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

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