Image Quality Metrics

Overview

Image Quality Metrics evaluates every spectral band of the current hyperspectral cube using one or more no-reference image-quality metrics.

IDCubePro supports BRISQUE, NIQE, and PIQE when the corresponding MATLAB functions are available.

Unlike full-reference image-quality methods, these metrics do not require an ideal reference image.

1. Load a hyperspectral dataset.

2. Open Image Quality Metrics.

3. Select BRISQUE, NIQE, PIQE, or a combination of metrics.

5. IDCubePro evaluates each spectral band independently.

6. Inspect the resulting quality-score curves.

7. Copy the results or export them to CSV if required.

BRISQUE stands for Blind/Referenceless Image Spatial Quality Evaluator.

BRISQUE estimates image quality from spatial statistics without requiring a reference image.

Lower BRISQUE scores generally indicate better perceived image quality.

NIQE stands for Natural Image Quality Evaluator.

NIQE measures how strongly an image differs from statistical characteristics expected for natural images.

It does not require a reference image or a training image from the current dataset.

Lower NIQE scores generally indicate better image quality.

PIQE stands for Perception-based Image Quality Evaluator.

PIQE estimates image quality by examining local spatial distortion and block-level image characteristics.

Lower PIQE scores generally indicate better perceived image quality.

Calculate runs every selected metric independently on every spectral band.

A progress window shows the current band and allows the calculation to be canceled.

Depending on cube size, number of spectral bands, computer performance, and selected metrics, calculation may require some time.

Before calculating the metrics, each spectral band is independently normalized to the range 0 to 1.

The minimum finite pixel value becomes 0 and the maximum finite pixel value becomes 1.

This allows the MATLAB image-quality functions to operate consistently across bands with different intensity ranges.

The resulting plot shows image-quality score as a function of wavelength or spectral band/component.

When wavelength information is available, the x-axis displays wavelength.

When the dataset is operating in band/component mode, the x-axis displays the band or component index.

Lower BRISQUE, NIQE, and PIQE values generally indicate better image quality.

Large changes in score between neighboring wavelengths may identify spectral regions with degraded spatial image quality.

Such changes can result from detector sensitivity, illumination, optical transmission, noise, low signal level, artifacts, or other acquisition characteristics.

Important

BRISQUE, NIQE, and PIQE were originally developed primarily for conventional images.

They are not hyperspectral-specific physical quality measures.

For hyperspectral imaging, IDCubePro applies them independently to individual spectral bands.

Therefore, the most useful interpretation is often the relative band-to-band trend rather than treating an individual score as an absolute hyperspectral quality measurement.

Image Quality Vs Signal-to-noise Ratio

Image Quality Metrics and Signal-to-Noise Ratio measure different properties.

SNR measures signal strength relative to noise.

BRISQUE, NIQE, and PIQE evaluate spatial image statistics and perceived image degradation.

A spectral band can therefore have relatively high SNR while still receiving a poor no-reference image-quality score, or vice versa.

The tool can help identify wavelength regions where spatial image quality becomes substantially worse.

This information may be useful when selecting spectral ranges for visualization, classification, clustering, dimensionality reduction, or other downstream analyses.

When comparing image-quality scores between datasets, acquisition and preprocessing conditions should be kept consistent whenever possible.

Relevant factors include exposure time, illumination, sensor gain, spatial resolution, spectral resolution, correction procedures, smoothing, and other preprocessing operations.

Copy Results places the spectral coordinate and calculated metric values on the system clipboard as tab-delimited text.

The copied data can be pasted into Excel, MATLAB, Prism, or another analysis program.

Export CSV saves the current results as a comma-separated file.

The first column contains wavelength or band/component position.

Additional columns contain the selected BRISQUE, NIQE, and PIQE scores.

Clear Plot removes the current results from the graph and clears the stored image-quality results.

It does not modify the loaded hyperspectral cube.

The availability line reports whether BRISQUE, NIQE, and PIQE are available in the current MATLAB installation.

Unavailable metrics are disabled automatically.

These metrics should not be interpreted as direct measurements of spectral accuracy, chemical specificity, spectral resolution, radiometric calibration, or biological relevance.

They evaluate spatial image characteristics of individual bands.

For a more complete assessment of hyperspectral data quality, consider these results together with SNR, calibration quality, spectral shape, detector characteristics, and application-specific validation.

Close the Image Quality Metrics help window.