VIPPrint: Validating Synthetic Image Detection and Source Linking Methods on a Large Scale Dataset of Printed...

Area of Science:

• Digital Forensics
• Image Analysis
• Computer Vision

Background:

• Printed documents are frequently linked to criminal activities, including terrorism and child exploitation.
• Printing and scanning processes can obscure image manipulation and the synthetic origin of images, posing challenges for forensic analysis.
• Existing research is hampered by a scarcity of large-scale reference datasets for developing and evaluating forensic algorithms.

Purpose of the Study:

• To introduce a novel, large-scale dataset of synthetic and natural printed face images.
• To facilitate algorithm development and benchmarking in the forensic analysis of printed and scanned images.
• To highlight the challenges in analyzing such images and stimulate further research.

Main Methods:

• A comprehensive dataset of printed synthetic and natural face images was compiled.
• Printer attribution methods were experimentally compared on the new dataset.
• State-of-the-art methods for distinguishing natural from synthetic images were evaluated on printed and scanned images.

Main Results:

• Printer attribution methods were tested, revealing complexities in forensic analysis of printed images.
• Established methods for identifying synthetic images proved ineffective on printed and scanned faces.
• The new dataset provides a valuable resource for advancing forensic image analysis.

Conclusions:

• The developed dataset and preliminary experiments underscore the difficulties in digital forensics for printed and scanned images.
• Further research is needed to develop robust methods for analyzing images that have undergone printing and scanning.
• The dataset is expected to accelerate progress in forensic image analysis and manipulation detection.