[This post was originally published in November of 2006]
Do you recognize the person depicted in this video? (QuickTime required; the movie is below the fold)
How about this one?
The first video is actually a “chimera,” formed by fusing half-images of two well-known faces together, then animated using 3D projection software. The second video shows us just the top half of another famous person’s head, rendered in the same way. If you’re like most people, you have a much easier time recognizing the face in the second movie than you do either of the faces in the first one. But why?
I can assure you, both of the faces in the first movie are just as famous as the face in the second movie, and you’re being given just as much information about each face: the top half comes from one person, and the bottom half comes from the second person.
These faces were created by Leslie Steede and Graham Hole to try to resolve key questions about how we recognize familiar faces. There is a great deal of research indicating that our memory stores three different types of information about faces: Information about the facial features themselves, semantic information about individuals (President, actor, rock star), and information linking particular faces to the semantic information. Much less is known about the particular ways facial features themselves are stored and accessed.
One thing that is known is that if we see a familiar face in motion, we’re much quicker to recognize it than if we see just a static photo. One explanation of why this is true is called the supplemental information hypothesis: that we actually recognize idiosyncratic movements of individuals. This was confirmed in a 2001 study by H. Hill and A. Johnson, where different actors’ facial expressions and other movements were rendered using a model of the same “average” face. Viewers were able to tell the difference between the actors even though the physical face was the same every time.
A second explanation, the representational enhancement hypothesis, holds that seeing a face in motion enhances the viewer’s ability to recognize static features. However, research supporting this hypothesis has been inconclusive.
Steed and Hole’s work attempted to better understand this second explanation. They created their animations using static images of famous faces including Bill Clinton, Russell Crowe, and Arnold Schwarzenegger. However, the animated portion of the display was generated automatically, so no supplemental information about the identity of any face was provided by the facial movements. If animated faces were recognized better than static ones, then this would offer clear evidence in support of the representational enhancement hypothesis.
In the experiment, volunteers saw 15 male faces (10 were famous and 5 were not) and asked to identify the person if he was famous. The viewers were divided into four groups, according to the type of face they saw (animated chimera, static chimera, animated half-face, static half-face). Then they were distracted for 15 minutes with an unrelated face-recognition task.
The next phase was the key: they viewed 40 static, full faces and were asked to respond as quickly as possible, indicating whether the face was famous or not famous. Ten of the famous faces were the same ones they had seen before — either from the top half of a chimera or a half-face. Another ten famous faces hadn’t been seen before, and the other faces were non-famous. Finally, they were asked to name the 20 famous faces they had seen based on intact full-face photos. Any data from unrecognized faces was discarded.
This is a classic priming study. If the faces viewed in the first phase of the task were recognized, they should make recall of same faces easier later on. Let’s take a look at the recognition rates for these faces:
As you can see, viewers were much better at identifying the half-faces compared to the chimeras, even though they saw the same portion of the face in each case. But perhaps, although the faces weren’t consciously recognized, they still primed participants to recognize the faces later. This chart shows how much faster viewers responded to famous faces they had seen before compared to new famous faces:
For chimeras, priming wasn’t significantly different from zero, while both static and moving half-faces did indeed prime viewers: they were faster to classify the famous faces they had seen previously in half-face form.
But most importantly, animation had no impact on reaction time. This suggests that motion, independent of the idiosyncrasies of a particular individual, does not aid us in recognizing faces.
Steede and Hole do point out that their research doesn’t necessarily mean that the representational enhancement hypothesis should be completely rejected. It’s also possible, for example, that we simply don’t see the chimeras as a familiar face even when we can recognize the component parts.
So who appeared in the two videos above? I’ll let you guess for yourself in the comments.
Steede, L.L, & Hole, G.J. (2006). Repetition priming and recognition of dynamic and static chimeras. Perception, 35, 1367-1382.
Steede, L.L., Hole, G.J. (2006). Repetition priming and recognition of dynamic and static chimeras. Perception, 35(10), 1367-1382. DOI: 10.1068/p5515