Demosaicing
Definition
The algorithm used to interpolate full-colour pixel values from the single-channel raw sensor data. Common methods include bilinear interpolation, gradient-based algorithms (AHD, PPG), and proprietary firmware implementations. Each method leaves a distinctive periodic correlation signature in the image.
- What it does
- Reconstructs all three color channels (red, green, blue) at every pixel from the single-channel Bayer mosaic.
- Forensic clue
- Leaves a spatial correlation pattern that varies by algorithm, useful for detecting local tampering and verifying camera authenticity.
- Common algorithms
- Bilinear, AHD, VNG, and deep-learning methods each produce distinct interpolation fingerprints.
Common questions
What is demosaicing and why does it matter in forensics?+
Demosaicing reconstructs the full color image from a Bayer mosaic, which is how digital camera sensors capture light. The reconstruction algorithm leaves a detectable spatial fingerprint, allowing forensic analysts to spot tampering by finding inconsistencies in the expected pattern.
Do different cameras produce different demosaicing fingerprints?+
Yes. The specific algorithm a camera uses to demosaice (bilinear, AHD, VNG, or deep-learning methods) creates a unique pattern. This means comparing an image's fingerprint against known camera signatures can reveal inconsistencies that point to splicing or other editing.
Can demosaicing artifacts help detect forgery?+
Definitely. When an image is edited or tampered with, the patched region typically does not match the camera's original demosaicing fingerprint. Analysts look for breaks in these correlation patterns to isolate suspicious areas.
Related terms
- Bayer CFA
- A colour filter array arranged in a repeating 2×2 mosaic of red, green (×2), and blue filters over the sensor. Each photosite...
- Brown-Conrady Model
- The standard parametric model for camera lens distortion, using radial coefficients k1, k2, k3 and tangential coefficients p1, p2. Camera calibration tools...
- Camera Noise Model
- A mathematical description of a sensor's noise as a function of signal level, combining signal-dependent shot noise and signal-independent readout noise. The...
- CFA (Colour Filter Array)
- The mosaic of colour filters placed over a sensor so that each photosite captures one colour channel. The Bayer pattern (RGGB arrangement)...
- CFA Phase
- The spatial offset of the CFA mosaic pattern in an image. Pasting a region from a different image or after a non-CFA-aligned...
- Colour Filter Array (CFA)
- A mosaic of spectrally selective filters placed over a camera sensor so that each photosite measures only one colour channel. The standard...
- Colour Space
- A defined mapping between numerical RGB values and real-world colours. sRGB is the web default with a smaller gamut; Adobe RGB covers...
- Device-Class Identification
- Establishing that an image was captured by one of a set of cameras sharing the same sensor architecture and firmware, rather than...
- Fixed-Pattern Noise (FPN)
- Any repeatable, spatially correlated noise in an image sensor, as opposed to random shot noise. PRNU and DCNU are both forms of...
- Frequency-Domain CFA Spectrum
- When a demosaiced image is transformed into the frequency domain (using a 2D DFT), the CFA interpolation introduces periodic peaks at spatial...
- JPEG Compression
- A lossy codec that divides an image into 8×8 pixel blocks, applies the discrete cosine transform, quantises the coefficients, and entropy-codes the...
- Noise Residual
- The signal remaining after a denoising filter removes the scene component from an image. For PRNU estimation, the residual approximates the sum...
Explained in these topics
- Digital Image Fundamentals: Pixels, Sensors, and FormatsThe interpolation step that reconstructs all three colour channels at every pixel from the single-channel Bayer mosaic. The algorithm used is camera-specific a...
- Lens Distortion, CFA Pattern and Demosaicing FingerprintsThe algorithm used to interpolate full-colour pixel values from the single-channel raw sensor data. Common methods include bilinear interpolation, gradient-bas...
- Noise and CFA Inconsistency DetectionThe interpolation process that reconstructs missing colour channels at each pixel from adjacent CFA samples. Different algorithms (bilinear, AHD, VNG, deep-lea...