Computers can now paint like Van Gogh and Picasso
http://qz.com/495614/computers-can-now-paint-like-van-gogh-and-picasso/
Computers have now learned how to paint in the styles of the most famous painters with the advancement of machine learning. Researchers have developed a system that will interpret the styles of the painters and convert digital photographs into paintings of those styles. The system incorporates a deep artificial neural which finds certain patterns in objects to create the paintings. The system, working like a brain, shares similarities between Google’s Deep Dream system, which converted images to digital dream-like landscapes.
The system learns from artist’s different styles by examining the use of color, shapes, lines, brushstrokes and recreates the image in the style. The system can scale the extent of how much the image transforms to match how intense it matches a certain artist’s work. Although it may seem like a very complicated filter on many common applications such as Instagram or Snapseed, the system wants to make a computer vision system to interpret human creation methodology.
Using neural networks to convert existing images by large-scale machine learning is not a new concept, but it is the beginning of machine learning systems trying to learn the process of human creativity in different art forms. The article mentions a machine learning system from MIT that attempts to generate new pieces of classical music that adhere to classical forms. It will still take time for a computer to get to the level of human creativity, but progress in neural networks and machine learning will make it much faster to imitate human creativity—one day.
This article relates to graph theory in our Networks class because the system uses deep neural networks by processing large amounts of data to create neural networks to slowly convert each image to an actual painting. The system will look at data and see what data is closest like it to convert any image to an artist’s style. This is just like seeing data having strong or weak connections with other data to make predictions on how to create a digital painting. The machine learning algorithm will recommend which way to transform the painting by examining the entire network and then decide which data to use. It then does the same process repeatedly until the desired effect is reached. The system then uses neural networks in a recursive manner to learn an artist’s style and apply it to a painting.