LAUSR

Speckle reduction is an inevitable preprocessing activity in many medical and satellite imaging applications. Particularly, ultrasonic data is prone to high-amplitude fluctuations leading to speckled appearance which hinders the image analysis phase. A non-local variational framework has been designed in this paper using symmetric maximum mean discrepancy (MMD) as the similarity measure. The speckle intensity is assumed as Gamma distributed in this framework as it stands close to the actual distribution in contrast to a Gaussian in many other works. The similarity metric designed based on the symmetric MMD tends to preserve local structures pretty well when compared to the ones based on the euclidean distance. This nonparametric kernel-based measure compares the distribution of the patches to determine the similarity instead of the intensity differences and therefore seems more effective in retaining the structures present in the images. A comprehensive experimental analysis is performed to demonstrate the performance of the model.