Neural-PRNU-Extractor
Project Overview
Neural-PRNU-Extractor is a modified PyTorch implementation of the FFDNet image denoising technique. Originally developed as part of the Signal, Image, and Video course for the master’s degree program in Artificial Intelligence Systems and Computer Science at the University of Trento, this project extends the capabilities of FFDNet.
Key Features
-
PRNU Extraction: Unlike the original FFDNet, this modified version focuses on the extraction of cameras’ PRNU (Photo Response Non-Uniformity).
-
Training Options: Neural-PRNU-Extractor provides the flexibility to train the network using the Wiener filter as a noise extraction strategy, in addition to the method outlined in the FFDNet paper.
Project Objective
The primary goal of Neural-PRNU-Extractor is noise reduction in images. It aims to predict and remove noise from images captured by digital and film cameras, enhancing their suitability for various applications in computer vision. The project can effectively predict noise levels (\(\sigma \in [0, 75]\)) in images.Additionally, Neural-PRNU-Extractor can compute and evaluate PRNU patterns based on a dataset of flat images, allowing for the identification of the specific device that generated an image. PRNU, or Photo Response Non-Uniformity, is a crucial factor in digital image sensors, commonly found in cameras and optical instruments, making it a valuable tool for device identification and verification.
The PRNU extraction pipeline can be seen here: Download PDF
Contributors
- Samuele Bortolotti
- Simone Alghisi
- Massimo Rizzoli