Answers are grounded in this paper and cite their sources.
Fuzzy logic model achieves competitive detection accuracy in image manipulation across various datasets, suggesting improved transparency over deep learning models.
This work introduces FLIDS (Fuzzy Logic-based Image Distortion Scoring), an interpretable and efficient system for image tampering detection based on hand-crafted features and fuzzy logic. FLIDS combines JPEG artifact analysis, edge consistency, co-occurrence entropy, and CFA disparities into a fuzzy rule-based system for assigning a tampering confidence score. In contrast to black-box deep learning systems, FLIDS prioritizes transparency and generalizability. Tests on CIFAR-10, MNIST, ImageNet Subset, and Deepfake datasets indicate FLIDS attains competitive accuracy compared to ResNet-18, Autoencoder, and hand-designed JPEG detectors in the majority of instances. FLIDS achieves 93.5% and 91.8% accuracy on CIFAR-10 and ImageNet Subset, respectively, as well as a balanced 90.2% on deepfake datasets. These findings point to FLIDS as a promising, interpretable solution to intricate deep learning systems in image forgery detection.
No takes yet. Share an insight, caveat, or question.
Shuriya et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: