Utilising LSTM Based Network for Image Forgery Detection

• Published in: 2023 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI) pp. 1 - 9
• Main Authors: Jaiswal, Arunima, Parmar, Aradhana, Sachdeva, Nitin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25-05-2023
• Subjects: Computational modeling; Convulational Neural Networks; Copy-Move Forgery; Deep learning; Forgery; Heating systems; Image Forgery Detection; Image segmentation; Machine Learning; Radon; Social networking (online)
• DOI: 10.1109/ACCAI58221.2023.10199262

Image Forgery refers to doing various kinds of modifications to the original image such that the actual meaning of it gets changed, due to which false information is spread everywhere. Moreover in this era where social media is at huge heights and images have become one of the most integral part of everyones life ,people tend to believe what they see and hence it's a very serous concern which needs a solution. There are many techniques present to detect image forgeries .In our research we have shown two major techniques to detect forgeries using resampling techniques. As Resampling is a key indicator of altered images, two approaches for detecting and localising picture modifications, employing resampling (by forming different samples) varied integral mechanisms. First approach, the Radon convert of these features is computed on all those image parts which are overlapping to each other, then a heatmap generated using classifiers of deep learning appproach and also gaussian random variable ,which is model for conditional probability values . To locate tampered sections, the Random Walker segmentation approach is utilised. For categorisation and identification of local tampered regions, the second method includes transmitting resampling characteristics computed on intersecting picture patches through an LSTM-based network. We have compared both approaches detection/localization performance and it shows that both strategies are effective at detecting and localising image forgeries of all levels at a very efficient level, according to the results of our research.