Imaging Systems and Optical Transfer Functions

The transfer-function picture is the daily tool of anyone who specifies an imaging instrument. Camera and lens makers publish MTF charts — contrast at 10 and 30 lp/mm versus field position — and a system's MTF is the product of the lens MTF, the sensor's pixel-aperture sinc, and any anti-alias filter. Semiconductor lithography lives right at the Abbe limit, printing half-pitch $k_1\lambda/\mathrm{NA}$: water immersion raised the effective NA past 1.35 and phase-shift masks engineer the object and pupil spectra to shrink $k_1$, before extreme-ultraviolet sources cut $\lambda$ to 13.5 nm. In astronomy, atmospheric turbulence stamps a random phase across the pupil and collapses the long-exposure MTF to that of a coherence patch of size $r_0$; adaptive optics cancels that pupil phase hundreds of times a second, restoring the diffraction-limited MTF of the full aperture — a real-time application of the Strehl and pupil-phase formalism. The most striking modern twist is that the Abbe limit binds only linear far-field imaging. Structured-illumination microscopy beats the sample against a known grating to fold high object frequencies down into the passband, doubling the effective cutoff; STED and single-molecule methods (PALM, STORM) exploit nonlinear fluorescence and localization to place emitters to a few nanometres. None violate the pupil autocorrelation — they add information the linear OTF cannot carry, so the transfer function still states exactly what one linear snapshot can and cannot see.

Treat the lens aperture as a bandpass filter for spatial frequencies. Fine object detail radiates steeply tilted plane waves; only tilts that still fit through the aperture are collected and re-focused. Detail finer than the cutoff radiates waves that miss the lens entirely, so no refocusing recovers it — only a wider aperture (higher NA) or a shorter wavelength collects those waves. This is why resolution is a property of the aperture, not the magnification: blowing the image up larger only magnifies the blur. The MTF is best read as a contrast-survival curve. Photograph a ladder of black-and-white bar targets of increasing spatial frequency: coarse bars come through crisp with contrast near unity, fine bars come through faded, and beyond the cutoff they wash out to uniform grey. The whole curve — not a single resolution number — characterizes the instrument, which is why lens makers publish MTF rather than a lines-per-millimetre figure.

The curve has exactly this shape because of geometry. The pupil-autocorrelation theorem says the OTF at frequency $f$ equals the fractional overlap area of the pupil with a copy of itself shifted by $s = \lambda z_i f$. Push to higher frequency and the disks slide apart, the overlap shrinks, and when the shift reaches a full pupil diameter they separate completely — that shift is the cutoff. Aberrations enter as wrinkles in the pupil phase: autocorrelating a wrinkled pupil with its shifted self lets the phases partly cancel, so the overlap integral drops, the MTF sags, and it can even go negative (contrast reversal). But the boundary where the disks separate never moves. Aberrations spend contrast; they cannot buy resolution.

Start from the operational statement of imaging. For an incoherent object — one whose points radiate with no fixed phase relationship, covering most self-luminous and diffusely-lit scenes — intensities add, and the shift-invariant instrument reproduces every object point as the same blur $h$.

where: $g_i$ — image-plane intensity; $g_o$ — geometric (magnification-scaled) image intensity; $h$ — intensity point-spread function, the image of a point; $*$ — 2-D convolution. The image is the object smeared by a fixed blur kernel — the sharper the kernel, the sharper the image. The convolution theorem converts the smear into a product. Fourier-transforming both sides replaces $h$ by the OTF and the object by its spatial-frequency spectrum; the modulus and phase of the OTF carry distinct information and get their own names.

where: $H$ — optical transfer function, normalized so $H(0,0)$ is the total collected energy; $f_x,f_y$ — spatial frequencies (cycles per unit length in the image plane); $\mathrm{MTF}$ — fraction of contrast transferred; $\mathrm{PTF}$ — phase, recording lateral shifts and sign reversals. The MTF answers "how much contrast survives," the PTF "does the pattern move or invert?" Both the PSF and the coherent transfer come from the aperture. Fraunhofer diffraction makes the amplitude PSF the Fourier transform of the pupil, so the intensity PSF is its squared modulus; for coherent light the system is linear in field and the amplitude transfer function is the pupil itself, rescaled so pupil position maps to spatial frequency.

