DEEP FORGERY DETECT: USING DEEP LEARNING-BASED FORGERY DETECTION TO IMPROVE SOCIAL MEDIA SECURITY
Keywords:
Images, Videos, Social media networks, Custom CNN,, LSTMAbstract
These days, surveillance cameras play a major role in both legal and security applications. Nonetheless, the images
and video recordings may be readily altered using a variety of video editing programs. The recorded data may be used as evidence
in a crime scene investigation. The result of a court action or inquiry may be greatly impacted by the integrity of the picture and
video. Therefore, before employing surveillance images and videos as evidence in any judicial proceeding, it is essential to verify
their legitimacy. Because they can not effectively use high-resolution encoded information, current deep learning-based methods
for image copy-move fraud detection are unable to discern the borders of tiny modified objects. The recovered frames in this
proposed article improve the accuracy of identification for tiny items by making use of the spatial information found in the feature
maps. A customized CNN-LSTM layer that employs transfer learning to differentiate between authentic and modified frames is
shown in the study. Ensemble Learning is used to evaluate the model. The algorithm's capacity to identify copy move forgeries
is improved by using sophisticated neural network designs, which enable the model to efficiently learn and extract intricate
information from both photos and videos. The model achieves 95.00% accuracy in detecting forged films with static backgrounds,
98.67% accuracy in detecting forged videos with moving backgrounds, and 95.08% accuracy in detecting and localizing the
forged image. A variety of social media photos and videos are used to evaluate the model. The experimental findings show that
the proposed model outperforms other current methods.