基于多尺度隐马尔可夫模型的CR影像降噪方法研究

在CR成像过程中不可避免的要引入各种干扰和噪声,只有弄清干扰图像信息的各种噪声来源、特征及其与信号的相互关系,才能有效地将之消除.在分析CR成像系统的基础上,文章指出影响CR图像质量的噪声主要是固有噪声和X线量子噪声,在统计规律上它们分别服从高斯分布和泊松分布.本文针对CR的固有噪声从小波系数的统计规律出发,根据固有噪声的特点,结合混合高斯模型描述小波系数的统计特征,采用两个状态的隐马尔可夫模型描述小波系数在尺度之间的相关性和依赖性,用最大期望值( EM)算法估计隐马尔可夫模型在各个尺度上的参数,然后按照尺度大小逐级对小波系数进行维纳滤波,最后是小波逆变换恢复图像.文章最后还给出了实验结果,并与其它降噪算法进行了比较.

MHMM-based computed radiography image-denoising method

In the process of computed radiography imaging, a lot of noises will be brought into the system. Therefore, only by knowing their resources, characteristics and relationship with signals, can the authors smooth them. On the basis of analyzing the computed radiography system in detail, the authors point out that there are two kinds of noises affecting the quality of a computed radiography image: Gaussian white noise and Poisson noise. Firstly, the statistical characteristics of wavelet coefficients and Gaussian white noise are summarized. Then, they are described with the mixture Gaussian model and hidden Markov model (HMM), which can fit the dependency of wavelet coefficients between scales. Finally, the novel algorithm developed by the authors is compared with other wavelet-based denoising methods.