where: $P(u,v)$ — pupil function, 1 inside the aperture and 0 outside, carrying phase $e^{ikW}$ for a wavefront aberration $W$; $\mathrm{ATF}$ — coherent (amplitude) transfer function, equal to the scaled pupil; $z_i$ — exit-pupil-to-image distance; $k=2\pi/\lambda$. The aperture is literally the coherent filter, hard-edged at the rim. The central theorem follows: because the incoherent PSF is a squared modulus, its transform is an autocorrelation, and unwinding the transforms shows the OTF is the pupil autocorrelated with itself — the equation the two figures drew.

where: $P^{*}$ — conjugate pupil, carrying the opposite aberration phase $-W$; $\lambda z_i f$ — pupil-plane shift for spatial frequency $f$. The transferred contrast at frequency $f$ is the fractional overlap of the pupil with a copy shifted by $\lambda z_i f$ — overlap area over total area. Because the pupil has finite radius $\rho_0$, the coherent transfer stops when $\lambda z_i f$ reaches $\rho_0$, while the autocorrelation spans twice the pupil radius and stops only at $2\rho_0$. That factor of two is the single most important number-fact of the subject.

where: $\mathrm{NA}=n\sin\theta_{\max}$ — image-space numerical aperture; $F_\# = f/D\approx 1/(2\,\mathrm{NA})$ in air — f-number; $\lambda$ — wavelength in the image-space medium. The incoherent system reaches twice the coherent cutoff, but with graded not flat contrast — "twice the cutoff" is not "twice as good." For the circular aperture the coherent PSF is the Airy pattern — the squared jinc function — whose first dark ring sets the classical resolution scale, and placing one Airy peak on a neighbour's first zero defines the Rayleigh criterion.

where: $J_1$ — first-order Bessel function; $r$ — radial image-plane coordinate; $r_{\mathrm{Airy}}$ — first-dark-ring (Rayleigh) radius. About 83.8% of the point's energy lands inside this ring (the Airy disk), and two points closer than $r_{\mathrm{Airy}}$ blur into one. The competing conventions differ only by a constant: Rayleigh $0.61\,\lambda/\mathrm{NA}$, Abbe $\lambda/(2\,\mathrm{NA})=1/f_c^{\mathrm{inc}}$, and Sparrow $\approx0.47\,\lambda/\mathrm{NA}$.

1 · Resolution of a microscope objective (the concrete anchor)

A dry objective has $\mathrm{NA}=0.95$ at $\lambda=550$ nm. Find its coherent and incoherent cutoffs, its Rayleigh and Abbe limits, and the contrast it transfers at the coherent cutoff.

Step 1. Coherent cutoff $f_c^{\mathrm{coh}}=\mathrm{NA}/\lambda=0.95/0.550\ \mu\mathrm{m}=1.73$ cycles/$\mu$m ($\approx1730$ lp/mm). Incoherent cutoff is twice this: $f_c^{\mathrm{inc}}=2\,\mathrm{NA}/\lambda=3.45$ cycles/$\mu$m, about 3450 line-pairs/mm — the finest intensity grating it passes at all.

Step 2. Rayleigh limit $d_{\min}^{\mathrm{Rayleigh}}=0.61\lambda/\mathrm{NA}=0.61\times550/0.95=\mathbf{353\ nm}$. Abbe limit $d_{\min}^{\mathrm{Abbe}}=\lambda/(2\,\mathrm{NA})=550/1.90=\mathbf{289\ nm}$, and as a check $1/f_c^{\mathrm{inc}}=1/3.45\ \mu\mathrm{m}=289$ nm — the Abbe limit is exactly the reciprocal of the incoherent cutoff.

Step 3. The coherent cutoff sits at $\tilde f=f_c^{\mathrm{coh}}/f_c^{\mathrm{inc}}=0.5$, so the incoherent MTF there is $\tfrac{2}{\pi}[\arccos 0.5-0.5\sqrt{0.75}]=\mathbf{0.391}$ — a grating of that fineness images with only 39% of its contrast. To reach 200 nm one goes to oil immersion, $\mathrm{NA}\approx1.4$, dropping the Abbe limit to $550/2.8=196$ nm. Cutoff 3450 lp/mm, Rayleigh 353 nm, Abbe 289 nm, MTF = 0.391 at the coherent cutoff.

2 · MTF of a camera lens at two test frequencies

A photographic lens works at $f/8$ with $\lambda=550$ nm. Find the Airy-disk radius, the incoherent cutoff in lp/mm, and the MTF at 30 and 60 lp/mm.

