Identify the response being measured
An edge captured by a camera measures the response of the recorded imaging chain: optics, sensor sampling and the processing retained in the file. SFR is a spatial frequency response measured by a defined method. To interpret it as optical MTF, the contributions of the capture and analysis chain must be understood. Report the test plane and units alongside the curve.
From the edge profile to a frequency curve
The analysis estimates the edge position and projects pixels onto distance normal to that edge. Samples from successive rows build an edge-spread function (ESF). Differentiating the ESF gives a line-spread function (LSF); windowing and a Fourier transform produce the frequency response, normalised to its low-frequency value. Edge fitting, binning, derivative correction and window choice are method-dependent and should be recorded with the software version.
- A slant supplies different subpixel positions across rows; it improves profile sampling without creating additional optical detail.
- Save the edge region and fitted profile with the result so fitting errors can be examined.
Choose the edge and analysis region together
Use a slight slant supported by the analysis implementation, with a region long enough to populate the profile and wide enough to include stable bright and dark plateaus. The applicable angle, contrast and region size depend on the method and sampling. Transmission through a chrome-on-glass boundary and reflection from a printed chart require different lighting arrangements.
| Item | What to examine |
|---|---|
| Edge direction | Fit the actual angle in the saved image; target artwork angle alone does not describe the camera alignment. |
| Bright and dark levels | Keep both levels within the usable response range and avoid clipping or unstable background illumination. |
| Usable edge | Inspect contamination, chips, reflections and neighbouring features inside the chosen region. |
| Field location | Use identified centre, edge or corner positions and record the tested orientation. |
Keep acquisition and processing comparable
Record focus, aperture, working distance, illumination, exposure and field position. State whether focus is fixed across the field or optimised at each region. Keep sharpening, noise reduction, demosaicing, tone response and resizing settings fixed for comparisons. RAW files may retain corrections or preprocessing; identify the actual pipeline rather than relying on the file extension.
Convert frequency without mixing coordinate planes
For unresampled pixels of pitch p millimetres, image-plane frequency equals cycles per pixel divided by p. With absolute lateral magnification |M|, object-plane frequency equals image-plane frequency multiplied by |M|. A result in cycles per pixel does not require a physical scale, but conversion to lp/mm requires the relevant pixel spacing and plane. Resizing or geometric correction changes sampling and can change the measured response.
Interpret MTF50 and processing effects
MTF50 commonly denotes the frequency where the normalised response reaches 0.5; MTF10 uses 0.1. State the crossing convention when the curve is non-monotonic. Sharpening can produce overshoot and response above 1, so a higher MTF50 may reflect processing as well as optical improvement. Compare the curve, edge profile, noise and original image together. Acceptance limits should correspond to the instrument task and agreed method.