Detection and localization of image and video manipulations
Year: 2023
Type: dissertation
Abstract: The Noiseprint method is an image forensic technique that extracts the noise residual pattern which contents the camera model fingerprint. This fingerprint is represented by a set of traces left on the image by the camera device when capturing it. Based on a review of the literature the Noiseprint network is obtained by means of a Siamese network with two CNN denoisers, which are trained with pairs of image patches coming from the same or different cameras and position on the image (labels +1 and -1 respectively). This thesis focuses on the localization of local forgeries in images created by editing programs based on deep learning, as well as, exploring the robustness of the method when the images are post-processed. Experimental results show that this forensic technique works successfully when the images are uncompressed. Hence, further research is needed to make the Noiseprint method a completely trusted localization system.
Author: Calsina Gamiz, Inés
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7112879209
