Skip to Main Content
This work proposes a novel speckle suppression method, called robust nonlinear wavelet diffusion. It shows that the log-transformed speckle can be approximated by Gaussian noise contaminated with long burst outliers. Consequently, we exploit this knowledge to design a speckle suppression filter within the framework of wavelet analysis. The outliers are removed by the combination of the robust-residual filter and nonlinear diffusion filter, and the Gaussian noise is eliminated by the wavelet soft-shrinkage technique. We validate the proposed method using synthetic and real echocardiographic images. The performance improvement over other traditional denoising filters is quantified in terms of noise suppression and structural preservation indices. Finally, using the denoised image, we improve the performance of the gradient vector flow snake by modifying its external force field, and we quantify the volume of left ventricle via segmentation applied to the echocardiographic image.