Perception II

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Sep 17, 2007

Slides

Contents

[edit] Readings

Same readings as for previous lecture. You are still responsible for making a comment on this lecture.

  • Perception in visualization. Healey. (html)
  • Graphical perception. Cleveland & McGill. (jstor)
  • Chapter 3: Layering and Separation, In Envisioning Information. Tufte.

Optional Readings

  • Gestalt and composition. In Course #13, SIGGRAPH 2002. Durand. (1-up pdf) (6-up pdf)
  • The psychophysics of sensory function. Stevens. (pdf)

[edit] Mcd - Sep 17, 2007 03:28:54 pm

Reading the Healey paper, I had some misgivings about Treisman's feature integration theory, which were ultimately reinforced later in the paper when I encountered similarity theory. It seemed to me that preattentiveness was not as strongly connected to individual dimensions themselves, but rather that deviance is preattentive.

Healey gives the following example, which I suspect isn't entirely true: "For example, a sloped line in a sea of vertical lines can be detected preattentively. However, a vertical line in a sea of sloped lines cannot be detected preattentively." In the former case, the vertical lines all share a common orientation, and the sloped line stands out. In the latter, if all the sloped lines are similar, i.e. sharing the same slope as below, where N-N similarity is high, then I would suspect the presence of a vertical line would be preattentive.
image:example.png
It's not that slope is asymmetrically preattentive, it's that a vertical line in a sea of randomly sloped lines (low N-N similarity) is not the same condition as a sloped line in a sea of vertical lines (high N-N similarity).

Looking at this example, I'm not totally convinced my suspicions are accurate, and the color-shape comparison this morning was convincing of a hierarchy of preattentiveness but similarity theory seems closer to what is actually going on that the notion that some features are inherently more or less preattentive than others.

[edit] Charlotte Wickham - Sep 17, 2007 08:47:20 pm

I was intrigued by the very brief discussion in class about the design of subway maps. A recent redesign (http://www.kickmap.com/) of the NY transport system has fueled some discussion on the web. Here is an interview with the designer of the current map and his reasoning behind his design choices http://gothamist.com/2007/08/03/michael_hertz_d.php.

The current NY subway map uses smooth curves to represent the routes. The new design uses horizontal, vertical and 45 degree sloped lines. I find the redesign aesthetically more pleasing. Of course with this type of visualization the real test is using it to navigate the transport system. Unfortunately I have never been to NY so I can't comment on the current map.

I've had no problem navigating the London system - the maps of the underground there are very geometrical and have no reference to above ground geography so an additional street map is necessary for navigation.

Once on the subway the important information is where and when to get off and where you can transfer to different routes. I find the addition of accurate street level information confuses the extraction of this vital information. It does however mean new users of the subway also need a street level map with the stations marked on it.

[edit] Mark Howison - Sep 18, 2007 05:45:03 pm

The paper I mentioned in class yesterday on ambiguous figures is:

Tsal, Y., & Kolbert, L. (1985). Disambiguating ambiguous figures by selective attention. Quarterly Journal of Experimental Psychology, 37A, 25-37.

"The present study was designed to investigate the hypothesis that the formation of a given percent of an ambiguous figure results from focusing attention on a focal area that contains features significant for this percept but not for the alternate one..."

Basically, the experiment they ran was with an ambiguous figure such as this one:

Image: Duck_rabbit.jpg

When participants' attention was drawn to the right side of the image, by having them identify a letter in that region of the screen before showing the figure, they were more likely to see the figure as a rabbit, and vice-versa.

In fact, if you just look at this figure, it is hard to see it as a rabbit without moving your attention toward the right side. Moving your attention to the left side almost immediately triggers the duck percept.

[edit] Robin Held - Sep 18, 2007 08:38:08 pm

In class it was asked why pie charts remain in common use, despite the clear evidence that they are not as effective as bar graphs. I think the main reasons are: 1) Very few people critically assess their own graphics 2) Pie charts are so prevalent that people don't question their quality. At first review, pie charts do seem like a logical choice to represent the respective portions of some whole occupied by various constituents. However, as Cleveland and McGill demonstrated, they cannot be used to reliably compare the sizes of those constituents. One needs very large differences in the portions to gain a rough idea of their relative sizes. Therefore, pie charts defeat their intended purpose and the data is not conveyed to the reader. Nonetheless, pie charts are still a default mode of visualization, so many authors will likely continue to use them for a long time.

