Figure 2: Synthesised input images and optimization results, using Total Curvature-squared as the regularization term within the cost function. | Scientific Reports

Figure 2: Synthesised input images and optimization results, using Total Curvature-squared as the regularization term within the cost function.

From: Image reconstruction from photon sparse data

Figure 2

We are using the Root Mean Square of each image against the ground truth to determine the quality of each reconstruction. The two data sets shown contain 50,000 and 5,000 photons. The first row of each set shows the ground truth, the respective synthesised data set and the reconstruction obtained using bootstrapping (with log-likelihood/pixel penalties of 0.484 and 0.175 for the 50,000 and 5,000-photon images, respectively), while the second row shows three further reconstructions with different log-likelihood/pixel () penalties for comparison (0.25, 0.5, 0.65 and 0.1, 0.15, 0.2 for the 50,000 and 5,000-photon images, respectively). The Root Mean Square normalised with respect to the average image intensity (RMS, inset) is shown for each image. Image sizes are all 240 × 240 pixels.

Back to article page