双树轮廓波变换域的磁共振图像降噪

Magnetic resonance image denoising in dual-tree Contourlet transform domain

为了改善磁共振(MR)图像的质量,提出一种基于双树轮廓波(DT-Contourlet)变换的MR图像降噪算法。研究了MR图像的噪声分布模型,认为这种噪声服从莱斯分布,从而推导了MR模平方图像的噪声参数估计方法。通过分析DT-Contourlet的塔型双树方向滤波器组结构,明确了DT-Contourlet不仅能保持轮廓波灵活的方向选择性,而且克服了传统轮廓波不具有平移不变性的缺点。在DT-Contourlet变换域,通过计算方差一致性测度,用局部自适应窗口估计阈值萎缩因子,对MR模平方图像的变换系数进行阈值萎缩。最后,经过DT-Contourlet反变换,实现了MR图像的降噪处理。实验结果表明,用本文算法降噪的MR仿真图像的峰值信噪比(PSNR)优于传统算法;与基于小波和轮廓波的方法相比,不同噪声方差下的PSNR平均提高了2.13dB和0.91dB。从视觉效果来看,该算法能在有效抑制MR图像噪声的同时,更好地保持图像的细节信息。

In order to improve the quality of Magnetic Resonance （MR） images, a denoising algorithm for a MR image using Dual-Tree Contourlet （DT-Contourlet） transform is proposed. The distribution model of noise of the MR image is investigated, and a method to estimate the noise parameters of the squared magnitude MR image is derived based on the assumption that such noise obeys Rician distribution. Then, the pyramidal dual-tree directional filter bank of DT-Contourlet is analyzed to show that DT-Contourlet maintains the flexibility direction selectivity of the Contourlet transform, and overcomes the shortcomes of the Contourlet in lack of shift invariance. After that, the locally adaptive window is used to compute the shrinkage factor to shrink the DT-Contourlet coefficients of the squared magnitude MR image in the DT-Contourlet domain by calculating the Variance Homogeneity Measurement （VHM）. Finally, the denoising algorithm to MR image is implemented via the inverse DT-Contourlet transform. Experimental results show that the Peak Signal-Noise Ratio （PSNR） of simulated MR images by proposed algorithm is superior to that by traditional algorithms. With different noisevariances, the PSNR of new algorithm is high 2.13 dB and 0.91 dB than those of wavelet-based and contourlet-based algorithms averagely. For visual quality, the proposed algorithm can reduce the noise in MR images effectively and retain more details simultaneously.