| In recent years, image denoising methods based on multiscale geometric analysis (MGA) become a research hotspot. Multiscale geometric analysis method solved the shortcomings of multi-dimensional signal representation of Wavelet transform, which widely used in image denoising, image enhancement, image fusion, image segmentation, image restoration, feature extraction and other related fields. Among MGA, Shearlet transform and nonsubsampled Shearlet transform (NSST) is the last multiscale geometric analysis tool, which is optimally sparse multidimensional representation for images, with the analysis ability of multiscale, multidirectional and local, and widely used in image denoising. This paper will use the nonsubsampled Shearlet transform to study ultrasound image denoising, and two aspects will be improved on the base of traditional NSST thresholding:(1) As the traditional Shearlet thresholding method performs the same threshold in each resolution, which is difficult to remove effetely the noise at each resolution. To solve this problem, a method that the threshold in each high-resolution of Shearlet domain weighted with scale function is proposed, on the basis of nonsubsampled Shearlet transform, and perform different threshold at various scale. This method will not only can remove the noise from various frequency, and preserve the image details as possible.(2) On the other hand, traditional NSST thresholding method was executed only at each high-solution, and low-solution NSST coefficients are still retained. At the low-solution has still a lot of low-frequency noise and weak edge, therefore, the KAD model method will be executed to remove the low-frequency noise and preserve the weak details as possible.By contrast the overall visual effect, local details and the gray scale curve of the particular scan line indicated that, the improved algorithm provides an improvement not only in noise reduction but also the preservation of edge features. |
PDF Full Text Request