How nikhil16072003/securing-video-content-against-deepfakes-with-dct-dwt-gan-based-watermarking works
Main Components
• Data Sources:
• Dataset Directory (contains raw video data or supporting media).
• Input Videos: input1.mp4 and possibly other sample videos.
• Deepfake Video: deepfaked.mp4 is likely used as a test case for robustness.
• Preprocessing & Transformation Modules:
• Modules or functions that apply DCT (Discrete Cosine Transform) and DWT (Discrete Wavelet Transform) to the raw video frames.
• Watermark Embedding via GAN:
• GAN Architecture: Represent the generator and discriminator components that implement watermark embedding and possibly validation against deepfakes.
• Training & Testing Pipelines:
• Train.ipynb: For training the GAN model/method.
• Test.ipynb: For evaluating the performance of the watermarking, including output generation (output_watermarked1.mp4).
• Evaluation & Visualization:
• Graph.xlsx: For storing or displaying performance metrics/charts.
• Any additional analysis (possibly integrated within the testing notebook).
Relationships and Interactions
• Data Flow: Begin with raw video input coming from the Dataset directory or given input video files.
• Preprocessing: Input videos feed into preprocessing modules that apply the DCT/DWT and possibly other necessary video processing steps.
• GAN Module: Processed data then goes into the GAN architecture where the generator embeds the watermark while the discriminator provides feedback.
• Output Generation: The output from the GAN (with an embedded watermark) is then compiled back into an output video file (output_watermarked1.mp4).
• Validation: Compare results against deepfaked.mp4 to ensure the watermark's integrity under tampering, with results documented in graph.xlsx.
Architectural Patterns & Design Principles
• Pipeline Architecture: Emphasize distinct stages—data ingestion, preprocessing, model training, and evaluation—all connected in a sequential data flow.
• Modularity: Clearly separate the transformation stages (DCT & DWT), the GAN-based watermarking, and the evaluation phases.
• Experimentation & Iteration: The usage of notebooks (train.ipynb and test.ipynb) suggests an iterative development and experimentation approach, which should be indicated in your annotations.
Relevant Technologies and Frameworks
• Signal processing libraries for DCT/DWT handling.
• Data visualization tools for handling Excel charts if needed.
• Use clear labels for each major component (e.g., "Input Video", "Preprocessing: DCT/DWT", "GAN Generator", "GAN Discriminator", "Output Watermarked Video", "Evaluation Module").