Self-Adaptive Magnetic Resonance Image Denoising Based On Lifting Wavelet

Magnetic resonance imaging(MRI) techniques have been applied widely in the medicine diagnosis and the scientific research for its unique characters such as non-invasive, pollution-free, imaging mechanism with several parameters and the capacity of reflecting in vivo organization, energy metabolism features etc. However, the noise caused by imaging mechanism, the outside environment as well as the patients will make the processing and application based on the MR images face two problems. One is that the noise in the background can decrease the lossless compression ratio seriously and the other is that the noise can make the boundary fuzzy so that the recognition and the analysis of the image detail become more difficulty. So denoising of the MR image has vital significance for the medical diagnosis and the scientific research.The wavelet transform has become one of the important tools in the field of image denoising. With the good time-frequency characters the denoising methods based on the wavelet not only can reduce noise but also keep the image details. So techniques based on thresholding of wavelet coefficients are gaining popularity as approaches to denoising image. In Most of the denoising methods, the noise models are assumed as Gaussian distribution. However, the noise in the MR images obeys Rician distribution. Unlike additive Gaussian noise, Rician noise is signal-dependent and consequently separating signal from noise is a very difficult task. Based on the analysis of present wavelet denoising methods, this paper studies a new thresholding method using a wavelet transform improved by lifting scheme for MR image.This paper mainly focuses on two aspects research work. Firstly, according to the theory of wavelet and its lifting scheme, we analyze how the property of the wavelet effect the construction error of the image and then propose a improved method to increase the vanishing moment of a wavelet by lifting algorithm. By increasing the vanishing moment, the performance of traditional wavelet is improved to have the ability of capturing more details with a better vibration, consequently enhances the reconstruction precision. Besides, lifting implementations of wavelet transform offers other two advantages in comparison to the traditional wavelet, e.g. faster and in-place calculations. Secondly, according to the MR image characteristic and the noise distribution property, a more self-adaptive threshold selection method is proposed to thresholding wavelet coefficients. On each decomposition level, the coefficient matrix of the high frequency is deblocked to several sub-matrixes. The decomposition level, the contrast and the absolute median of a selected sub-matrix are combined to determine the threshold used to process the corresponding coefficients of the sub-matrix. So the thresholds determined by this method have a better self-adaptive performance.A large number of experiments on MR image are performed. The simulation results indicate that the denoising algorithm on MR image proposed in this article obtains a better performance, especially for MR image with lower signal-noise ratio.