Computer scientists from the University of Bath have written a software which transforms your ordinary photographs and movies into cubist works of art and animation reminiscent of Picasso. They trained their software to identify important elements of a face, such as a nose, eye or mouth, until the computer learned how to recognize them on its own. This was achieved by giving the software a kind of 'aesthetic sense.' Then, by "using photographs of a subject taken from multiple points of view, the software automatically picks out important areas within the image, which are cut out as chunks. The chunks are statistically shuffled and a few of them randomly selected and distorted into a 'cubist' composition ready for digital painting." The software is not yet publicly available, but software and animation companies have expressed interest.
Here are more details on how this software works.
In order to create the software, the researchers had to teach the computer how to pick out the elements of photographs that, until now, only humans have been able to recognise as important.
By giving the computer an ‘aesthetic sense’, Dr Peter Hall and Dr John Collomosse from the Department of Computer Science, were able to create a series of automated artworks with new effects, such as making a Picasso cubist-style picture from ordinary photos.
The researchers fed the computer a series of pictures where they had identified the aesthetically important elements, such as a nose, eye or mouth. Gradually the computer learned how to recognise these important elements and overlook the more obvious contrasts between edges or borders, which is the limit of what computers can do at the moment.
Using photographs of a subject taken from multiple points of view, the software automatically picks out important areas within the image, which are cut out as chunks. The chunks are statistically shuffled and a few of them randomly selected and distorted into a 'cubist' composition ready for digital painting, creating a new kind of automated art that was impossible before.
Below are four frames of a source video after image registration. Features have been tracked as a planar group, and each active contour subsequently relaxed to yield a salient feature boundary (shown in yellow). The corresponding Cubist painting is on the right. (Credit: John Collomosse).
The images above and below have been extracted from the chapter 3 of the Ph.D. Thesis of John Collomosse, "Higher Level Techniques for the Artistic Rendering of Images and Video." This chapter is called "Painterly and Cubist-style Rendering using Image Salience" (PDF format, 39 pages, 13.3 MB).
And here is a gallery of images illustrating the application of our rendering algorithm. (Credit: John Collomosse).
The software also can work with animations. The Video Paintbox can transform real video clips into artistically stylized animations with a minimum of manual effort.
When will you be able to take the pictures of yur kid and turn it into a Picasso masterpiece? Not yet.
For now, the software is still in development and will be limited to use by professional animators involved in the research. However, there has been some interest from software and animation companies and the team are keen to take these leads forward.
Source: University of Bath news release, August 25, 2004; and various web pages