Computer Algorithm Can Copy Artists' Styles
- September 02, 2015 14:51
Researchers in Germany claim that a computer algorithm can copy the styles of artists to create new works in about an hour.
Their paper "A Neural Algorithm of Artistic Style," submitted to the Nature Communications journal, outlines how a “convolutional neural network” (CNN) can take a foundation image (eg., photo) and transform it into a work that accurately mimics the artistic style of such greats as Munch, van Gogh or Picasso.
Leon Gatys, an author of the paper (with Alexander S. Ecker and Matthias Bethge) and an University of Tuebingen PhD student, wrote:
“The key finding … is that the representations of content [the foundation image] and style [of specific artworks] in the convolutional neural network are separable.
“That is, we can manipulate both representations independently to produce new, perceptually meaningful images.”
