Hyperspectral Texture Metrology Based on Joint Probability of Spectral and Spatial Distribution.

Area of Science:

• Remote Sensing
• Image Analysis
• Computer Vision

Background:

• Texture characterization is crucial for interpreting hyperspectral images.
• Existing methods often lack physical relevance and direct interpretability.
• Capturing both spectral and spatial complexity remains a challenge.

Purpose of the Study:

• To introduce a novel, physically relevant texture feature for hyperspectral images.
• To develop a generic formulation for capturing spectral and spatial complexity.
• To validate the proposed feature against established methods in diverse applications.

Main Methods:

• A relative spectral difference occurrence matrix (RSDOM) was formulated.
• RSDOM was constructed in a multireference, multidirectional, and multiscale context.
• Performance was evaluated on texture classification (HyTexiLa), content-based image retrieval (ICONES-HSI), and land cover classification (Salinas).

Main Results:

• RSDOM achieved 98.5% accuracy in texture classification.
• It yielded 80.3% precision for top 10 retrieved images in CBIR.
• Land cover classification reached 96.0% accuracy post-processing.
• RSDOM outperformed GLCM, Gabor filter, LBP, SVM, CCF, CNN, and GCN.

Conclusions:

• RSDOM provides a metrologically valid texture representation.
• The feature demonstrates superior performance across multiple hyperspectral imaging tasks.
• RSDOM offers advantages in feature size and applicability regardless of spectral characteristics.