Ultrasound Image Denoising Using Improved Image Decomposition Method

摘要

Ultrasound images are very noisy. Along with system noise, a significant noise source is the speckle phenomenon caused by interference in the viewed object. The noise of ultrasound image was always deemed as multiplicative noise, but the present multiplicative noise models are not fitting for it very well, because the noise is more complex than the simple multiplicative one. So we should consider separate the real part from large noised image using image decomposition for the purpose of denoising. The VO (Vese-Osher) model can decompose image into smooth area and the oscillating patterns, but it unable to deal with image with large oscillating part very well. To deal with this problem, we modify the VO model by add a fourth order filter and we call this improved decomposition methods as mixted VO model. The combined algorithm takes the advantage of both filters since it is able to preserve edges and separate large oscillating part. The minimization problem must be realized through solving partial differential equations with complex finite difference scheme which leads to low efficiency. To simplify this, the Split Bregman algorithm for the mixted VO model is proposed and then we use it for ultrasound image denoising.