A computer has been taught to distinguish between different arrangements of the elements and principles of design across artists and over time. But don't worry, it's nowhere near human.
Much of what makes this possible is the amount of artwork that has been digitized, ie, machine readable, ie, machine consumable. We can now feed our computers the same diet of cultural nutrients that we as photosensitive creatures enjoy, giving it a fair chance.
The computer can't see the things we see in a work of art. It can't see the richness and it can't assign meaning anywhere near the way we can. But it can do some things even better than us, especially if we give it a concise metric with which is can "familiarize" itself with the artwork.
The primary metric used here was one of complexity-entropy, where each pixel in an image was measured in terms of the complexity of its spatial patterns. A fully white square, or a square filled with all manner of colors and lines. Each pixel value is then added to give the picture an overall score.
The score of the different artworks in the database shifted in tandem with major accepted periods of artistic periods, making it look like this program has figured something out, can recognize something about our artifacts of cultural assimilation.
Machine learning enables physics-inspired metrics for analyzing art
Aug 2018, phys.org
No comments:
Post a Comment