ACQUINE (Aesthetic Quality Inference Engine) is a machine-learning based online system of computer-based prediction of aesthetic quality for color natural photographic pictures. (i.e. an image search engine with aesthetic ratings) We believe that this is an important step in computer science research because it shows that computers can learn about and exhibit emotional responses to visual stimilus like humans do. It has been developed at Penn State since about 2005. Dr. Ritendra Datta (now with the Google engineering office in Pittsburgh) was the main developer, working with Prof. Jia Li and Prof. James Z. Wang. The system was placed online for public use in April 2009. This is work-in-progress and hence it undergoes algorithmic changes from time to time, in an effort to improve performance. The work is Patent Pending. If you are interested in licensing this technology from Penn State, please contact James Wang.

What is ACQUINE Designed to Handle?

ACQUINE is designed mainly to assess the aesthetic quality of color professional photographs.

It is NOT designed for computer graphics, artificially-produced diagrams, figures, paintings, composite pictures, or casual family photos. Please do not upload objectionable photos or any photo containing private information that you do not wish to share with others. Full color photos are more suitable than black-and-white photos. The system is not designed to assess very low resolution images. Our recommended resolution is for photo images with 600×600 or more pixels. Acquine is not designed to rate the look or the attractiveness of a person.

As for most computer-based systems, it is possible to find special cases where the system is clearly not functioning as intended. For instance, one may find that a very poor quality photo gets a good score, or an award-winning photo gets a low score.

A rule of thumb is that if the aesthetic quality of a photo is obvious to most people, it may not be worthwhile to seek Acquine’s opinion on it because Acquine may assign funny scores in such cases. There are three ways in which Acquine can be seen in action:

How to Use Acquine to Score Photographs:

Uploading a Photo from your Local Disk

Providing an URL of a Photo on the Web

Show Photos Randomly Sampled from the Photo.net Website

Why are Some Reasonable Photos Rated so Poorly?
Whereas it is not rocket science, automatic assessing photo aesthetics is a new scientific research area and is not simple. We do not expect our friends to always agree with us on visual aesthetics. We should therefore not expect Acquine to always agree with us on visual aesthetics. We are essentially trying to “teach” Acquine about those human emotional reactions to visual stimulus. Aesthetics is an important first step in this direction. It can take time for the Acquine to learn about our world. So please be patient. If your photos are not rated as highly as you would expect, it may not always mean that your photos are not of high aesthetic quality. You are encouraged to upload such photos to sites like photo.net to get human assessments and critiques.

When humans assess the aesthetic quality of photos, score inflation is commonly seen. This is probably because we all want to be more encouraging to others. However, Acquine tries to give you a score that is not inflated. That is, the score actually goes from zero to 100, with 50 being the score of a typical average-quality professional photograph. In another word, a photo with a score of 50 is still a good picture. Acquine is tough!

Source: ACQUINE

This entry was posted on Thursday, May 14th, 2009 at 7:03 pm and is filed under Innovations, Newcomers. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.