A Benchmark for Despeckling Filters

Abstract:
The validation of filters for SAR image denoising is mostly performed through personal choices of experimentation on a set of author-selected synthetic and true SAR images, which makes the objective elucidation of their benefits over well-established filtering methods hard. Following the example of other scientific fields, this situation can be alleviated by providing a set of reference images to test new methods on thorough quantitative measures of quality. We propose an extensible benchmark for speckle denoising techniques. We also stress the need to establish a repository with a pool of error-free and correctly codified denoising filters and quality estimators.