Image and video forgeries are becoming more advanced with the latest developments of artificial intelligence and other digital editing software that have highly developed and take the artificial intelligence and the digital editing software to another level that becomes impossible to identify without the need to intervene in the data. This becomes a serious challenge to the forensic investigation department, journalism and cyber security. Although they are very popular, traditional convolution neural networks (CNNs) have serious limitations in identifying the most delicate manipulations because they do not allow the preservation of spatial hierarchies. To fill in this hole, we introduce a hybrid tamper detection system based on combining Capsule Networks and Error Level Analysis (ELA). CapsNet preserves spatial relation to detect the fine-grained anomalous areas, whereas ELA focuses on non uniformities throughout the compression artifacts to determine the areas where tampering has occurred. The suggested approach was tested on the common forensic datasets and showed higher accuracy and resilience than conventional CNN-based models in both the detection of image and video forgery. The study presents a stable and understandable method of authenticating digital media that can be used to provide a robust defense against misinformation, cybercrime, and deepfake filtering.
Borgaonkar et al. (Wed,) studied this question.