MECH5720 Sensors and Signals

How MECH5720 is assessed

The four numerical rows sum to 100%. The final supervised written examination has a separate hurdle. The major-project pitch is a required hurdle within the project but carries no additional weight beyond the 20% project total.

What MECH5720 covers

The unit moves from camera formation and sampling through multiplexed acquisition and frequency-domain analysis, then applies those tools to coded apertures, lensless reconstruction, light fields, depth and an evidence-led imaging project.

The focus then shifts from conventional image capture to computational imaging, where the measurement is deliberately encoded and a reconstruction algorithm recovers the desired scene information.

How the analytical thread develops

Multiplexing introduces linear measurement models and the distinction between having enough equations and having a well-conditioned inverse.

Two-dimensional Fourier analysis then supplies the language for convolution, filtering, sampling and transfer functions. Coded-aperture work connects a measured point-spread function to its optical and modulation transfer functions, exposing why throughput alone does not guarantee stable recovery.

Lensless imaging generalises the system to a forward operator and uses regularisation and ADMM to negotiate data agreement with an image prior. Light-field imaging retains ray direction, enabling post-capture refocusing and depth inference from disparity.

What good work looks like

The unit rewards evidence-led engineering.

A credible result states the geometry, calibrates the assembled system, uses a meaningful baseline, keeps parameters and units visible, and tests a conclusion with more than an attractive output image. Residuals, repeated captures, resolution targets, focal stacks and known-depth scenes provide complementary checks.

When an assumption fails across field position, depth, view or wavelength, the conclusion should narrow to the validated range.

How assessment shapes preparation

Four group laboratory submissions contribute 40 percent, the major project contributes 20 percent, and the supervised two-hour open-book written examination contributes 40 percent. The examination carries a written-exam hurdle.

The project comprises a report and talk, with a required pitch inside the project process.

Effective revision therefore combines calculations with diagnosis: write the forward model, identify the limiting assumption, compute carefully, predict the artifact and propose a control that could disprove the explanation.

MECH5720 cheatsheet for the final exam

AskSia publishes a free MECH5720 cheatsheet for the final exam, a cheat sheet that follows the official permitted-materials rule for this assessment and condenses the topics in this guide.

Open the MECH5720 cheatsheet.

Worked example · free

Locate the sensor for a thin-lens image

Key terms

Forward model

A mathematical mapping from the scene through the imaging system to the expected sensor measurement.

Point-spread function

The measured image of a point source, describing how a system distributes point energy in space.

Modulation transfer function

The magnitude of the optical transfer function, showing retained contrast as spatial frequency changes.

Spatial frequency

The rate at which image intensity varies across distance or samples, commonly measured in cycles per pixel.

Nyquist limit

One half cycle per sample, above which sampled frequencies fold into aliases without prior filtering.

Condition number

A measure of how strongly errors or noise can be amplified when a linear system is inverted.

Coded aperture

A patterned opening that deliberately shapes a point-spread function for computational reconstruction.

Wiener filtering

Regularised frequency-domain recovery that limits division by weak transfer coefficients.

Adjoint operator

The transpose-conjugate mapping that backprojects measurement-space information into scene coordinates.

Regularisation

A constraint or penalty that stabilises an inverse problem by favouring plausible solutions.

Light field

A ray-based representation that retains both spatial position and propagation direction.

Epipolar-plane image

A two-dimensional slice of a light field in which scene points trace depth-dependent lines.

Disparity

The image-coordinate displacement of a feature between views, used with calibrated geometry to infer depth.

Data residual

The difference between a recorded measurement and the measurement predicted by an estimated scene.

MECH5720 FAQ

Questions students ask in each MECH5720 chapter

What makes imaging computational?

The optics and sensor produce an encoded measurement that is interpreted with an explicit model and algorithm. Computation is not merely post-processing; it is part of how the system captures and separates information for a defined task. Chapter 01 · Computational Imaging Systems

What does the thin-lens equation omit?

It omits lens thickness, principal-plane detail, aberrations, diffraction, field dependence and assembly tolerances. It is valuable for first-order geometry, while physical calibration locates best focus and characterises departures in the real camera. Chapter 02 · Image Formation and Sensor Physics

Does a smaller aperture always sharpen the image?

Stopping down can reduce aberration and increase depth of field, but it also reduces light and eventually increases diffraction blur. The best setting depends on scene depth, exposure constraints, optical quality and the spatial frequencies needed by the task. Chapter 03 · Classical Imaging Tradeoffs

Is an invertible code automatically a good code?

No. An invertible matrix can still have nearly dependent rows or columns, causing strong noise amplification. Inspect singular values or condition number and test recovery under expected noise and calibration errors. Chapter 04 · Multiplexing and Computational Relighting

Why is phase important for images?

Phase carries the relative alignment of frequency components that forms edges and recognisable structure. Preserving magnitude while scrambling phase can destroy scene geometry even though the energy at each frequency is unchanged. Chapter 05 · Multidimensional Signals and Fourier Analysis

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. Chapter 06 · Coded Apertures and Transfer Functions

Is the adjoint the inverse?

Not in general. The adjoint backprojects measurement information into scene coordinates, but it does not undo blur, mixing or information loss unless the operator has special orthonormal structure. Chapter 07 · Lensless Imaging and ADMM

What does a lenslet array trade?

A fixed detector divides samples between spatial position and ray direction. More pixels beneath each lenslet improve angular sampling but leave fewer independent spatial samples across the scene. Chapter 08 · Light Fields and Depth

What belongs in a strong project comparison?

Use a falsifiable question, matched geometry and exposure, an appropriate baseline, recorded calibration and parameter settings, repeated measurements, a metric with stated units and an honest account of failures and validity limits. Chapter 09 · Project Evidence and Exam Synthesis

How to study for the exam

Revise in linked passes rather than treating each imaging method as an isolated device. First draw the forward chain from scene to optics or code, sensor, calibration, reconstruction and metric. Next retrieve the equations that quantify each link: thin-lens geometry, sampling limits, convolution and Fourier transforms, linear measurement models, regularised inversion, ADMM updates and disparity depth. Then practise diagnosis.

For every artifact, name the assumption that would produce it and design a controlled capture that could distinguish that cause from noise, poor tuning or a different model error. Rework the laboratory calculations with changed numbers and explain the engineering consequence of each result.

For the project material, convert every broad claim into a comparison with fixed geometry, a baseline, repeated measurements and a stated range of validity. In the final week, use the mixed practice set under time pressure and audit units, signs, conditioning and evidence boundaries.

Because the exam is open book, optimise notes for navigation: one index page, compact equations with variable definitions, and a symptom-to-test table. Check current administrative instructions separately before the examination.