Online system rates images by aesthetic quality
May 5, 2009(PhysOrg.com) -- An online photo-rating system developed at Penn State is the first publicly available tool for automatically determining the aesthetic value of an image, according to a Penn State researcher involved with the project.
James Z. Wang, associate professor of information sciences and technology, is one of the principal researchers on the Aesthetic Quality Inference Engine (ACQUINE), a system that judges the aesthetic quality of digital images. Wang said this tool is a significant first step in recognizing human emotional reaction to visual stimulus.
ACQUINE, which has been in development since 2005 and was launched in April 2009, can be found online at http://acquine.alipr.com . Users can upload their own images for rating or test the system by providing a link to any image online. The system provides an aesthetic rating within seconds.
Wang said the system extracts and uses visual aspects such as color saturation, color distribution and photo composition to give any uploaded image a rating from zero to 100. The system learns to associate these aspects with the way humans rate photos based on thousands of previously-rated photographs in online photo-sharing Web sites such as photo.net.
"In its current form, we've seen more than 80 percent consistency between the human and computer ratings," Wang said. "The improvements to the system that are currently under development show promise to get even higher performance.
"Furthermore, aesthetics represents just one dimension of human emotion. Future systems will perhaps strive to capture other emotions that pictures arouse in people," he said.
According to Wang, there also are opportunities to link the rating system directly to cameras so that when a photo is taken, the photographer can instantly see how it might be perceived by the public.
Wang worked with Ritendra Datta, a recent Penn State Ph.D. recipient and Jia Li, associate professor of statistics at Penn State. Funding for this project was provided by the National Science Foundation as part of ongoing research about the relationship between computers and visual concepts. The researchers previously used similar technology to detect authentic Vincent van Gogh paintings.
Provided by Penn State
Feb 04, 2010
I tried it on a horse wang, it scored 17.
Seemed to work like a charm, but then I tried a particularly gross looking cow-pie and it scored a 51.
I tried uploading a few pictures and was disappointed in the results. Try, instead, looking through their random selections and you will surely see pictures that you like which are poorly rated.
Since different people will like different things and give widely different ratings on pictures, by having software analyze the results of a human "poll" grading the quality a set of pictures, you are left with a very confused algorithm which can only predict the AVERAGE rating that a group of humans would give to a picture.
It is the same problem with politicians following opinion polls too closely, or a person placing too much credence in what others think: the result is often chaos and confusion. A person must follow his or her own best judgement, and since this photo rating system cannot (yet) think for itself, it cannot truly make credible selections.
In addition, even if it COULD think, it would often come up with different results than each of us. People seldom agree completely on what is good or bad art.
What I think is important here is that they are improving the ability of software to analyze photos based on prescribed criteria. The Google Similar Pictures experiment could benefit from more research in this field. I have yet to see it present me with pictures that I would consider to be similar to the one that I start out with. I always end up getting more pictures from the same website or with similar captions.
Leonardo 100; Leonardo with graffiti 100; Program 0.