Feature Asymmetry Anisotropic Diffusion for Speckle Reduction

Ultrasonography has been increasingly used in the clinical diagnosis and therapy, but doctors often suffer great difficulties to interpret the ultrasound data due to the speckle that severely degrades the image quality. In this paper, we propose to reduce speckle noise in ultrasound images using feature asymmetry anisotropic diffusion (FAAD). The proposed approach is an adaptive diffusion process that can preserve the image features while suppressing the noise by incorporating a local phase -based edge detector, called feature asymmetry (FA), into the forward -and -backward diffusion. Unlike the intensity -based operators, the FA measurement is theoretically intensity -invariant and it can effectively discriminate the edges from noise even if they have similar gradient response. This property is very essential for the subsequent diffusion process because it supervises FAAD to perform the forward diffusion in speckled regions for noise removal and inhibit the smoothing on the edges with different image intensities, resulting in better preservation of the low contrast edges. Meanwhile, the backward diffusion in our FAAD can better protect the intensity contrasts of features by reserving the diffusion process happened at features. In addition, the parameters involved are automatically computed in order to enhance the robustness of the proposed approach so that it can be adapted to different images without repetitive parameter tuning. We validate the proposed approach on clinical ultrasound images and compare segmentation accuracies on despeckled results. Experimental results demonstrate that our approach performs better than state-of-the-art despeckling methods in terms of speckle reduction and edge preservation.