34 Image processing metrics
Work in Progress
Suggestions of all kinds for this book draft are welcome — whether it’s fixing small errors, raising bigger questions, or offering new perspectives. Please share comments through GitHub Issues. To make feedback easier to address, please point to the section you have in mind — by section number or a short snippet of text.
34.1 Image processing metrics overview
| Metric / Standard | Category | Purpose | Notes / Comments |
|---|---|---|---|
| ISO 12233 Slanted-Edge MTF / SFR | Resolution / Sharpness | Measures system MTF using the slanted-edge method (ESF → LSF → MTF) | Industry standard for camera/lens resolution testing. |
| MTF50 / MTF30 / MTF10 | Resolution / Sharpness | Single-number sharpness metrics derived from the MTF curve | Common in Imatest, DxO, DPReview. |
| Acutance (ISO 20462/Imatest) | Resolution / Sharpness | Integrates MTF with a weighting to yield overall sharpness | Weighting often not based on vision; used as a device metric. |
| Siemens Star Resolution | Resolution | Radial resolution and asymmetry detection | Good for checking directional resolution limits. |
| Line Pair Resolution (lp/mm) | Resolution | Classical chart-based resolution measurement | Used in optical labs and film era; simple but widely used. |
| Random Texture MTF (ISO 19567-2) | Texture Preservation | Measures texture rendering in presence of noise reduction/demosaicing | “Dead-leaves” chart; used heavily by DxO. |
| Texture Frequency Loss (TFL) | Texture Preservation | Quantifies suppression of fine texture | Derived from random texture MTF. |
| ISO 15739 Sharpness Metric | Resolution / Texture | Sharpness in real scenes with noise present | Part of ISO imaging performance for DSC. |
| Noise Power Spectrum (NPS) | Noise | Frequency-dependent noise measurement | Standard for sensor evaluation in machine vision & medical imaging. |
| Signal-to-Noise Ratio (SNR, SNR(dB)) | Noise | Basic engineering noise metric | Device-centric, not perceptual. |
| Fixed-Pattern Noise (FPN) | Sensor Noise | Measures spatially fixed noise contributions | Includes DSNU (dark signal) and PRNU (photo-response). |
| Photo-Response Nonuniformity (PRNU) | Sensor Noise | Measures variance of pixel gain | Fundamental for CMOS/CCD characterization. |
| Dark Signal Non-Uniformity (DSNU) | Sensor Noise | Measures dark-current spatial variation | Temperature-dependent; sensor-level metric. |
| Pixel Aperture / Sensor MTF | Sensor Resolution | Analytic or measured MTF of sensor pixel structure | Includes optical blur + charge diffusion. |
| Charge Diffusion Width / Spot Spread | Sensor Blur | Spatial blurring caused within the photodiode | Used in CMOS/CCD modeling and sensor datasheets. |
| Ringing / Overshoot / Undershoot Metrics | Artifact Quantification | Measures artifacts from sharpening, deblurring, demosaicing | Extracted from the edge profile. |
| Aliasing Energy (Spectral Folding) | Artifacts | Measures energy above Nyquist that folds into lower frequencies | Used to evaluate sampling, CFA, & lens interactions. |
| Demosaicing Artifact Metrics | Artifacts | Quantifies false color or zippering | Often computed in Fourier domain or via local statistics. |
| PSNR (Peak Signal-to-Noise Ratio) | Fidelity | Pixelwise distortion measure | Device-centric; widely used in compression and pipeline tests. |
| MSE / RMSE | Fidelity | Raw pixel error | No perceptual component. |
| Linear Distortion Metrics (Gamma / Transfer Function) | Linearity / Fidelity | Measures system response curve | Used in display and camera characterization. |
| ISO 9241-305 Display MTF | Display Resolution | Resolution of display measured with a camera | Standard for spatial performance of displays. |
| Pixel Structure / Subpixel MTF | Display / Sensor Resolution | MTF from subpixel arrangement (OLED, LCD, µLED) | Important for VR/AR and microdisplays. |
| Display Uniformity Maps | Display Uniformity | Spatial variation in luminance or contrast | Not HVS-based; purely physical uniformity metric. |
| IEEE P1858 (CPIQ) Spatial Metrics (non-perceptual subset) | Integrated System Metrics | Various device-centric spatial fidelity and texture metrics | Standardized test suite used by industry. |
| Veiling Glare Index / Flare | Optical Artifact | Measures stray-light performance (e.g., contrast loss) | Classical photographic metric. |
| RMS Granularity (Film) | Texture / Noise | Noise texture metric from film imaging | Included for completeness in imaging history. |
34.2 Component metrics
34.2.1 General measures
PSNR, MSE, RMSE
34.2.2 Sensor measures
PRNU
DSNU
Charge diffusion
34.2.3 Display
Mura
Gamma related
Image quantization
34.5 Distortion metrics
Are there any?
34.6 Texture metrics
Deadleaves is something …
Graphics texture synthesis algorithms, such as Eero’s and the competition.
34.7 High dynamic range metrics
Veiling glare
Dynamic range
34.8 Computer vision metrics for color
These are just vector inner product in whatever RGB space the user happens to be using. No reference to human color vision.
A lot of these might have been used in those directional cosine errors in Zheng’s dissertation. But it is the cosine metric for similarity of large vectors that needs to be included somewhere.