Video
BibTex
@inproceedings{Azami:2021:10.20380/GI2021.09,
author = {Azami, Rosa and Mould, David},
title = {Artistic Recoloring of Image Oversegmentations},
booktitle = {Proceedings of Graphics Interface 2021},
series = {GI 2021},
year = {2021},
issn = {0713-5424},
isbn = {978-0-9947868-6-9},
location = {Virtual Event},
pages = {68 -- 76},
numpages = {9},
doi = {10.20380/GI2021.09},
publisher = {Canadian Information Processing Society},
}
Abstract
We propose a method to assign vivid colors to regions of an oversegmented image. We restrict the output colors to those found in an input palette, and seek to preserve the recognizability of structure in the image. Our strategy is to match the color distances between the colors of adjacent regions with the color differences between the assigned palette colors; thus, assigned colors may be very far from the original colors, but both large local differences (edges) and small ones (uniform areas) are maintained. We use the widest path algorithm on a graph-based structure to obtain a spanning tree over the set of regions, then traverse the tree to assign colors in a greedy fashion. Our method produces vivid recolorings of region-based abstraction using arbitrary palettes. We demonstrate a set of stylizations that can be generated by our algorithm.