Slashdot Mirror
What Makes a Photograph Memorable?
Hugh Pickens writes "Researchers have developed a computer algorithm that can rank images based on memorability. They found that in general, images with people in them are the most memorable, followed by images of human-scale space — such as the produce aisle of a grocery store — and close-ups of objects. Least memorable are natural landscapes. Researchers built a collection of about 10,000 images of all kinds for the study — interior-design photos, nature scenes, streetscapes and others, and human subjects who participated through Amazon's Mechanical Turk program were told to indicate, by pressing a key on their keyboard, when an image appeared that they had already seen. The researchers then used machine-learning techniques to create a computational model that analyzed the images and their memorability as rated by humans by analyzing various statistics — such as color, or the distribution of edges — and correlated them with the image's memorability. 'There has been a lot of work in trying to understand what makes an image interesting, or appealing, or what makes people like a particular image,' says Alexei Efros at Carnegie Mellon University. 'What [the MIT researchers] did was basically approach the problem from a very scientific point of view and say that one thing we can measure is memorability.' Researchers believe the algorithm may be useful (PDF) to graphic designers, photo editors, or anyone trying to decide which of their vacation photos to post on Facebook."
tits
Exposed breasts
Appearance of Forrest Gump is undoubtely a plus.
Achille Talon
Hop!
An unfeasibly large anus.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Please don't tell me that entities selected through Amazon's Mechanical Turk pass for subjects in MIT-level scientific research. Should I start taking MIT less seriously?
What a strange study. From the summary, the researchers sampled a group of individuals by presenting them with random photographs and rating not their _memorability_ but rather recall in asking them to press a key "when an image appeared that they had already seen". This is much different than what I believe makes a photograph memorable--which typically involves some sort of an emotional response to the subject in photograph. For instance, nature pictures taken on a journey to me personally would be very memorable--even though the study suggests otherwise.
If you're in marketing and want people to "recall" your product, yeah sure, this study is relevant. But, it's sort of misleading labeling it memorable as it suggests an emotional response and this study does not address that.
By the way, the definition of "memorable" is the quality of being worth remembering--very different from recall.
Of course things with people and animals (or representations thereof) are more memorable than landscapes. Our minds have evolved to put greater emphasis on things that are a threat or opportunity. Besides, landscapes are generally classified as that only because they're outdoors and don't have any other distinguishing characteristics that would put it into another group.
Knowledge Brings Fear
getting out of a pool, ...
Nonsense. It is widely known amongst the connaisseurs of fine art photography that Ansel Adams is a hack precisely because he never did any supermarket aisles. Seriously, who that guy thought he was? All high and mighty with his view camera and large format b&w negatives... well I got news for you pops! Only pretentious poseurs and high school emo kids and art school losers think b&w is artsy fartsy, to the rest of us is just pretentious self indulging tripe! Get with the program man. Time to get like a Sony Powershot or something, they have amazing colors!
If that photo wasnt isolated, but associated with external ( sound, light, smell, etc) or internal (i.e. feelings, idea associations, complex toughts, etc) things, getting back those things will make easier to get back those photos.
They should consult with more photographers. One thing is obvious: the most-memorable pictures have a central point of focus...something to grab your interest. The least memorable images in the TFA have nothing to grab your attention. That applies to a mixture of subject matter as well as a single subject, such as landscapes.
The TFA gave short shrift to aesthetics, too--where in the photo the central point of focus most favorably may be placed, such as the Rule of Thirds and Golden Sections. These go back to Da Vinci...not new ideas.
An algorithm is of mild interest, yet Van Gogh's Starry Night, a landscape, is far more interesting than his Starry Night Over the Rhone, a painting that includes a man and a woman walking arm-in-arm by the river.
Said that one thing that makes a photo interesting is a pic of something common in one location, that is shown in another location where that thing is not common. There is no way an algorithm could describe that.
I am Slashdot. Are you Slashdot as well?
Other criticism of the study aside, a group of people who might be interested in how well pictures are remembered after short glances are advertisers and marketers.
they fscked that study up real good.
personally I remember a really good sunset, or an awesome view of nature far more than I remember a friggin grocery store. heck there are two that I go into regularly and one I rarely go into and the one I rarely visit I can't find anything in it.
human sized spaces aren't more memorable. what it contains and that means to the viewer is what is rememberable.
