Correlating Low-Level Image Statistics with Users’ Rapid Aesthetic and Affective Judgments of Web Pages

Summary:
In this article, the authors were concerned with how well websites evaluated based on decomposed low-level images compared to what actual users thought about the pages themselves. Users were asked to base their decisions on four main design dimensions:
Attractiveness, Pragmatic Quality, Hedonic Quality: Identity, and Hedonic Quality: Stimulation (hedonic quality refers to how pleasurable and interesting someone thinks the page is). The users rated the thirty web pages used on a seven point scale along the gambit of each dimension.

The same set of pages were also analyzed by a very structured computer algorithm. Simply put, the algorithm attempts to break the page into pieces that contain an equal amount of information. These quadrants of the image are then measure for balance, symmetry, equilibrium (is the image lopsided or centered), and the total number of quadrants created (an example of the breakdown is shown to the right). The authors used some pretty advanced math, so if you want to check it out for yourself I recommend you read the last page of the article, which is covered in algorithms.

Overall, the algorithm's computation of a page's awesomeness had a strong correlation with what the test users thought as well. Below, you can see how the different measures of attractiveness match up between human and machine. The authors stress the impact of these results- a computer can be used to predict how visually appealing a design is to end-users. This is both a more cost-effective evaluation of a design and a breakthrough in the understanding of how humans make decisions about visuals. Because the testers were only allowed to view the actual page for 150 milliseconds, the results prove that humans make low-level decisions in a similar process as the authors' algorithm does, just in a much more effective way.

Discussion:

This paper killed me. With a title like that, I was not looking forward to reading heaps of procedure and math. However, in the end it turned out to be pretty dang interesting. I wasn't aware that any research was being conducted as to simulating the likes and dislikes of humans based on images and designs. If this work were expanded, then computer programs could screen TV shows, commercials, movies, websites, and photographs to determine if humans would actually find them appealing before launching them publicly or to testers. That's a little bit creepy and a little cool.