Is it also called the projection-slice theorem?

Yes. That name emphasizes the relationship between a projection and a Fourier-domain slice.

Does it apply only in two dimensions?

The idea extends to higher dimensions, with the dimensions of projections and Fourier slices changing accordingly.

Why is the Fourier origin shared by projections?

In the ideal consistent model, the zero-frequency projection value corresponds to the total integral of the object, independent of projection orientation.

AI Glossary / Computer vision and multimodal AI / Advanced

Fourier slice theorem

Definition

The Fourier slice theorem states that the one-dimensional Fourier transform of a parallel projection of a two-dimensional function equals a line through the origin of that function's two-dimensional Fourier transform. The line's orientation corresponds to the projection direction under the chosen coordinate convention.

What it means

A projection combines information along parallel lines, such as summing values through an object at a particular angle. The theorem connects that collection of line sums with a specific part of the object's frequency-domain representation.

It helps explain how projections from different angles can support reconstruction. Each angle supplies another central line of Fourier information. More complete and adequately sampled angular coverage supplies information that a single projection cannot provide.

The word “slice” here refers to a slice in Fourier space, not simply a visible cross-section of the physical object. The theorem is a mathematical relationship underlying tomographic methods; turning it into a reliable practical reconstruction also requires sampling, noise and acquisition geometry to be addressed.

How it works

Let pθ(t) denote the integral of an object function along lines whose normal direction makes angle θ with the horizontal axis. Taking the one-dimensional Fourier transform with respect to t yields Pθ(ω).

With consistent transform conventions, Pθ(ω) = F(ω cos θ, ω sin θ), where F is the object's two-dimensional Fourier transform. Thus each projection provides values along a radial line through the Fourier origin.

A reconstruction method can combine frequency information from several projections or use an equivalent formulation such as filtered backprojection. Finite detector spacing, limited angles and noise leave practical gaps or uncertainties. Special acquisition geometries require appropriate adaptations rather than uncritical application of the simplest parallel-beam formula.

Three practical examples

1. Tomographic reconstruction: A system uses projections from multiple directions to estimate an object's cross-sectional structure, with the theorem explaining their frequency coverage.

2. Industrial imaging research: Engineers simulate projections of a known object to study what limited viewing angles leave poorly determined.

3. Computational-science education: A small image is summed along rows and columns, and the transforms of those sums are compared with corresponding lines in its two-dimensional transform.

Think of it this way

Think of learning about a woven pattern by receiving different directional summaries. Each summary constrains part of the pattern, but one direction leaves many possibilities unresolved.

The analogy is incomplete because the theorem gives an exact transform relationship under its assumptions, not merely a loose resemblance. The summaries are line integrals, and the corresponding information lies in frequency space.

Worked example

Take a tiny discrete array with rows [1, 2] and [3, 4]. Summing along the vertical direction gives the horizontal projection [4, 6]. Its two-point discrete Fourier transform is [10, −2], using the standard unnormalized forward convention.

The two-dimensional transform of the original array is [[10, −2], [−4, 0]] with this row-and-column ordering. The horizontal frequency line at zero vertical frequency is [10, −2], matching the transform of the projection.

This small discrete example illustrates the same algebraic connection for an axis-aligned sum. It does not show arbitrary-angle interpolation, continuous line integrals or a complete reconstruction system. Different transform normalizations may change scaling, so conventions must be consistent before comparing numerical results.

Illustrative case study

Fictional case: Fir Imaging tests a reconstruction setup that can view a sample from only a restricted range of angles. Its developers plot the corresponding Fourier coverage and see that some directions remain poorly constrained.

They compare reconstructions against known synthetic objects and report orientation-dependent errors rather than a single attractive example. Additional assumptions can make the images look cleaner, but the team keeps clear which information came from measurements and which came from the reconstruction model. The case illustrates the significance of missing angular data, without promising a particular solution to the acquisition limit.

Common misconceptions and mistakes

A projection is not generally a photograph of an object's exterior. The theorem's simple form assumes a defined line-integral model.

The Fourier slice is not a physical slice through the object. Confusing these spaces makes the reconstruction argument misleading.

One projection does not determine an arbitrary two-dimensional object. It supplies one line of frequency information, leaving many possibilities elsewhere.

The theorem does not remove practical reconstruction problems. Noisy data, finite sampling, inconsistent geometry and limited angles still matter. A direct discrete implementation also requires care because radial frequency samples do not automatically coincide with a convenient Cartesian grid.

Business use

The theorem helps teams understand the information available from projection-based imaging. It is useful for discussing acquisition coverage, reconstruction limitations and the role of additional assumptions, even when a software package hides the mathematical details.

Evaluation should use known references and the intended acquisition geometry. Distinguish mathematical validity of the relationship from performance of a particular discretization or reconstruction algorithm. A visually plausible result from sparse projections should not be treated as proof that all internal structure was measured accurately.

Check your understanding

A system measures only a few projection angles. Does the theorem fill all the unmeasured Fourier directions automatically?

Answer: No. It identifies where the measured information belongs. Unmeasured directions remain unconstrained by those projections unless additional evidence or assumptions are introduced. Reconstruction methods must handle that limitation explicitly.

Related terms

Radon transform
Fourier transform
Inverse problem
tomography
filtered backprojection
projection data
limited-angle reconstruction

Learn next

Verify the tiny row-and-column sum example using a discrete transform. Then study the Radon transform and draw how projection angles correspond to radial lines in Fourier space. Compare complete-angle and limited-angle synthetic reconstructions before interpreting results from real measurements.

Sources

Primary educational sources: A. C. Kak and Malcolm Slaney, Principles of Computerized Tomographic Imaging, author-hosted contents and Chapter 3 reference, https://www.slaney.org/pct/pct-toc.html ; UCL Mullard Space Science Laboratory, “The Fourier Backprojection Method,” https://www.mssl.ucl.ac.uk/www_solar/moses/moses-web/Pages/fourier_backprojection.htm . The UCL explanation was read; the linked book chapter could not be fetched during this check. The discrete calculation is original.

Previous: Fourier analysis Next: Fourier space