MECH5720 Chap.6 Coded Apertures and Transfer Functions
Coded Apertures and Transfer Functions
A coded aperture reshapes the point-spread function so that computation can recover scene information. The Fourier transform of the measured PSF is the optical transfer function; its magnitude is the modulation transfer function. Stable design seeks useful throughput and broad frequency support rather than a bright response with deep spectral nulls.
Calibration must include the assembled optics and should test variation across field position and depth before one kernel is used globally. A point source provides the spatial impulse used to measure the complete system. The recorded pattern includes the aperture, propagation, lens elements if present, detector integration, alignment and fabrication errors.
A binary mask file describes only the intended opening, so substituting it for a physical PSF ignores effects that reconstruction must undo. Dark subtraction, unsaturated exposure and a point sufficiently small for the desired resolution are basic calibration controls. Fourier transformation yields a complex optical transfer function.
Its magnitude reports contrast support by frequency; its phase reports displacement and asymmetric response. The zero-frequency coefficient reflects total throughput, whereas near-zero coefficients elsewhere identify patterns that were scarcely recorded. A bright mask with deep nulls may therefore be less useful for inversion than a dimmer mask with broader, more even support.
Compare absolute spectra for photon efficiency and normalised spectra for conditioning. Deconvolution divides or approximately divides by the transfer response. Plain inverse filtering amplifies noise and mismatch wherever the magnitude is small. Wiener-style regularisation places a positive term in the denominator, reducing unstable recovery at the cost of bias and lost detail.
The regularisation choice should be swept on a controlled target, with residual, target error and visible artifacts considered together. A single sharpness statistic can reward ringing and cannot establish accuracy. Shift invariance and depth invariance are hypotheses to test. Measure PSFs at centre and corners with range fixed, then at several ranges with field position fixed.
Systematic corner change calls for a spatially varying model, restricted field or optical redesign. Range change calls for depth-specific kernels or a volumetric model. One global parameter cannot repair the wrong operator. Resolution targets connect the transfer curve to image structure. Contrast around a Siemens star or across bar frequencies should be computed with declared normalisation and exposure.
Directional codes need angular analysis because radial averaging can conceal a weak orientation. Report calibration geometry, retained spectral range, reconstruction settings and field or depth limits so the final image remains traceable to information the camera actually measured.
What this chapter covers
- 01
Coded-mask design
- 02
Measured point-spread functions
- 03
Optical transfer function
- 04
Modulation transfer function
- 05
Wiener deconvolution
- 06
Field and depth calibration
Worked example · free
Regularise a weak transfer coefficient
- +1Compute H squared = 0.01 and the denominator 0.01 + 0.02 = 0.03.
- +1Compute the numerator HY = 0.006.
- +1The regularised estimate is 0.006/0.03 = 0.20.
- +1Plain inversion gives Y/H = 0.60.
- +1Explain that regularisation suppresses the unstable weakly supported coefficient.
Key terms
- Coded aperture
- A patterned opening used to encode scene information in the sensor measurement.
- Point-spread function
- The spatial response of the complete imaging system to a point source.
- Optical transfer function
- The complex Fourier transform of a point-spread function.
- Modulation transfer function
- The magnitude of the optical transfer function.
- Wiener filter
- A regularised inverse that limits recovery where transfer support is weak.
Coded Apertures and Transfer Functions FAQ
Why should the measured PSF replace the mask design?
The design omits diffraction, aberration, fabrication error, placement, detector response and alignment. A measured PSF captures the assembled system under the calibration condition used by reconstruction.
Does equal open area mean equal reconstruction quality?
No. Equal open area approximately controls throughput, but the masks can distribute frequency support very differently. Compare normalised transfer magnitude for conditioning and absolute response for photon efficiency.
What causes ringing after deconvolution?
Ringing can arise from weak transfer frequencies, sharp regularisation transitions, boundary assumptions or a mismatched kernel. Test calibration at the scene depth and field position before assuming that parameter tuning is the only remedy.
Exam move
Trace PSF to OTF to MTF and state what each representation contributes. Compare masks on both throughput and normalised frequency support. Work small Wiener calculations, then diagnose whether an artifact is caused by weak support, boundary handling, calibration mismatch or regularisation strength. Practise reading a two-dimensional transfer map before reducing it to a radial curve, especially for directional masks.
For every parameter sweep, track residual, target accuracy and visible artifacts together. Design a calibration grid across field position and depth, then decide whether one kernel remains defensible. End by explaining why matched open area controls throughput but cannot establish equal conditioning or equal reconstruction stability. Include phase response when a spatial shift or asymmetric artifact is part of the diagnosis.