提出一种基于小波域内统计建模的低剂量计算机X 线断层(CT) 图像去噪新方法。利用非下采样Contourlet 变换(NSCT)获得具有平移不变性的多尺度、多方向频率子带；结合噪声特点,通过统计参数预置改进隐马尔可夫树(HMT)模型,加速构建层间、方向间不同子带系数的概率转移矩阵,采用期望最大(EM)算法训练获得边缘概率密度；设计Bayes 最大后验概率(MAP)估计器对图像噪声进行建模与滤除。实验表明:相比小波HMT 去噪、Contourlet 软阈值去噪等同类方法,该方法提高了噪声估计精度,使图像峰值信噪比(PSNR)明显增加,细节信息更清晰。

A new low-dose computed tomography (CT) image denoising method based on statistical modeling in wavelet domain is presented. The shift -invariant nonsubsampled Contourlet transform (NSCT) is performed to obtain the multi -scale and multi -direction frequency subbands,then the proposed hidden Markov tree (HMT) is modeled to train the transitional probability matrix by a proposed statistically modeling acceleration,from where the marginal probability density between different subbands is built by expectation maximum (EM) algorithm,and finally the noise is modeled and filtered by a Bayes maximum a posteriori (MAP) estimator. The results indicate it achieves higher peak signal-noise ratio (PSNR),better edge detail and visual quality than other similar wavelet-based and Contourled-based denoising methods.