结合多分辨率修正曲率配准的层间插值

针对体数据各坐标轴分辨率不一致,导致医学及工业三维图像重建时出现边界台阶状结构、细节断裂或缺失等问题,提出了基于多分辨率修正曲率配准的层间插值方法。该方法采用反投影重建形式增强图像细节,解决配准图像清晰度和对比度弱的问题;利用三次卷积插值构造切片的低分辨率图像以保留图像细微结构,提高配准精度。采取从低分辨率粗配准到高分辨率精细配准的策略减少计算时间,提高计算效率。利用切片图像对应像素间存在对称形变结构的特征,建立了修正曲率模型估计形变场,解决了配准时单向形变不一致的问题。最后,通过离散余弦变换(DCT)的数值解析方案对构建的形变场估计函数进行优化,利用最终形变场数据对切片进行线性插值,计算出层间图像。实验结果表明,提出的算法能够消除现有方法插值图像的边缘模糊现象。与线性插值算法相比,提出方法的均方差(MSD)减小了40%,高于对称曲率模型,且耗时仅为该模型的20%左右,满足应用要求。

Slice interpolation on multilevel modified curvature-based registration

When 3D medical and industrial images are reconstructed, the differences of volume data resolution in three directions often lead to detail missing and surface discontinuous. Therefore, a slice interpolation algorithm was proposed based on multilevel modified curvature-based registration. With the method, a back projection method was used to enhance the image details and to improve the definition and contrast of the matched images. Then, the cubic convolution interpolation was used to construct a low resolution image for preserving image detail and to improve the matching accuracy. A new scheme from low resolution coarse registration to high resolution fine registration was proposed to reduce the calculation time to improve the calculation efficiency. Furthermore, a modified curvature-based registration model was built based on the symmetric transformation between the pixels in the slice images to resolve the problem of transformation inconsistency in image matching. Finally, the Discrete Cosine Transform, (DCT)was used to minimize and optimize the joint criterion of a deformation field and the deformation field data were used to perform the linear interpolation for the slice to obtain the slice image. Experiment results show the proposed method eliminates the blurred edges of the slice interpolation image. As compared with the linear interpolation method, the Mean Square Deviation(MSD) of the proposed method has reduced by 40%, which is higher than that of the classical curvature mode and the time consuming also is 20% that of the classical one.