Filtering & Enhancement

Overview

Close Filtering & Enhancement help.

1. Load a hyperspectral dataset.

2. Open Correct / Filtering & Enhancement.

4. Adjust the available parameters.

6. The main status changes to Processing while the operation is running.

7. The working cube is updated and the display is refreshed.

8. Use Reset Filters to restore the original dataset when available.

Important

Filtering changes the current working cube in memory. It does not automatically overwrite the original file on disk.

Applies local averaging in the image plane independently for each spectral band.

The filter size defines the square spatial neighborhood. A value of 3 uses a 3 x 3 neighborhood. If an even value is entered, IDCubePro increases it to the next odd integer.

Larger kernels produce stronger spatial smoothing but can blur edges and small objects.

• Spatial pixel noise is present.
• Small local intensity fluctuations should be suppressed.
• Spectral band positions should remain unchanged.

This operation smooths spatial structure. Fine features smaller than the selected neighborhood may be reduced.

Applies Gaussian smoothing to the hyperspectral cube using MATLAB imgaussfilt3.

Sigma controls the width of the Gaussian distribution. Larger sigma values produce stronger smoothing.

Defines the Gaussian kernel size. IDCubePro forces the value to an odd integer.

Important

Because this is a 3D operation, smoothing can extend through both spatial and spectral dimensions. This is different from the Spectral Mean Filter, which is restricted to the spectral axis.

Strong 3D smoothing may broaden or suppress narrow spectral features.

Smooths each pixel spectrum along the band dimension using a [1 1 N] averaging kernel.

Defines the number of neighboring spectral bands included in the moving average. Even values are increased to the next odd integer.

Reduces high-frequency spectral noise while preserving the spatial image geometry.

Large windows can broaden peaks, reduce narrow absorption features, and lower spectral resolution.

Asymmetric Least Squares (als)

Provides baseline-oriented spectral smoothing using an asymmetric least-squares model.

Light: lower smoothness and fewer iterations.

Balanced: general-purpose default.

Strong: stronger baseline smoothing.

Smoothness controls the baseline penalty. Larger values favor a smoother baseline.

Asymmetry controls unequal weighting around the fitted baseline.

Iterations controls repeated reweighting.

Broad spectral baseline or background variation interferes with narrower spectral features.

Excessive smoothing can alter scientifically meaningful broad spectral structure.

Applies hyperspectral denoising through the denoiseNGMeet routine available to IDCubePro.

Light preserves more detail.

Balanced is the default compromise.

Strong applies more aggressive denoising.

Sigma represents the assumed noise level.

Spectral Subspace controls the reduced spectral representation.

Iterations controls repeated denoising passes.

NGMeet can be computationally intensive for large cubes. The main IDCubePro status remains Processing while the denoising callback is running.

Strong denoising can suppress weak spectral or spatial features together with noise.

Applies spatial-spectral total-variation denoising through SSTV_2.

Lambda, mu, and nu control the regularization terms used by the SSTV algorithm.

Iterations controls optimization length.

Light uses weaker regularization and fewer iterations.

Balanced provides a general-purpose setting.

Strong uses larger regularization values and more iterations.

Noise is present across both spatial and spectral dimensions and preservation of piecewise-smooth structure is desirable.

Over-regularization can flatten subtle spatial textures or spectral variation.

FFT filters operate along the spectral dimension and require a valid wavelength vector.

Preserves slowly varying spectral components while suppressing rapid spectral fluctuations. This can reduce high-frequency spectral noise.

Emphasizes rapid spectral changes while suppressing broad slowly varying background structure.

The cutoff determines which spectral-frequency components are retained or suppressed. The useful value depends on the spectral sampling and scientific objective.

FFT filtering can introduce spectral artifacts when cutoff settings are too aggressive or when spectra contain discontinuities.

• Standard Deviation Normalize

Parameter dialogs do not change the main status simply by opening. Processing is shown when an Apply callback actually begins computation. Ready is restored when the callback finishes.

The current dispatcher launches mssgolay_1_2025b. Its Apply callback is located in that separate function, so Processing/Ready status for Savitzky-Golay must be added there rather than in this dispatcher.