Kim Kwangdon, Lee Kisung, Lee Hakjae, Joo Sungkwan, Kang Jungwon
Department of IT-Convergence, Korea University, Anam-ro 145, Seongbuk-gu, Seoul, South Korea.
Department of Bio-Convergence Engineering, Korea University, Hana B, Anam-ro 145, Seongbuk-gu, Seoul, South Korea.
Jpn J Radiol. 2018 Jan;36(1):59-67. doi: 10.1007/s11604-017-0697-9. Epub 2017 Oct 30.
We aimed to develop a gap-filling algorithm, in particular the filter mask design method of the algorithm, which optimizes the filter to the imaging object by an adaptive and iterative process, rather than by manual means.
Two numerical phantoms (Shepp-Logan and Jaszczak) were used for sinogram generation. The algorithm works iteratively, not only on the gap-filling iteration but also on the mask generation, to identify the object-dedicated low frequency area in the DCT-domain that is to be preserved. We redefine the low frequency preserving region of the filter mask at every gap-filling iteration, and the region verges on the property of the original image in the DCT domain.
The previous DCT2 mask for each phantom case had been manually well optimized, and the results show little difference from the reference image and sinogram. We observed little or no difference between the results of the manually optimized DCT2 algorithm and those of the proposed algorithm.
The proposed algorithm works well for various types of scanning object and shows results that compare to those of the manually optimized DCT2 algorithm without perfect or full information of the imaging object.
我们旨在开发一种填补间隙算法，尤其是该算法的滤波掩模设计方法，通过自适应和迭代过程而非手动方式使滤波器针对成像对象进行优化。
使用两个数值体模（Shepp-Logan和Jaszczak）生成正弦图。该算法以迭代方式工作，不仅在填补间隙迭代中，而且在掩模生成中，以识别离散余弦变换（DCT）域中要保留的对象专用低频区域。在每次填补间隙迭代时，我们重新定义滤波掩模的低频保留区域，并且该区域在DCT域中趋近于原始图像的特性。
每个体模案例先前的DCT2掩模已通过手动方式得到良好优化，结果显示与参考图像和正弦图几乎没有差异。我们观察到手动优化的DCT2算法结果与所提出算法的结果之间几乎没有差异。
所提出的算法适用于各种类型的扫描对象，并且在没有成像对象的完整或全部信息的情况下，其结果与手动优化的DCT2算法的结果相当。
