VF3 has been tested on the MIVIA LDGraphs dataset, a new and very wide dataset composed by dense graphs whose size ranges from 300 to 10.000 nodes.
VF3 has been compared with the following four state of the art algorithms (whose executables used in the experimentation can be requested, together with VF3, at the following link):
- VF2, the predecessor of VF3.
- RI: it is based on tree search with backtracking (like VF2 and VF3) and it was the winner algorithm in the ICPR 2014 contest on graph matching for biological databases.
- L2G: it is based on tree search with backtracking (like VF2, VF3 and RI) and it was the second algorithm in the ICPR 2014 contest on graph matching for biological databases.
- LAD: it is based on a constraint propagation approach; it should have some advantage when dealing with very symmetric or repetitive structures, and should be less sensitive to the density of the graphs.
Some graphics useful to confirm the effectiveness of VF3 are reported below, for both unlabeled and labeled graphs.
|
|
|
| η = 0,2 | η = 0,3 | η = 0,4 |
|
|
|
| η = 0,2 | η = 0,3 | η = 0,4 |
|
|
|
| η = 0,2 | η = 0,3 | η = 0,4 |