Part of it is due to a literal interpretation of the word "memorable". There's a difference between easy to remember, which is what they've measured, and deemed worthy of committing to memory, which is more a more arbitrary consideration. If you see images of a supermarket isle, it might be memorable in that you'll be able to tell you've already seen that picture, but it's not memorable in the sense that most people use the word. It's not that they deem the picture special, it's just that our brains are able to recognize it without effort...you need to remember unimportant things like that in order to navigate places you've been to before, even if these places are completely uninteresting.
My field of work has nothing to do with this, so I may be off in this belief, but I would expect humans who live their lives in less urban settings to have an easier time remembering pictures of natural landscapes. We've trained ourselves to navigate supermarkets, but for most of us, a bunch of trees are just a bunch of trees. Those who actually can navigate the woods without getting lost probably have trained their brain to pay more attention to the differences, and thus should be able to remember if they've seen a picture of a particular natural landscape vs another far more easily.
I address this question in a (to be published) book on the psychology of entertainment where I explore the concept of novelty. Although mere newness is not enough to make something memorable, if something combines a strong design structure or a vivid one, and is both personally and culturally novel, its memorability is greatly increased. When we are young (immature experientially), almost everything is novel and gets consideration as we take in perceptions. Repeated patterns in the environment are assimilated into recognizers, so that we can detect what is unusual and possibly a threat. (Ie. that which is out of place invokes attention, leading to better chance of survival from potential threats.) I believe that the same mechanisms, with varied parameters, then serve multiple purposes including artistic perception. mechanics of reading, and so on. I am engaged in an ongoing effort to embed this principle in hybrid symbolic and neural recognizer systems, as part of a larger effort. Anyway, I leave the take-away point that memorability is a function of both perceptual system operation and interpretive deep systems drawing on culturebases, hence novelty and memorability is dependent on individual (per person) frameworks.
Come on, you know that this was done as "pure science," and that the justifications were tacked on at a later time. I worked in a research lab over a summer studying machine learning as applied to conversational branching in natural language processing. The specific problem I was working on had to do with picking cost-evaluation algorithms which would allow for a fully expressed Markov tree, while minimizing the solution space - so, almost nothing to do with NLP at all. You can bet that when someone who knew nothing about my research asked what I did, I came up with some lame "real world" application, instead of taking a few hours to discuss the nuances of machine learning, and why it was terribly interesting that a Monte Carlo algorithm could eventually stumble upon a solution in very specific circumstances where Q-Learning never would.
Any of them ironically labled..
"Its not my weiner."
http://www.jsonline.com/blogs/news/123109063.html
-Hack
Got Geometrodynamics? Awe, too hard to figure out? Too bad.
As part of some personal projects for learning 3D modelling using Blender, I tried reconstructing the interiors of some places I had visited. Could easily remember those areas where I had walked through or sat down in, but anywhere more than three or more metres away, I couldn't remember.
I guess if you were walking through a jungle trail, remembering the junctions on that trail would be more important that the surrounding vegetation.
Though, many supermarket customers get really annoyed when the supermarket decides to rearrange the layout and suddenly everything has been reshuffled around. Sometimes the store managers have to send out Sherpa guides to help shoppers find what they are looking for.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Boobies! And no, not the jiggly kind. I mean the bird, you perverts.
Enlightenment is a pipe dream. So where's the pipe?
Oddly, these Internet memes don't seem that bad, with the mellowing of time. Here is what I think of them now...
tubgirl: blow up dolls are funny
lemonparty: reminds me of Barney Frank
goatse: what is the medical term for that?
Eh, YMMV.
This issue is a bit more complicated than you think.
caused a buffer overflow
Table-ized A.I.
You have never seen one of these?
http://www.anseladams.com/category_s/4.htm
memorability Looking at the images, there's definitely familiar objects helping recognizability, but I saw a much more computationally-easy pattern. Look at the edge detection results of the images. Now count the internally terminated edges (of some length at-least 5px on the thumbs) vs the "lines" that extend to the end of the image.
Memorability = (internally-terminated edges) / (edges touching border)
At a glance, this appears much more likely. Why? I think people put more importance on fully-framed "objects" as life demands a high ability to detect objects and we do so with corners. The more unterminated edges, the less we can trust that we see "objects" rather than just data, so it's thrown-out as noise.
The formula should be easy to test: 1. Edge detect 2. "Walk" edges: remember ones that don't split, end, or considerably change direction (circles ok) in 5px (appropriately scaled) 3. Of all the #2 found edges, mark those that touch edges 4. Memorability = [ #2 (all) - #3 ] / #3
Let me know how it goes...
Science & open-source build trust from peer review. Learn systems you can trust.