeiting. This model suggests constructing a convolutional neural network with a structure that is distinct from previous studies in which we attempt to interpret the features derived from each layer of convolution to recognize a variety of picture manipulation using automated feature recognition. Three convolutional layers, one fully interconnected layer, and a SoftMax classifier constitute the suggested system. Our study utilizes our own data collection as the training data, which includes genuine pictures, spliced images, and further enhanced replicates with retouched and re-compressed images. Experimental findings make it abundantly obvious that the proposed network is optimal and versatile.