2018Kim FourierError: Difference between revisions
(Created page with "== Citation == Trampert, P.; Wang, W.; Chen, D.; Ravelli, R. B. G.; Dahmen, T.; Peters, P. J.; Kübel, C. & Slusallek, P. Exemplar-based inpainting as a solution to the missi...") |
No edit summary |
||
| Line 1: | Line 1: | ||
== Citation == | == Citation == | ||
T. Kim, J. Haldar. The Fourier radial error spectrum plot: A more nuanced quantitative evaluation of image reconstruction quality. Proc. IEEE Intl. Symposium on Biomedical Imaging, 2018. | |||
== Abstract == | == Abstract == | ||
In the modern biomedical image reconstruction literature, the quality of a reconstructed image is often numerically quantified using scalar error measures such as mean-squared error or the structural similarity index. While such measures provide a rough summary of image quality, they also suffer from well-known limitations. For example, a substantial amount of information is necessarily lost whenever the characteristics of a high-dimensional image are summarized by a single number. In this work, we introduce the Fourier radial Error Spectrum Plot (ESP), which provides a novel and more nuanced assessment of error by decomposing the error into its different spatial frequency components. The usefulness of ESP is illustrated in the context of MRI reconstruction from under-sampled data. In addition, we demonstrate that the extra dimension of insight provided by ESP can be used to improve the performance of existing image reconstruction techniques. | |||
== Keywords == | == Keywords == | ||
Latest revision as of 12:13, 29 June 2018
Citation
T. Kim, J. Haldar. The Fourier radial error spectrum plot: A more nuanced quantitative evaluation of image reconstruction quality. Proc. IEEE Intl. Symposium on Biomedical Imaging, 2018.
Abstract
In the modern biomedical image reconstruction literature, the quality of a reconstructed image is often numerically quantified using scalar error measures such as mean-squared error or the structural similarity index. While such measures provide a rough summary of image quality, they also suffer from well-known limitations. For example, a substantial amount of information is necessarily lost whenever the characteristics of a high-dimensional image are summarized by a single number. In this work, we introduce the Fourier radial Error Spectrum Plot (ESP), which provides a novel and more nuanced assessment of error by decomposing the error into its different spatial frequency components. The usefulness of ESP is illustrated in the context of MRI reconstruction from under-sampled data. In addition, we demonstrate that the extra dimension of insight provided by ESP can be used to improve the performance of existing image reconstruction techniques.
Keywords
Links
https://ieeexplore.ieee.org/abstract/document/8363523/