Can cubic B-splines interpolate data?

Yes, when coefficients are calculated to satisfy the interpolation conditions; raw samples are not generally the required coefficients.

Does cubic mean degree four?

No. Cubic means degree three, although some sources use order four for that degree.

Is every cubic image interpolator a B-spline method?

No. Several different cubic constructions exist.

AI Glossary / Computer vision and multimodal AI / Advanced

Cubic B-spline

Definition

A cubic B-spline uses piecewise polynomials of degree three as locally supported basis functions. Weighted combinations represent smooth curves or signals, including continuous image models for resampling. At simple interior knots, the usual cubic spline has continuous first and second derivatives. Knot repetition and boundary choices can change continuity and endpoint behavior.

What it means

Instead of treating an image as isolated pixel values, a spline representation describes a function that can be evaluated between sample locations. The coefficients in that representation are not generally identical to the original samples. This distinction is essential: directly using samples as coefficients for a cubic B-spline basis usually smooths them, whereas properly calculated coefficients can produce interpolation that passes through the samples. It is therefore misleading to say that cubic B-splines can only approximate and can never interpolate. The construction and coefficient calculation determine the result.

How it works

Choose a knot sequence, degree three and coefficients, then sum coefficient-weighted basis functions. Each basis function has local support, so only nearby coefficients contribute at a location. In uniform sampled-signal applications, a prefilter or equivalent coefficient solve can prepare the representation for interpolation. Evaluation then supplies values at desired positions. Boundary assumptions remain part of that calculation. The basis, coefficient estimation and resampling procedure should be distinguished, because merely naming the cubic basis does not specify whether the final operation interpolates, smooths or extrapolates.

Three practical examples

An image-processing researcher evaluates a spline representation at new coordinates during an image transformation, checking its boundary treatment and coefficient preparation.
A graphics developer builds a smooth curve from control coefficients and inspects how changing one coefficient affects the nearby curve segment.
A student compares direct sample-as-coefficient evaluation with a fitted interpolating spline, demonstrating why the two constructions need not agree at original sample locations.

Think of it this way

Imagine building a smooth surface from overlapping shaped supports. Their heights are control settings, not necessarily points that the final surface must pass through. If you want the surface to hit specified measurements, you must solve for suitable settings. This analogy separates coefficients from observations. It also explains local support: changing one setting affects the nearby region supported by its basis function rather than necessarily changing the whole curve in the same way.

Worked example

For an original uniform one-dimensional illustration, the centered cubic B-spline basis has values 1/6, 2/3 and 1/6 at offsets −1, zero and one. If neighboring coefficients are [0,6,0], evaluation at the center gives 0/6+6×2/3+0/6=4, not six. Thus treating a sample value of six as the center coefficient does not automatically reproduce that sample. To interpolate a whole sample sequence, coefficients must instead satisfy the corresponding system of equations, including boundary conditions. This small calculation shows why “basis coefficient” and “sample value” are different concepts; it is not a complete coefficient solution or an image-quality comparison.

Illustrative case study

A fictional mapping team changes its image-resampling routine and notices that fine markings look softer. It investigates whether the new routine expects raw samples or precomputed spline coefficients. A small test at original grid locations reveals that the configuration does not reproduce the input samples as intended. The team corrects the preparation step and then evaluates transformed images for the actual mapping task. It also reviews boundaries and repeated resampling. This scenario illustrates diagnosing a representation mismatch rather than claiming that one interpolation family is always visually superior or appropriate for every kind of image.

Common misconceptions and mistakes

“Cubic” identifies polynomial degree, not a guarantee that the curve passes through its coefficients. The basis's positivity does not imply that every interpolating reconstruction is a simple convex average of raw samples, because calculated coefficients can differ substantially from those samples. Another mistake is using cubic B-spline and all bicubic interpolation methods as interchangeable names. Knot choices, boundary conditions and coefficient preparation matter. Extrapolation outside the fitted interval is also a separate policy, and smooth continuation does not make predictions beyond available data reliable.

Business use

Spline representations can support image transformation, curve modeling and numerical workflows where smooth evaluation between samples is useful. A business evaluating a software change should check that the intended interpolation or smoothing behavior is actually configured. Preserve the relevant boundary and coefficient settings for reproducibility. Compare task outcomes and detail retention on representative inputs rather than relying on the word “cubic” as a quality ranking. Where geometric or intensity measurements matter, validate the full resampling process and avoid assuming that a visually smooth output is necessarily a more accurate representation.

Check your understanding

Why did the center evaluate to four?

Answer: The center basis weight was two thirds and the neighboring coefficients were zero.

What would be needed to reproduce a specified sample sequence?

Answer: A suitable coefficient solve or prefilter with defined boundary behavior.

Why must the word “order” be checked?

Answer: Its convention can differ from polynomial degree.

Related terms

B-spline basis
knot vector
spline coefficient
interpolation
prefiltering
image resampling

Learn next

Evaluate a uniform cubic basis at integer locations and compare coefficient values with resulting function values. Then fit a cubic spline through a small sample set and verify interpolation at the original coordinates. Inspect boundary and extrapolation behavior separately.

Sources

SciPy, BSpline documentation: https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.BSpline.html Unser, Splines: A Perfect Fit for Signal and Image Processing: https://bigwww.epfl.ch/preprints/unser9902p.pdf The coefficient calculation, support analogy and mapping-team case are original educational illustrations. The entry qualifies the source glossary's approximation-only shorthand.

Previous: Cryptographic splitting Next: Cuboid (computer vision)