[edit] Hazel Onsrud - Sep 18, 2007 09:16:27 pm

The conversation in class about the subway map prompted me to think of a couple of other examples of diagrams, that contrary to the perception guidelines proposed, see to use straight lines, often with drastic corners. For example, circuit diagrams are often drawn with a mess of straight lines...I attempted to draw a few with curved lines and found as the complexity of the diagram grew, the more I appreciated the straight line norm. Thus, I think the clarity of the overall concept and details here, like for the subway map, again outweighs our perception tendencies.

On a totally different note, this class also reminded me of a board game I played as a child...the only link I could find to it was here: http://www.lemniscateinc.com/board.html, but it was neat because you had to concentrate your attention on what the other players perceived...and from what I remember, individual folks seemed to stress different components of each visual cue, and coming up with a pattern for what types of things they saw was what helped you win the game.

[edit] James O'Shea - Sep 19, 2007 07:42:36 pm

I think the Healy paper provided a good overview of the field and I enjoyed the accompanying images and applets. The Change-Blindness examples seem to have the desired effect on people, and I was wondering how they generated them. Do they just guess which parts to omit and try them out? The effect must have something to do with the visual saliency of the the image in different areas, and it seems like one could experimentally derive change-blindness examples by automatically testing different parts of the image. Visual saliency is an important measurement for many algorithms in computer vision and graphics, and there are many different energy functions used to calculate it. It would be interesting to test some of these functions on the change-blindness images to see how well they predict whether the difference between a pair of images will be detected.

[edit] Ken-ichi - Sep 19, 2007 09:20:34 am

I'm having trouble understanding gestalt. I've encountered this word in entomology, where there are a dizzying array of species and identifying them is usually a matter of counting body parts, and comparing shapes and sizes, and walking through lengthy dichotomous decision trees. People with a lot of experience, though, are often able to look at an insect and call out the family (or the species) without even looking through a dissecting scope, and they sometimes claim they were able to make the ID "by gestalt," i.e. they just know it when they see it. I can do this with some birds, where I'll just see something flutter away out of the corner of my eye and immediately know what it is, through a combination of color, size, movement, sound, and context.

However, Monday's lecture and the optional readings haven't clearly defined for me what gestalt is. Is it even a noun? What exactly is meant by the sentence (from the optional presentation on gestalt) 'The shape with the best “Pragnanz” is the figure'? Is "Pragnanz" a state in which lines continue, objects obey consistent grouping, and figure/ground distinctions are clear?

[edit] James Andrews - Sep 19, 2007 12:52:59 am

Hazel -- good point about circuit diagrams. They tend to disambiguate crossing points by marking 'actual' joins with a solid dot and by making a little U shape in the line when one wire crosses 'over' another; I imagine once the ambiguity at crossings is sorted out so clearly, the straight lines become more readable. That style seems like it might be the best way to draw any relatively large network of connections, actually.

Ken-ichi: Here's a page that seems nice and has definitions: [1]

From that page: "Gestalt is the German word for "form," and as it applied in gestalt psychology it means "unified whole" or " configuration." The essential point of gestalt is that in perception the whole is different from the sum of its parts."

... so when you recognize a bird or a bug, it's not by breaking down individual components explicitly; rather your mind seems to recognize the whole all at once.

And Pragnanz apparently means "good form," where "good means symmetrical, simple, and regular."

[edit] Amanda Alvarez - Sep 19, 2007 08:41:20 pm

Prägnanz = conciseness.

I wonder whether anyone has experimentally evaluated whether images with good continuity, grouping and Prägnanz actually have higher information content. To me, this would be the step needed to take Gestalt theory from being just a description to actually being an explanation; we like nicely continuous lines because they are more informative (more informative because they agree with all the natural regularities we have witnessed before?). There seems to be more information introduced through layering (discussed by Tufte, eg. Kanisza triangles, or 1+1=3). Then again, knowing that we are sensitive to perceptual differences (and that it is differences or changes which are often most informative) leads me to question whether the predictability of Gestalt principles could possibly be informative. Considering how much Tufte likes his data-ink though, I'm betting there has to be some crazy link between Prägnanz and information.

[edit] Christopher Volz - Sep 19, 2007 08:53:25 pm

Healy's paper got me thinking about a couple of things, namely that there seems to be a hierarchy of perception and that you are very limited in the amount of information you can convey at the "pre-attentive" layer and that, of course, certain demarcations are going to take perceptual precedence over others. I expect this is useful if you need to convey information at a glance, but if you're making a graphic that is intended to be studied, it's probably just good information to know.

