Metrics - ISET/isetcam GitHub Wiki

• Supporting Concepts overview
• Metrics tutorials
• Spatial CIELAB (S-CIELAB)

---

What metrics measure

ISETCam's image-quality metrics quantify how well a simulated (or real) imaging system reproduces a scene, computed from the output of the image processor — and, for spatial and color metrics, compared against a calibrated display model, since image quality is ultimately about what a viewer would see. Metrics fall into three groups: spatial, color, and noise.

Spatial metrics

• ISO 12233 slanted edge — the standard method for measuring a system's spatial frequency response (MTF) from an image of a slanted edge. ISO12233 / ieISO12233 compute it; sceneCreate('slanted bar') creates the target. See GUI utilities for the interactive version, and s_metricsMTFSlantedBar in Examples for a scripted one.
• Acutance — a perceptually weighted measure of edge sharpness, combining the MTF with a model of the human contrast sensitivity function (ISOAcutance, cameraAcutance). See s_metricsAcutance in Examples.
• SQRI — Barten's Square Root Integral, a single-number spatial quality metric that integrates contrast sensitivity across spatial frequency (ieSQRI). See t_metricsSQRI in Image processing, color, metrics, and displays.
• SSIM — the structural similarity index; examples/metrics/s_metricsSSIM.m shows ISETCam's usage pattern (SSIM itself is provided by MATLAB's Image Processing Toolbox).

Color metrics

• CIELAB color difference — deltaEab, deltaE94, and deltaE2000 compute the CIE76, CIE94, and CIE2000 color-difference formulas between two CIELAB values, in increasing order of perceptual accuracy and complexity. deltaEuv computes the older CIE 1976 u'v' difference. t_metricsCielab (see Tutorials) plots these; s_metricsMacbethDeltaE (see Examples) is a complete automated pipeline that renders a Macbeth Color Checker and evaluates its rendered color error.
• Spatial CIELAB (S-CIELAB) — a spatial extension of CIELAB that accounts for the eye's reduced sensitivity to high-frequency color errors before computing the color difference. See Spatial CIELAB for the calculation and current status.

Noise metrics

• PSNR (peak signal-to-noise ratio) — iePSNR.
• vSNR (visible SNR) — a perceptually weighted SNR metric (cameraVSNR, xyz2vSNR). See s_metricsVSNR in Examples.
• Pixel- and sensor-level SNR (shot noise, read noise, DSNU, PRNU) are covered on the Sensor noise model page; s_sensorSNR in Examples works through the underlying formula.

Recommended tutorials and examples

• Image processing, color, metrics, and displays — t_metricsCielab, t_metricsColor, and t_metricsSQRI.
• Examples — s_metricsMacbethDeltaE, a fully automated end-to-end color-error evaluation.

Related FISE concepts

The online Foundations of Image Systems Engineering (FISE) develops image-quality metrics, contrast sensitivity, and color difference as conceptual background.

Boundaries

These metrics evaluate the output of the ISETCam pipeline against a reference or a standard target. They are not themselves models of the human visual system beyond the contrast-sensitivity and color-difference formulas built into a given metric; for detailed visual-system modeling, see ISETBio.