DeepTamperNet Identification: A Hybrid Preprocessing and Feature Extraction Fusion Pipeline for Robust Deepfake Detection

The proliferation of crimes especially with the misuse of digital content embedded with advanced technology to manipulate individuals has largely accelerated in neoteric times. Simulated data has established trust amongst individuals, thereby making them vulnerable to several frauds that manipulate their identity. This manipulation has led to severe loss interms of finances and in also risking several lives. Deepfakes are one such cybercrimes that has increased gargantuanly, thereby necessitating the impeccable technological mechanisms to agnize fakes, tampering and manipulation. Although existing research identifies deepfakes, the approach of static identification makes it impossible to identify the tampered data dynamically while in the process of manipulation. This determination overcomes the gap evinced in the existing studies by incorporating tampering levels in a dynamic data. The tampered heatmap levels are shown dynamically to comprehend the level of tampering, thereby enabling users to distinguish between manipulated and genuine data. The paper entails a detection pipeline through the processes of pre-processing and feature unsheathing using deep heuristic techniques. The Multi-Task Cascaded Convolutional Neural Network (MTCNN) is used for face detection embedded with different pre-processing techniques. The study analyses the tamper score which accelerates the evaluative quantification of manipulated data. A hybrid deep learning model such as the ResNet18, MobileNetV2 and EfficientNetB0 fused with hybrid heuristic edge algorithm is combined to form the "DeepTamperNet" that aids in stratifying the dynamic data into fake and genuine. This research aids in curbing cybercrimes through efficient tampering cognizance deliverables that can be used as video tampering detectors