Model-based edge detector for spectral imagery using sparse spatiospectral masks.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Edge detection is crucial for analyzing spectral imagery.
• Isoluminant edges, defined by color but not intensity changes, pose a challenge for traditional methods.
• Prior knowledge of spectral signatures aids in material identification.

Purpose of the Study:

• To develop novel model-based algorithms for spectral edge detection.
• To specifically address the challenge of detecting isoluminant edges.
• To improve accuracy and efficiency compared to existing edge detection techniques.

Main Methods:

• Developed two model-based algorithms utilizing spectral-band ratios to define edge signatures.
• Integrated spectral signatures into a sparse joint spatiospectral nonlinear operator.
• Employed a classifier-enhanced extension for adaptive feature accentuation.
• Validated algorithms using airborne hyperspectral and mid-infrared imagery.

Main Results:

• Proposed algorithms demonstrated superior accuracy in edge detection, particularly for isoluminant edges.
• Outperformed benchmark algorithms like multicolor gradient (MCG) and hyperspectral/spatial detection of edges (HySPADE).
• Achieved significant data compression through efficient band selection, reducing operations per pixel by 71x compared to MCG.

Conclusions:

• The developed algorithms offer a robust solution for spectral edge detection, especially in challenging scenarios with isoluminant edges.
• These methods provide a more accurate and computationally efficient approach to analyzing spectral imagery.
• The band selection strategy enables substantial data reduction, making the algorithms practical for large datasets.