DeepJurist: A Hybrid Deep Learning Architecture for Detecting Fake Visual Evidence in Judicial Systems

The rise in manipulated visual evidence poses a significant threat to judicial systems, potentially leading to incorrect judgments, wrongful convictions, or acquittals of guilty parties. Such outcomes can erode public trust in legal institutions and undermine the fairness of trials, especially when deepfake videos or doctored images are used to distort the truth. In this context, developing accurate detection methods is critical to maintaining the integrity of judicial decisions. Despite advancements in face forgery detection, most models struggle to effectively capture both spatial and temporal inconsistencies in videos, limiting their application in courtrooms. Addressing this gap, a hybrid deep learning approach is proposed to detect fake visual evidence using the FaceForensics++ dataset, a benchmark for detecting facial manipulations. The proposed model integrates Convolutional Neural Networks (CNN) for feature extraction, Bidirectional Long Short-Term Memory (Bi-LSTM) for sequencing, and MultiHead Attention mechanisms to enhance the capture of temporal dependencies in video frames. This combination enables the model to identify subtle manipulations across frames. The proposed method achieved a high accuracy of 95.40%, with precision, recall, and F1-scores of 93.71%, 97.68%, and 95.65%, respectively, outperforming established transfer learning models such as VGG16, ResNet50, and EfficientNet. This model holds practical potential for legal systems by offering a robust tool for authenticating visual evidence. Its high detection rates could reduce the risk of false judgments and contribute to the integrity of court proceedings, emphasizing the importance of reliable visual verification in critical decision-making scenarios.