基于Contourlet阈值法的锥形束CT图像去噪研究

目的：将多尺度分析工具之一的Contourlet变换运用到锥形束CT（CBCT）图像去噪领域,并对Contourlet不同阈值去噪方法进行探讨。提出基于Contourlet变换结合半软阈值方法对锥形束CT去噪,并论证去噪效果。方法：利用Contourlet变换的多尺度多方向性以及平移不变性,对低分辨率锥形束CT图像进行拉普拉斯塔形滤波和方向滤波多层分解后得到变换系数,随后对变换系数采用不同阈值方法进行处理,最后逆序反变换得到去噪后图像。通过软阈值和硬阈值方法在Contourlet变换中的应用,提出半软阈值结合Contourlet变换方法对锥形束CT图像去噪。通过对头,胸,盆腔各10例临床锥形束CT图像的去噪,比较三种阈值去噪效果。结果：半软阈值法在胸部和盆腔部锥形束CT图像去噪中比Contourlet硬阈值去噪在PSNR上平均高出1.40 d B和3.11 d B,但在头部锥形束CT图像处理中无优势,而Contourlet软阈值去噪后的锥形束CT图像在消除噪声的同时,信号自身的能量被消弱最多。结论：本文半软阈值法在一定程度上修正了硬,软阈值函数的缺陷,结合Contourlet变换在处理图像几何结构方面的优势,为锥形束CT图像去噪提供了一个新思路。

Denoising Study of Cone-Beam CT Images Based on the Contourlet Threshold Method

Objective： Contourlet transform, as one of the multi-scale analysis tools, was applied to the Cone-Beam CT （CBCT） denoising in this article. Meanwhile, different Contourlet thresholding denoising methods were discussed. A denoising method was proposed based on Contourlet transform combined with？semi-soft thresholding method for Cone-Beam CT denoising, and proved its effect. Methods： As Contourlet transform was utilized the advantages of in its multi-scale, multi-direction and its shift invariance, the Cone-Beam CT of low resolution were multilayer？decomposed by Contourlet transform which consisted of Laplacian pyramid and Directional filter banks. Then, the transform coefficients were filtered by different thresholding methods. Finally, denoising images were reconstructed through reverse transformation. Comparisons between the soft-thresholding method and the hard-thresholding method were made based on the Contourlet transform. A denoising method was proposed based on Contourlet transform combined with semi-soft thresholding method for Cone-Beam CT denoising. We compared different threshold methods in order to find the optimal One, and made statistical analysis and comparison with the denoising qualities of clinical images of different body parts. Results： The study showed that the soft-thresholding method and the hard-thresholding method had their different advantages. In the chest and pelvic part, the semi-soft threshold- ing denoising method performs much better than the hard-thresholding method by an average of 1.40 dB and 3.11 dB. But in the head part, it had no advantage. And by denoising of the soft-thresholding method, the energy of the Cone-Beam CT image was attenuated most. Conclusions： In this paper, the semi-soft thresholding method based on Contourlet transform which can distinguish between noise and edge, combined the advantages of the hard-thresholding method and the soft one. It had proved that the Contourlet semi-soft thresholding method might be a brand-new way for Cone-Beam CT i