Removing speckle in SAR images while maintaining spatial resolution: from classical image processing techniques to unsupervised deep learning
Title: Removing speckle in SAR images while maintaining spatial resolution: from classical image processing techniques to unsupervised deep learning
Abstract: Synthetic Aperture Radar (SAR) imaging is a coherent imaging technique widely used for Earth observation. This technique has the ability to see through clouds and provide complementary information to optical imaging. Like other coherent imaging systems (sonar, ultrasound medical imaging, coherent optics) it suffers from the speckle phenomenon: images display a strong multiplicative noise.
In this talk I will describe how despeckling techniques have evolved from non-linear filtering up to the most recent self-supervised deep neural network techniques and how they account for the statistical model of speckle.