An Anisotropic Diffusion Filter Based on Multidirectional Separability

Extracting tongue contour from high noised ultrasound image is a key issue of observing speech production procedure. Anisotropic diffusion has been widely used in reducing speckle noise of ultrasound images but it is not very effective in preserving edges and tends to blur them. Hence the blurred edges hamper the succeeding contour-based pattern analysis or modeling. In this study, we modify the standard SRAD (speckle reducing anisotropic diffusion) to improve its edge detection and suppress the intrinsic edge blurring effect of SRAD by exploiting the multidirectional separability. We experimented with both synthetic and real ultrasound images by SRAD and the proposed approach. The extracted contours in denoised images by SRAD and the proposed approach are compared in terms of the corresponding accuracy, both subjectively and objectively. The results show the proposed approach performs better than the conventional SRAD and more accurate contours can be obtained for post processing.