In the following we give some extra detail about the methods we compare against in our multi comparison.

We implemented the method described in

Kartic Subr, Cyril Soler, Fredo Durand,
Edge-preserving Multiscale Image Decompostion based on Local Extrema, ACM Transactions on Graphics (TOG) Volume 28, Issue 5 (Proc. SIGGRAPH Asia 2009).

We tried both applying the method to RGB channels and to our separated CMY channels, followed by Neugebauer reconstruction using Eq. 1 in our paper. The method performed slightly better on RGB inputs, so we are showing these here.

We generated results using the "Descreen" plugin for the GIMP image editing application.

The plugin essentially implements the method described in this paper:

C. J. Stanger, T. Tran, and B. Smith,
Descreening of color halftone images in the frequency domain,
Proc. SPIE 7866, 78661H, 2011.

We generated results using the original author's implementation of

H. Siddiqui, C. A. Bouman,
Training-based de-screening,
IEEE Trans. Image Proc. 16, 3, 789--802, 2007.

The software may be obtained here. Unfortunately, we were not able to scan our images on a HP Scanjet 8250 that the method was trained for. Instead, we scanned the images on a very similar HP Scanjet 8300. Nevertheless, the results might be somewhat degraded due to that. We obtained a license for the commercial Descreen 5 plugin for Photoshop. A trial version of the plugin may be obtained here.

We generated the results using the Gaussian blur implementation in Adobe Photoshop CS5.

We compare against the method described in

Z. Farbman, R. Fattal, D. Lischinski, R. Szeliski,
Edge-preserving decompositions for multi-scale tone and detail manipulation,
ACM Trans. Graph. 27, 3, Article 67, 2008.

We used the original author's implementation.

We compare against results generated by following the online tutorial here. Essentiall, Gaussian smoothing followed by median filtering and unsharp masking.