Continuum Hull Removal

Overview

Continuum Hull Removal normalizes each pixel spectrum by an interpolated convex-hull continuum.

The purpose is to reduce broad spectral baseline and slope effects so that absorption-band depth and shape can be compared more directly.

The transformation is applied independently to every pixel spectrum in the hyperspectral cube.

For each spectrum, IDCubePro:

1. Uses the wavelength vector as the spectral sampling axis.

2. Adds small artificial anchor positions just outside the spectral range.

3. Builds a convex hull around the spectrum.

4. Keeps the valid hull points belonging to the original spectral samples.

5. Interpolates a continuum between those hull points.

6. Divides the original spectrum by the interpolated continuum.

Because every pixel can have a different spectral shape, the convex hull is calculated independently for each pixel.

Interpretation Of The Output

After continuum removal, values near 1 generally represent portions of the spectrum lying near the continuum.

Absorption features commonly appear below 1.

Deeper absorption features generally produce lower continuum-removed values, although interpretation still depends on spectral quality, wavelength sampling, and the application.

Raw reflectance spectra can differ because of broad brightness changes, spectral slope, illumination, sample geometry, or other slowly varying background effects.

Continuum removal reduces those broad effects and emphasizes relative absorption structure.

This can make absorption-band position, depth, width, and shape easier to compare across spectra.

Continuum removal is commonly useful for reflectance-type spectra containing absorption features.

Examples include mineral, vegetation, food, pharmaceutical, chemical, tissue, and material spectra where relative absorption structure is scientifically meaningful.

• Spectra contain substantial noise.
• The wavelength range does not adequately define the shoulders around an absorption feature.
• Very few spectral bands are available.
• Saturated or invalid bands distort the convex hull.
• Strong artifacts create artificial hull points.
• An interpolation method produces overshoot or oscillation.

Continuum removal is generally most meaningful after obvious acquisition artifacts have been corrected.

Depending on the dataset, this may include reference correction, removal of invalid bands, denoising, or other instrument-specific preprocessing.

The exact order should be selected according to the measurement physics and kept consistent across datasets that will be compared.

IDCubePro uses matrices.Wavelengths when it exists and contains one wavelength value per spectral band.

If a valid wavelength vector is not available, the tool falls back to band index 1, 2, 3, and so forth.

Using real wavelengths is preferable whenever spectral positions have physical meaning.

The interpolation method controls how the continuum is reconstructed between convex-hull support points.

makima is Modified Akima interpolation.

It is the default and is a good general-purpose choice because it is smooth while usually avoiding the strong oscillations that can occur with conventional cubic splines.

linear connects neighboring hull points with straight lines.

It is simple, stable, easy to interpret, and usually avoids overshoot.

Use linear when you want a conservative piecewise-linear continuum.

nearest uses the value of the nearest hull point.

It produces step-like continuum segments and is usually less natural for smoothly varying spectra.

next uses the next sample value between hull points.

It also produces step-like behavior and is mainly useful for specialized interpolation needs.

pchip uses shape-preserving piecewise cubic interpolation.

It is smooth and usually limits overshoot.

pchip is a strong alternative to makima when a smooth shape-preserving continuum is desired.

cubic uses cubic interpolation.

It can provide a smooth continuum but may behave differently from shape-preserving approaches near sharp spectral curvature.

v5cubic reproduces legacy MATLAB cubic interpolation behavior.

Use it mainly when compatibility with an older analysis workflow is required.

spline uses a smooth cubic spline.

It can produce a very smooth continuum but may overshoot between hull points.

Inspect results carefully when using spline, especially with sparse or noisy spectra.

WHICH METHOD SHOULD I START WITH?

Start with makima for most datasets.

Compare pchip when shape preservation is especially important.

Use linear when you want the most conservative and transparent interpolation.

Use spline only when you have verified that overshoot does not distort the continuum.

For scientifically important datasets, compare the continuum-removed spectra produced by two or more reasonable interpolation methods.

If absorption depth or shape changes substantially with the interpolation method, the result may be sensitive to continuum definition and should be interpreted cautiously.

Apply performs continuum removal on every pixel spectrum in the current myData.Images cube.

