Noiseprint
Definition
A CNN-based camera-model fingerprint extractor by Cozzolino and Verdoliva. Applied to deepfakes, it reveals inconsistency between the camera fingerprint in the genuine background and the absent or different fingerprint in the synthesised face region.
- Method
- CNN-based camera-model fingerprint extraction from noise residual
- Detects
- Copy-move splicing and deepfakes via fingerprint inconsistency
- Inventors
- Cozzolino and Verdoliva
Common questions
What does Noiseprint actually do?+
Noiseprint analyzes the noise residual in a digital image to extract a camera-model fingerprint. Every camera has a unique fingerprint from its sensor and processing pipeline. If parts of an image have fingerprints that don't match the dominant pattern, those regions are flagged as potentially manipulated.
How does Noiseprint detect deepfakes?+
In a deepfake, the synthesised face region typically lacks the camera fingerprint present in the genuine background. Noiseprint spots this inconsistency. The fake face either has no fingerprint or a different one than the rest of the image, signaling that it wasn't captured by the same camera.
Is Noiseprint reliable for copy-move detection?+
Noiseprint flags regions where the noise fingerprint breaks the dominant pattern, which is the key signature of copy-move splicing. However, its accuracy depends on the image quality, camera model coverage in the training data, and whether post-processing or compression has degraded the noise pattern.
Related terms
- Affine transform estimation
- The recovery of the geometric transformation (rotation, scaling, shear) relating a copied region to its source, using matched keypoint pairs. Geometrically consistent...
- Chromatic aberration
- Colour fringing at high-contrast edges caused by differential refraction of light wavelengths by a lens. The pattern is lens-specific and inconsistent across...
- CNN residual detector
- A convolutional neural network trained on the high-frequency residual image, the difference between the original and a de-noised version, to classify whether...
- Copy-move forgery
- A manipulation that clones a region from within the same image and pastes it elsewhere, typically to hide an object or repeat...
- Detection generalisation
- The capacity of a trained detector to correctly identify deepfakes produced by generators not seen during training. Low generalisation is the central...
- Frequency-domain artefact
- A periodic or statistical anomaly in the Fourier spectrum of an image or audio signal introduced by the generation pipeline's upsampling, filter,...
- Physiological signal
- A biological process visible in video, such as eye blinking, rPPG (remote photoplethysmography), and head micro-motion from the cardiac cycle, that deepfake...
- rPPG
- Remote photoplethysmography. A technique for measuring heart rate from subtle periodic colour changes in facial skin caused by blood-volume pulses. Authentic video...
- SIFT (Scale-Invariant Feature Transform)
- A keypoint detection and description algorithm that extracts local feature descriptors invariant to scale, rotation, and partial illumination change, enabling matching of...
- Splicing
- A manipulation that inserts content from a completely different source image into the target, producing a composite that may look coherent but...
Explained in these topics
- Copy-Move and Splicing DetectionA CNN-based forensic tool that extracts a camera-model fingerprint from the noise residual of an image; regions where the fingerprint is inconsistent with the...
- Deepfake Detection: Signal, Physiological, and Semantic MethodsA CNN-based camera-model fingerprint extractor by Cozzolino and Verdoliva. Applied to deepfakes, it reveals inconsistency between the camera fingerprint in the...