Assessing Performance of Multitemporal SAR Image Despeckling Filters via a Benchmarking Tool

In this work we present a novel benchmarking framework for assessing the quality of despeckled multitemporal SAR images on an objective basis. Taking cues from a recent work providing a useful tool for measuring the performance of single-channel despeckling filters, here we extend that analysis to multitemporal filters and datasets. Due to the lack of reference speckle-free real SAR images, the quantitative performance parameters introduced here are evaluated on simulated time series of canonical scenes obtained through a reliable and well-assessed SAR simulator which accounts for both geometrical and electromagnetic properties of the scattering surface. First we analyze image quality over datasets with homogeneous reflectivity in time. Then, for a more realistic performance assessment in practical situations, ad-hoc quality parameters are introduced to measure the effects of time-varying scene characteristics on the despeckled dataset. The consistency of the proposed framework is tested on state-of-the-art multitemporal despeckling algorithms.