双能CT利用两组不同能谱下的衰减信息, 准确分割两种基材料。 在实际应用中, 物体内部材料结构复杂、 组分多元化, 想了解其内部结构信息往往需要获取三种及三种以上基材料图像。 常规CT是连续混合谱束, 获取的投影信息与单能重建算法不匹配, 重建图像中各基材料的衰减系数存在误差。 由于工业领域材料的密度普遍较大, 所以重建图像中基材料的噪声更严重, 影响各组分表征准确度, 尤其对于衰减系数接近的材料区分难度更大。 为实现双能数据分解得到多幅高质量基图像, 除了重建图像本身存在的噪声影响外, 材料分解模型中系数矩阵的选取也十分重要。 然而重建图像中的衰减系数值与理论衰减系数值存在偏差, 重建图像中密度接近的不同材料的衰减系数相近甚至相等, 导致待分解像素的材料三元组选取错误, 降低了材料分解精确度。 因此, 提出一种在层间约束下的三维块匹配多成分分解方法。 该方法引入有质量体积守恒和每个像素至多包含三类材料约束的多材料分解模型, 将三维结构相似性信息加入到像素材料成分选取中, 利用三维结构信息进行约束求解达到降低噪声污染的目的, 获取大致分解的初始基图像; 再利用三维块匹配方法对初始基图像进行匹配, 对各基材料图像进行三维特征约束分类, 分类后选取含有该类基材料的最优材料成分三元组进行多材料分解, 得到更准确的组分表征图。 金属纯模体和花岗岩两组实验中, 与已有方法的结果图进行对比, 层间约束下三维块匹配分解方法对衰减系数接近的工业材料的识别更准确, 各组分表征图结构更完整, 图像质量更好, 细节处理也更精确。 金属纯模体实验中的定量分析表明, 相比已有方法, 该方法的PSNR值和SSIM值分别提高了5%～6%和31%～35%。 验证了该算法的有效性和鲁棒性, 在常规CT系统下实现了更精确的多成分分解。

Dual-energy CT uses two groups of attenuation information under different energy spectra to accurately segment two kinds of basis materials. In practical applications, the internal material structure of the object is complex, and the composition is diversified. It is often necessary to obtain three or more basic material images to understand its internal structure information. Conventional CT is a continuous mixed spectral beam. The projection information obtained does not match the single-energy reconstruction algorithm, and there are errors in the attenuation coefficients of each basic material in the reconstructed image. The density of materials in the industrial field is generally larger and the noise of the basic material in the reconstructed image is more serious, affecting the accuracy of each component’s characterisation, especially for the materials with similar attenuation coefficients. In order to realize dual-energy data decomposition to obtain multiple high-quality basis images, in addition to the influence of noise in the reconstructed image, the selection of the coefficient matrix in the material decomposition model is also very important. However, there is a deviation between the attenuation coefficient value in the reconstructed image and the theoretical attenuation coefficient value. In the reconstructed image, the attenuation coefficients of different materials with similar densities are similar or even equal, resulting in the wrong selection of the material triplet of the pixel to be decomposed, which reduces the accuracy of material decomposition. Therefore, it proposes a 3D block-matching multi-component decomposition method with inter-layer constraints. In the method, a multi-material decomposition model with mass volume conservation and the constraint that each pixel comprises three types of materials at most is introduced. Three-dimensional structure similarity information is added to the selection of pixel material components. Constraint solution is carried out by utilizing the three-dimensional structure information to reduce noise pollution, and an initial basis image, roughly decomposed, is obtained; Then use the three-dimensional block matching method to match the initial basis image, and classify the three-dimensional feature constraints of each basis material image. After classification, select the optimal material composition triplet containing this type of basic material to carry out a multi-material decomposition model and obtain a more accurate component characterization diagram. In two groups of experiments on pure metal phantoms and granite. Compared with the results of the existing methods, the three-dimensional block matching decomposition method under interlayer constraints is more accurate in identifying industrial materials with similar attenuation coefficients, the structure of each component characterization image is complete, the image quality is good, and the detail processing is accurate. The quantitative analysis in pure metal phantom experiments shows that the PSNR and SSIM values of the proposed method are increased by 5%~6% and 31%~35%, respectively, compared with the existing methods. The effectiveness and robustness of the algorithm are verified, and more accurate multi-component decomposition is achieved under the conventional CT system.