Comparison to
Lai et al. 2009
Lai et al. 2009
|
We compare against method described in Yu-Kun Lai and Shi-Min Hi and Ralph R. Martin, Automatic and Topology-Preserving Gradient Mesh Generation for Image Vectorization, ACM Transactions on Graphics 23(3), 2009 We used the author's implementation to create these results. Please note: most vectorization tools, such as the algorithm described in the Lai et al.'s paper, are designed for much larger input images. In the paper we give a detailed explanation for why they perform less successfully on tiny pixel art inputs. The results below do not make a statement about the quality one can achieve on input images within the design range of these tools, they merely show that there is a need for specialized algorithms for very tiny images. It should also be noted that our algorithm does not generalize to large images with noise or anti-aliased edges. Please see Lai et al's paper to see some very nice results on such challenging images. |
Boo
Super Mario World, © Nintendo |
 |
 |
 |
| Input | Nearest (16x) | Our Method (16x) | Lai et al., 2009 |
Mario and Yoshi 2
Super Mario World, © Nintendo |
 |
 |
 |
| Input | Nearest (8x) | Our Method (8x) | Lai et al., 2009 |
Bomber 4
Super Bomberman, © Hudson Soft |
 |
 |
 |
| Input | Nearest (16x) | Our Method (16x) | Lai et al., 2009 |
Yoshi 3
Super Mario World, © Nintendo |
 |
 |
 |
| Input | Nearest (16x) | Our Method (16x) | Lai et al., 2009 |