A Noisy SAR Image Fusion Method Based on NLM and GAN

Area of Science:

• Remote Sensing
• Image Processing
• Artificial Intelligence

Background:

• Speckle noise in synthetic aperture radar (SAR) images degrades image quality and hinders downstream applications.
• The grayscale nature of SAR images limits their direct applicability due to human visual system's sensitivity to color.

Purpose of the Study:

• To develop an effective method for fusing noisy SAR images, addressing both noise reduction and the incorporation of color information.
• To enhance the utility of SAR imagery for human interpretation and subsequent processing tasks.

Main Methods:

• A pre-processing step using nonlocal matching to group similar image blocks.
• Generative adversarial networks (GANs) to generate noise-free fused SAR image blocks with enhanced color and spatial resolution.
• Aggregation of processed image blocks to form the final fused image.

Main Results:

• The proposed method successfully generates fused SAR images that are noise-free and possess color information.
• Experimental results on SEN1-2 datasets demonstrate superior fusion performance compared to existing state-of-the-art methods.
• The method exhibits robustness against various levels of image noise.

Conclusions:

• The presented noisy SAR image fusion method significantly outperforms current techniques.
• The integration of nonlocal matching and GANs offers a promising direction for enhancing SAR image quality and applicability.
• The developed technique addresses key limitations of SAR imagery, paving the way for broader applications.