Step 1. Airy radius $r_{\mathrm{Airy}}=1.22\,\lambda F_\#=1.22\times0.550\ \mu\mathrm{m}\times8=\mathbf{5.37\ \mu m}$. That exceeds a $\sim4\ \mu$m pixel, so at $f/8$ the lens is diffraction-limited, not pixel-limited — stopping down further only softens it.

Step 2. Cutoff $f_c^{\mathrm{inc}}=1/(\lambda F_\#)=1/(0.550\times8)\ \mu\mathrm{m}^{-1}=\mathbf{227\ lp/mm}$. The test frequencies are $\tilde f=30/227=0.132$ and $\tilde f=60/227=0.264$.

Step 3. At 30 lp/mm: $\tfrac{2}{\pi}[\arccos 0.132-0.132\sqrt{1-0.132^2}]=\tfrac{2}{\pi}[1.4384-0.1308]=\mathbf{0.832}$. At 60 lp/mm: $\tfrac{2}{\pi}[\arccos 0.264-0.264\sqrt{1-0.264^2}]=\tfrac{2}{\pi}[1.3035-0.2546]=\mathbf{0.668}$. These are the ceilings a real $f/8$ lens is graded against; aberrations only pull them lower. 83% contrast at 30 lp/mm, 67% at 60 lp/mm — the diffraction-limited best for f/8.

3 · Is a slightly defocused lens still diffraction-limited?

A lens carries residual defocus with peak-to-valley wavefront error $\lambda/4$. Estimate its RMS error and Strehl ratio, and decide.

Step 1. For defocus across a circular pupil the RMS is the peak-to-valley over $2\sqrt3$: $\sigma_W=(\lambda/4)/(2\sqrt3)=\lambda/13.9\approx\lambda/14$ — which is why the "$\lambda/4$ PV" and "$\lambda/14$ RMS" rules are the same statement. The Strehl estimate is then $S\approx e^{-(2\pi\sigma_W/\lambda)^2}=e^{-(2\pi/14)^2}=e^{-0.2014}=\mathbf{0.82}$.

Step 2. The Marechal threshold is $S>0.8$; since $0.82>0.80$ the lens just qualifies as diffraction-limited — its MTF hugs the ideal curve, and the defocus costs a few percent of peak intensity without moving the cutoff. Any more defocus and the Strehl falls through 0.8 and the MTF visibly sags. RMS $\approx\lambda/14$, Strehl $\approx0.82$ — yes, but barely.

The transfer-function view is the modern language, but its physical heart — that resolution is set by how much of the diffracted spectrum an aperture can gather — was seen with startling clarity a century before Fourier optics had a name, by a man staring down a microscope for a lens company.

Texts. Goodman, Introduction to Fourier Optics, 3rd ed., Ch. 6 (coherent and incoherent imaging, the OTF) · Hecht, Optics, 5th ed., Ch. 13 (Fourier optics, OTF and MTF) and Ch. 10 (Fraunhofer diffraction, the Airy pattern) · Born and Wolf, Principles of Optics, 7th ed., Ch. 8-9 (diffraction and aberrations) · Saleh and Teich, Fundamentals of Photonics, 3rd ed., Ch. 4 · Demtroeder, Laser Spectroscopy, Vol. 1 (spectrographs and imaging resolution). Key numerical anchors for this article: the circular-aperture MTF satisfies $\mathrm{MTF}(0)=1$, $\mathrm{MTF}(1)=0$, $\mathrm{MTF}(0.5)=0.391$; the Rayleigh radius follows from the first zero of $J_1$ at 3.8317, giving $3.8317/\pi=1.220$ and $r=0.61\,\lambda/\mathrm{NA}$; the Airy disk encircles 0.838 of the total energy; the cutoffs are $f_c^{\mathrm{inc}}=2\,\mathrm{NA}/\lambda=1/(\lambda F_\#)$ and $f_c^{\mathrm{coh}}=\mathrm{NA}/\lambda$; the Strehl estimate gives $S\approx e^{-(2\pi\sigma_W/\lambda)^2}=0.82$ at $\sigma_W=\lambda/14$; and the worked MTF values are 0.832 at $\tilde f=0.132$ and 0.668 at $\tilde f=0.264$.

Prerequisites. Numerical Aperture and Acceptance Angle · Fourier Transforms and Convolution · Fraunhofer Diffraction. Leads to. Integrated Photonic Waveguides and Circuits · Microring Resonators and Optical Cavities on Chip · Total Internal Reflection and Optical Guiding.

Ask the tutor: Tell me your lens or objective's NA (or f-number) and wavelength, and I will walk you through its cutoff, Airy disk, and full MTF curve step by step.