Study On Ultrasound Image Speckle Noise Reduction Algorithm

Ultrasonic testing is widely used in medical diagnose due to its noninvasive nature, real-time and low cost, which uses ultrasonic reflection and diffraction characteristics to detect whether internal organs or the surface is abnormal. Ultrasound images contain two types of noise, one is additive electrical or thermal noise, and another one is speckle. The speckle, a form of locally correlated multiplicative noise, is caused by the interference of the energy from randomly distributed structure scatters, which degrades the contrast resolution in ultrasound images and obscures the diagnosing details. Ultrasound images are not only affected by speckle noise but also influenced by the thermal noise. Traditional methods are not ideal to remove speckle and noise, since they are not better in preserving edges and details.To overcome the limitations of existing methods, the paper gives different improved schemes with the noise of ultrasound images; the main research work is as follows:1. We construct the two-dimensional ultrasound speckle model to study the location of distribution characteristics and magnitude of the statistical properties, and discuss that the density of scatter is related to ultrasound speckle. Finally, the clinical ultrasound speckle is simulated.2. We improve the original bilateral filtering algorithm, which is proposed based on Gaussian noise. We know the filtering effect of bilateral filter is poor and it can not preserve edges well. To preserve fine details and edge in a noisy image, it is necessary to propose a scheme for the edge detection and noise detection before filtering to make the improved method remove Gaussian noise or impulse noise. In view of the impulse noise filter, the Gaussian function in the bilateral filter is improved as Rayleigh distribution function to suppress the speckle.3. We proposed the Rayleigh-trimmed bilateral filtering algorithm which is a "detect and remove" filtering mechanism and makes full use of the statistical characteristics. Firstly, the target pixel is classified as noise, speckle or noise-free based on the sorted quadrant median vector (SQMV) scheme by calculating a reference median in the filtering window and in turn comparing with the target pixel. Afterwards, bilateral filter and Rayleigh-trimmed filter are switched to filter noise and speckle, respectively, while keeping the noise-free pixel unchanged. Visual and numerical indices on synthetic and real ultrasound images are used to evaluate the performance of the proposed method;:experiment results.indicate that the-algorithm is effective on speckle and noise suppression and fine details preservation.