The Research Of Medical Ultrasound Image Denosing Algorithms Based On Multiscale Geometric Analysis

Ultrasound imaging has been widely used in clinical diagnosis with advantages of high security, no damage, convenient, low cost and so on. However, the quality of the ultrasound image will be reduced greatly, because multiplicative speckle noise is produced during imaging.The poor ultrasound image will not only affect the doctor's diagnosis, but also increase the difficulties of the image processing and analying subsequently. Therefore, the research of speckle noise suppression has been one of the important topics of medical ultrasound imaging technologies.Multiscale geometric analysis, which is a hot topic in image processing area currently, was successfully used in area of image denoising, fusion and so on, because it overcame the defects that wavelet could not optimally present singularity of line and surface in image. Specially, Contourlet and NSCT are widely used in imge denoising for the advantages of decomposing images in multiscales and multidrections. The Bayesian threshold was improved in this paper, which based on Contourlet and NSCT:Traditional Bayesian threshold was amended with different subband energy distributions of Log transformed ultrasound images in Contourlet and NSCT domain. In the Contourlet domain, the amended threshold got better denoising effect than Bayesian threshold denoising based on wavelet and scale adaptive threshold denoising based on Contourlet. Further more, it also gets obvious better denoising effect in the NSCT domain.In addition, Bayesian threshold was modified again with energy distribution characteristi-ics of subband images and an adjustable factor, which based on difference of noise variance on different scales in transform domain. The second modified threshold was used to denoise in NSCT domain, and better denoising effect was obtained.