A new metric for measuring structure-preserving capability of despeckling of SAR images

In this paper, a new metric called despeckling structure loss (DSL) is proposed for performance assessment of despeckling algorithms with a focus on the preservation of structural features such as edges and textures. The ratio image of a synthetic aperture radar (SAR) image to its despeckled result can clearly indicate the loss of structural features caused by the despeckling process. By taking into account characteristics of the best and worst structure preservation in despeckling, the DSL metric examines the presence of structural features in ratio images by using local correlations between the ratio image and the noise-free reference image at edge points. The DSL is shown to be a monotonic function of structure loss bounded between the best and worst cases, leading an objective and quantitative measure of the structure-preserving capability of despeckling algorithms. Experimental evaluations of the proposed metric have been carried out on simulated SAR images including one generated by SAR raw signal simulator. Despeckled results using five typical algorithms clearly demonstrate the efficiency of the DSL metric. In comparison, the commonly used despeckling metrics including the signal-to-mean-square error ratio, the edge correlation index, the Pratt's figure of merit, the structural similarity and the equivalent number of looks of ratio images, fail to keep a consistency with the structure loss shown in despeckled results as well as ratio images.