ABSTRACT
High-resolution synthetic aperture radar (SAR) systems with various imaging modalities, including polarimetric, interferometric, multi-frequency, and multi-temporal information, are widely used in Earth observation and remote sensing. The occurrence of speckle in SAR images, however, affects the performance of all image processing techniques and may thus prevent full exploitation for various applications. Over the past 40 years, speckle reduction methods have been developed in the literature, highlighting the importance of this topic. Despite this extensive knowledge, speckle removal is still an open problem that is far from being fully solved. In this chapter, a comprehensive overview of despeckling methods for single and multi-channel SAR images is given and specifications of the different methods are pointed out. It provides an overview of nonlocal filtering techniques, focusing on the definition of the similarity measures between two noisy patches in an image. Different filtering performance evaluation methods are presented to evaluate denoised images. These methods focus on the use of reference and referenceless metrics, where the former are commonly used to evaluate results on synthetic datasets for which a reference is available, while the latter define denoising indicators that are useful for evaluating denoising performance on real datasets. The chapter further includes several experiments that are conducted with real SAR data, while it discusses the effectiveness of denoising techniques. Finally, upcoming speckle reduction methods based upon machine learning, such as deep learning, are discussed as a new generation of despeckling techniques.
