Adaptive Contourlet Fusion Clustering for SAR Image Change Detection

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Multi-temporal synthetic aperture radar (SAR) image analysis requires robust change detection methods.
• Existing unsupervised methods often struggle with noise and detail preservation.

Purpose of the Study:

• To propose a novel unsupervised change detection method for multi-temporal SAR images.
• To enhance the accuracy of identifying changed regions by effectively fusing information and incorporating non-local data.

Main Methods:

• Adaptive Contourlet fusion is employed to integrate complementary information from difference images.
• Different fusion rules are applied to Contourlet coefficients for low and high-frequency bands.
• A fast non-local clustering algorithm (FNLC) is developed, incorporating both local and non-local information to classify fused images.

Main Results:

• The proposed method generates a binary image highlighting changed regions.
• Contourlet fusion effectively restrains details in unchanged regions while emphasizing changed areas.
• FNLC reduces noise impact while preserving crucial details of changed regions.

Conclusions:

• The novel adaptive Contourlet fusion clustering method demonstrates state-of-the-art performance.
• The approach is effective for real-world applications on both small and large-scale SAR datasets.
• The integration of local and non-local information in FNLC significantly improves change detection accuracy.