We have focused, as before, in the Evolutionary Art capabilities of OSGiLiath. This time we have chosen OpenCV rather than Processing for performance and also because it includes more metrics and image analysis.
OpenCV has been compiled for several platforms (MacOS X, Ubuntu…) and added to the project, with their appropriated Java bindings.
New kind of Artistic Primitives have been added: the Patches. A patch is just a piece of other image. A script in Perl made by @jjmerelo has also been added to the project, with a lot of patches extracted from Caltech101 image benchmark. Individuals can now be a vector of patches, each one in a different positions. The service OpenCVCollageDrawer extracts this information to create a “collage” made of patches, and the MatchingFitness evaluates the individuals to “approach” them to a predefined image, but other feature extraction operations have been added (such as the Histogram).
The Human Guidance module has also been improved: now each image selected increases in 1 vote, and the other decrease 0,33 votes. Help us selecting images here! We will use the knowledge gathered for future experiments.
The first results, only applying the regular matching are not very outstanding, but it’s a start: