Detail-preserving despeckling under multiscale anisotropic diffusion in medical ultrasound images

In this paper, a detail-preserving despeckling method is proposed for speckle suppression of displayed ultrasonic images. Our method is performed under the framework of anisotropic diffusion, and it is summarized by three steps: classification, outlier shrinkage and anisotropic diffusion. A normalized modulus multiscale product is defined and used instead of gradient as edge detection operator in the process of diffusion. Diffusion threshold is adaptively determined using classified homogeneous speckle regions to produce a robust result. To avoid speckle spreading, outliers in homogeneous regions are removed before each iteration of anisotropic diffusion. Experiments on synthetic and real displayed ultrasonic images show that the proposed method compares favorably to some other anisotropic diffusion-based and outlier shrinkage-based methods.