The current interpolation method is read from the dropdown at the time Apply is clicked.

Processing is pixel-wise because each spectrum can have its own convex hull.

Large hyperspectral cubes can therefore require substantial computation time.

The progress dialog updates periodically rather than on every pixel to reduce user-interface overhead.

The progress dialog can be canceled.

If cancellation occurs before the processed cube is committed back to myData, the current working cube remains unchanged.

After successful processing, myData.Images is replaced by the continuum-normalized working cube.

The spatial dimensions, number of spectral bands, and wavelength vector are preserved.

This function does not intentionally overwrite myDataOriginal.

The original dataset can therefore remain available for the broader IDCubePro Reset Preprocessing workflow when that original copy exists.

After continuum removal, IDCubePro clears cached display images and refreshes the current display.

Both the modern IDCubePro interface and the classic/legacy interface are supported by the refresh logic.

IDCubePro records the continuum-removal operation in processing history.

The history entry includes the interpolation method, cube dimensions, wavelength range, and the fact that the working dataset was changed.

Continuum removal is often used before measuring relative absorption depth.

A simple depth measure may be related to 1 minus the continuum-removed reflectance at the absorption minimum.

However, the scientifically appropriate band-depth metric depends on the application and should be defined consistently.

Continuum removal can help identify the wavelength of an absorption minimum because broad spectral slope has been reduced.

Band position can still be affected by noise, sampling interval, smoothing, and overlapping absorptions.

The transformed spectrum can also be used to compare band width, asymmetry, shoulders, and other shape characteristics.

These measurements are only meaningful when spectral resolution and signal-to-noise ratio are adequate.

Noise can introduce artificial local maxima that influence the convex hull.

If the continuum appears unstable, consider appropriate denoising or smoothing before continuum removal.

Avoid excessive smoothing that could alter real absorption features.

Continuum definition near the beginning and end of the spectral range can be less reliable because there may be insufficient spectral information outside the feature.

Interpret absorption features near spectral boundaries with additional caution.

The algorithm protects against division by a zero or invalid interpolated continuum by replacing extremely small continuum values with a numerical epsilon.

This prevents Inf and NaN values but does not make physically invalid input spectra scientifically meaningful.

When comparing continuum-removed spectra across samples, use consistent wavelength ranges, spectral sampling, preprocessing, and interpolation method.

Changing any of these can alter the continuum and therefore the apparent absorption depth or shape.

Continuum-removed spectra may be useful as features when class differences are driven by relative absorption structure rather than absolute brightness.

However, continuum removal is not automatically beneficial for every classifier or dataset.

Its value should be evaluated using an appropriate validation strategy.

Continuum removal changes the original spectral intensity scale.

Do not interpret continuum-removed values as the original reflectance, radiance, fluorescence intensity, or concentration.

Use the transformed data only for analyses whose meaning is compatible with relative continuum normalization.

1. Load and inspect the hyperspectral dataset.

2. Perform necessary reference correction and artifact removal.

3. Remove unusable spectral bands if appropriate.

4. Open Continuum Hull Removal.

7. Inspect representative spectra after processing.

8. Verify that absorption features are enhanced without obvious interpolation artifacts.

9. If necessary, compare pchip or linear.

10. Continue downstream analysis using one consistent method.

Common Problem - Oscillating Continuum

If the continuum shows unrealistic oscillation, use makima, pchip, or linear instead of spline-like interpolation.

Common Problem - Unstable Results

If small changes in interpolation method produce large changes in the normalized spectrum, inspect noise, spectral range, hull support points, and feature boundaries.

Common Problem - Very Few Hull Points

A poorly sampled or unusual spectrum may not produce enough valid hull points for a meaningful continuum.

The function reports an error when a valid hull cannot be computed.

Common Problem - Absorption Disappears

If a feature becomes weak or distorted after continuum removal, check whether the convex hull is crossing the feature incorrectly or whether the spectral range does not adequately define the continuum shoulders.

Continuum removal is a preprocessing transformation, not a physical calibration by itself.

Results depend on wavelength range, spectral sampling, noise, interpolation method, preprocessing, and the shape of the measured spectrum.

Interpret continuum-removed features within the measurement and application context.

Close Continuum Hull Removal help.