A New Image Quality Index for Objectively Evaluating Despeckling Filtering in SAR Images

摘要

Synthetic aperture radar (SAR) images are corrupted with a multiplicative granular-like noise pattern known as speckle. The goal for a despeckling filter consists of suppressing the speckle while preserving all the scene features such as texture, point-type targets, and, especially, edges. There exist several speckle filtering techniques and a relevant number of image quality indexes to evaluate the performances of a filtering operation on an SAR image. However, assessing the superiority of a filter over other is not a trivial issue. In this work, we present a new referenceless estimator (-ratio estimator) based on the ratio edge detector which allows helping in objectively evaluating a filter realization on SAR images. The proposed estimator operates on the ratio image obtained as the point-to-point ratio between the original image (noisy image) and the filtered image. An ideal filter operation on an image implies that, in areas where speckle is fully developed, the ratio image should have the features of pure speckle and no geometric content. The new estimator measures the remaining geometric content within the ratio image. This new estimator is easy to compute and it provides an excellent metric to rank a filtering operation on real SAR images.