Artificial Image Tampering Distorts Spatial Distribution of Texture Landmarks and Quality Characteristics

Authors: Tahir Hassan, Aras Asaad, Dashti Ali, Sabah Jassim

Published: 2022-08-04 15:13:00+00:00

Comment: 6 pages, 7 figures, 3 tables

AI Summary

This paper investigates the detection of artificial image tampering, specifically face morphing attacks, by analyzing distortions in face photographs. It extends previous work using Persistent Homology (PH) of texture landmarks and introduces a novel Multi-Characteristics Image Quality (MCIQ) feature vector. The study demonstrates that both the spatial distribution of texture landmarks and image quality characteristics are significantly altered by tampering, providing effective features for explainable tamper detection.

Abstract

Advances in AI based computer vision has led to a significant growth in synthetic image generation and artificial image tampering with serious implications for unethical exploitations that undermine person identification and could make render AI predictions less explainable.Morphing, Deepfake and other artificial generation of face photographs undermine the reliability of face biometrics authentication using different electronic ID documents.Morphed face photographs on e-passports can fool automated border control systems and human guards.This paper extends our previous work on using the persistent homology (PH) of texture landmarks to detect morphing attacks.We demonstrate that artificial image tampering distorts the spatial distribution of texture landmarks (i.e. their PH) as well as that of a set of image quality characteristics.We shall demonstrate that the tamper caused distortion of these two slim feature vectors provide significant potentials for building explainable (Handcrafted) tamper detectors with low error rates and suitable for implementation on constrained devices.

Key findings

The proposed MCIQ and PH-based feature vectors effectively detect morphing attacks with high accuracy, often achieving low False Rejection Rates (FRR) and False Acceptance Rates (FAR) below 4%. While performance varied between different databases and PH feature types, the methods demonstrated good generalization capabilities in cross-database validation tests. The PH features, particularly with barcode statistics and improved preprocessing, outperformed previous PH-based approaches.

Approach

The paper proposes two main approaches: one based on Persistent Homology (PH) of texture landmarks and another using a novel Multi-Characteristics Image Quality (MCIQ) feature vector. For PH, Uniform Local Binary Patterns (ULBP) are used to extract texture landmarks, from which persistence barcodes are generated and then featurized into Betti curves and statistical measures. The MCIQ approach computes a 50-dimensional spatio-statistical feature vector by analyzing various image quality metrics across image blocks. A Cubic SVM classifier is then used for detection.

Datasets

AMSL DB, Utrecht DB

Model(s)

Cubic SVM (for classification), Persistent Homology (PH) and Multi-Characteristics Image Quality (MCIQ) for feature extraction.

Author countries

UK, Canada

← Previous