Unsupervised despeckling performance evaluation for SAR images
Long Sun, Lei Zhu · 2016
To evaluate despeckling performance of SAR images using the equivalent number of looks(ENL) and edge keeping index(EKI), a new unsupervised evaluation method is proposed. First, ratio-based edge strength map(RESM) and direction information are calculated by anisotropic Gaussian kernel(AGK) bi-windows. Second, SAR image is divided into homogeneous regions and edge regions by thresholding RESM with a threshold estimated by an adaptive method, and thin edges can be extracted by the non-minimun suppression. Third, each local ENL of all pixels in homogeneous regions is estimated, and each local EKI of all pixels in edge regions is estimated. Then contrast curves can be obtained by statistical method for all local ENL and EKI. Experimental results show that the proposed method need not select estimation regions by artificial operation, and can get over insufficient evaluation and avoid the deviation estimation produced by the selection of different estimation regions.