A Novel Image Falsification Detection Using Vision Transformer (Vi-T) Neural Network

Abstract

Image classification and image recognition are significantly impacted by the creation of Vision Transformers. It is no wonder that CNN architectures utilize multiple layers and a considerable number of hyper-parameters are necessary for training, as they are much more complex. While training and implementing, CNN uses a considerable number of resources. Alternately, Vision Transformers are the innovative architecture of neural networks that allows it to take in an image and break it into little pieces, then use an attention mechanism to search for correlations in the pieces. For localization of the attention mechanism, the transformers are well trained on small visual patches. With the introduction of image manipulation tools, much of the picture data available on the Internet today are designed to fool consumers into thinking the image is genuine. To authenticate photographs, we must utilize many different neural networks. The main purpose of this research was to detect forgery and locate the tampered image utilizing the transformers and attention mechanism concept. Using the RTX-3080 graphic card, the study effort is implemented on two benchmark datasets. These datasets are used to get the evaluation results, and they are compared to the state-of-the-art techniques. During the training and validation process, training accuracy of 98% and validation accuracy of 97% are achieved on benchmark datasets.