A comparative study of motion estimations and super resolutions on digital breast tomosynthesis projection images

摘要

Super Resolution (SR) is the image processing technique that reconstructs a high resolution (HR) image from multiple low resolution images. Image registration is a necessary part of SR reconstruction. The accuracy of motion estimation in registration step affects the quality of the result HR image directly. In this paper, we study the feasibility of SR in Digital Breast Tomosynthesis (DBT) projection images. We compare the frequency and spatial domain methods of motion estimation in DBT projection image registration. Moreover, we compare different SR reconstruction methods that include bi-cubic interpolation, iterative back projection, and robust SR. The results show that motion estimation in the spatial domain is more precise than that of frequency domain. For SR reconstruction, robust SR algorithm outperforms the others.