I was also reminded how our attention can often get the better of us, while we focus on one thing or action we can often completely miss others; that is while counting trees we don't realize we're lost in the forest (if I may bastardize the expression). Which brings me to the winner of the 2004 Ig Noble prize in Psychology -- the invisible gorilla on the basketball court

[edit] Kenrick Kin - Sep 22, 2007 11:01:41 pm

Here's something I found much more difficult than the examples we did in class, hehe. The picture gradually changes and you have to find the difference.

http://newyorkdesign.hp.infoseek.co.jp/menu06.html

[edit] Athulan - Sep 23, 2007 06:37:41 pm

I don't see all the features Healey lists in Table 1 in his paper as being preattentive. I think they depend a lot on the relative difference between the unique feature amid the rest of the features. For example, Maneesh put an example of curvature in the lecture and Healey has one in the table, I felt that the lecture example was much more preattentive than the Healey example as the lines in the former were much thinner compared to those in the latter. This made the curved feature stand out more prominently, while in Healey's the curved feature did not stand out prominently at all.

Similarly, I felt that "closure" would depend on a lot on how closed the other curves are - this would work only if the other curves were relatively open. I think its difficult to denote preattentiveness in an absolute sense the way Healey does in his examples. Of course, this makes the it harder to define as it will be a function of multiple variables, and these variables will change depending on which preattentive feature is trying to he highlighted.

[edit] David Jacobs - Sep 23, 2007 08:52:22 pm

Kenrick: After about 15 frustrating minutes, I can safely say that gradually changing scenes are much harder to detect changes in. I've been trying to figure out what exactly it is that makes it so much more difficult. On the one hand, since the change is gradual, there's nothing that "pops" when the scene changes. On the other, since the change in this case is a simple change in opacity of the changed region, it's much easier to spot the change on a static image when it's halfway through. I think the reason it works so well lies in the way we look at these pictures. I find myself getting bored looking at the overall picture, and eventually examining small regions in great detail, searching for anything that stands out. The problem is that I only stand a chance of finding the changing region if I happen upon not only the right place, but at the right time.

[edit] Daisy Wang - Sep 24, 2007 08:56:11 am

Human perception is important for creating visualization, because the effectiveness of a visualization to get-across some idea, information or point of view, is dependent on how people perceive the visualization. It seems to be obvious, but I never have expected so much difference in human perceiving different dimensions in a visualization. However, after some try-outs in class, I am convinced. I think it is this realization that different visualization will account for its effectiveness should be passed on to every graph makers. And a visulaization tool should on the one hand being able to suggest good ways for visualization, but on the other hand, also provide reasons for choosing one over the other. Because, besides the rules of human perception, each designer has their own emphasis and goals.

This will give a graph designer better flexibility. This will also transfer the weight on the software to decide the de-facto visualization to represent the data, to the designer themselves to design effective visualization, taking into account both general rules of human perception and their own goals in representing the visualization.

[edit] Karen Hsu - Sep 29, 2007 10:47:02 pm

Although not preattentive, our discussion of the topic somehow reminded me of Dinosaur Comics, a webcomic whose creators reuse the same artwork and change only the dialogue.

Here is one that I found amusing and somewhat appropriate: http://www.qwantz.com/fanart/dinosaur%20comics%20perception%201.PNG

[edit] N8agrin - Sep 30, 2007 10:45:29 pm

Cleveland and McGill's study provided a number of interesting findings about how capable the visual system is at measuring various visual properties. As we discussed in class, one of the most significant findings was that many of the testers had quite a bit of difficulty in distinguishing between angles, suggesting that pie charts are an inefficient method for displaying comparative data. I was somewhat confused about their discussion of the ability to perceive outlined rectangles, as in Figure 12 of their paper. It's unclear to me if in the second plot, the white areas are a visual cue to the overall area of the bars, or if they are meant to represent their own values. If I read the paper correctly, I believe they claim that outlines of rectangles tend to make the encoded values easier to extrapolate, however I would argue that in this visualization it's still difficult to compare A and B to one another.

Tufte discusses layering and perception of positive and negative spaces, as well as visual phenomena. Particularly noteworthy to me were his examples of negative space and how they eye can perceive the missing shape as being in front of other images in the visualization. In particular are his suggestions regarding the use of negative space which might create a recursive pattern or moire vibrations, both visually distracting and potentially unintelligible.



[add comment]
Personal tools