HyTexiLa: High Resolution Visible and Near Infrared Hyperspectral Texture Images.

Area of Science:

• Materials Science
• Computer Vision
• Spectroscopy

Background:

• Hyperspectral imaging captures detailed spectral information across numerous bands.
• Texture analysis is crucial for material characterization and identification.
• Existing datasets may lack the required spectral and spatial resolution for advanced texture studies.

Purpose of the Study:

• To introduce the HyTexiLa dataset, a collection of close-range hyperspectral images of 112 materials.
• To detail the calibration and distortion correction methods for hyperspectral image acquisition.
• To explore hyperspectral texture analysis using spectral complexity and advanced feature extraction techniques.

Main Methods:

• Acquisition of high spectral and spatial resolution reflectance images using hyperspectral sensors.
• Data calibration and correction for image acquisition distortions.
• Spectral analysis using non-negative matrix factorization (NMF) for spectral complexity.
• Extension of local binary pattern (LBP) operators for hyperspectral texture analysis.

Main Results:

• The HyTexiLa dataset provides detailed spectral and spatial texture information for 112 materials.
• Non-negative matrix factorization revealed generally low spectral complexity for individual textures.
• Hyperspectral texture classification performance improves with an increased number of spectral bands.
• Opponent band local binary pattern features demonstrated superior performance in texture classification.

Conclusions:

• The HyTexiLa dataset is a valuable resource for studying material textures using hyperspectral imaging.
• Increasing spectral resolution enhances the ability to classify material textures.
• Advanced feature extraction methods, like opponent band LBP, are effective for hyperspectral texture analysis.