Other Reconstruction Methods
This module provides alternative methods for reconstructing images from measurements, which are not well described by other modules in the library.
Learned Filtered Back-Projection
The simplest method for reconstructing an image from a measurements is to first map the measurements to the image domain via a non-learned mapping, and then apply a deep network to the obtain the final reconstruction. This idea was introduced by Jin et al. “Deep Convolutional Neural Network for Inverse Problems in Imaging” for tomographic reconstruction.
Artifact removal architecture \(\phi(A^{\top}y)\). |