CBIR-SAR System Using Stochastic Distance

Area of Science:

• Remote Sensing
• Image Processing
• Geophysics

Background:

• Synthetic-Aperture Radar (SAR) imagery is crucial for Earth observation.
• Content-Based Image Retrieval (CBIR) systems require robust feature extraction and similarity measures.
• Existing SAR image retrieval methods may not fully capture textural nuances.

Purpose of the Study:

• To develop and evaluate a novel CBIR system for SAR images using stochastic distance.
• To accurately model SAR data intensity using the GI0 distribution and estimate its parameters.
• To assess the retrieval performance using Mean Average Precision (MAP) across different SAR sensors.

Main Methods:

• Estimating roughness (α^) and scale (γ^) parameters of the GI0 distribution using Maximum Likelihood Estimation and Log-Cumulants.
• Employing triangular stochastic distance to measure similarity between query and database SAR images.
• Evaluating performance with Mean Average Precision (MAP) on synthetic and real SAR data from UAVSAR, OrbiSaR-2, and ALOS PALSAR sensors.

Main Results:

• The proposed CBIR-SAR system achieved high MAP values, exceeding 0.833 for real SAR images across all polarization channels.
• The system demonstrated effectiveness in retrieving heterogeneous regions in synthetic images.
• Results confirmed the method's sensitivity to texture, relying on accurate parameter estimation.

Conclusions:

• The developed stochastic distance-based CBIR system offers effective retrieval for SAR imagery.
• Accurate estimation of GI0 distribution parameters is critical for successful image retrieval.
• The method shows promise for applications in analyzing forest and urban areas using SAR data.