Page 1 of 1

Stereo window placement

Posted: Sun Jun 16, 2019 1:59 pm
by stereomaton
I just want to introduce important notion in stereo photography (it applies to video too) for enthusiasts who come to stereo through stereopi:
the stereo window.

In traditional photography, there is a frame in which all elements are included.
While framing, among other things, it is important to pay attention to what is cut by the edges.
The 3D equivalent is the stereo window though which we see all the elements.
Being aware of the edges is tremendously more important in this context (we will see why thereafter).

The edges of your picture create a delimitation beyond which no pixels are visible. Typically, this is a rectangular shape. But because your content has depth, this rectangle get a position in the third dimension too (or more precisely, the content has a depth relative to this rectangle). And because man cannot see on the sides (no pixels) it acts like a window we see the scene through. That's why it is named stereo window.

As stereo(photo/video)graphers, we have some freedom to choose were the window is placed in our scene, and we can also make some elements go through the window (out of the screen). There are essentially two limits:
1) There shall not be too much disparity
This means that no element in the photograph shall be more than a certain distance apart (i.e. not too far from where the same element would be if the two photographs were superposed). Actually, this distance depends a lot of geometric factors such as the size of the screen/print and the distance you watch it from. For the spectator, it shall not exceed 6cm apparent (perspective into account) size. Note that this distance is far smaller for objects which pop out of screen.
This constraint is easily achieved as soon as the closer element is not too close, say about 1.5m-2m with the base (distance between cameras) of 65mm of the stereopi. For closer objects, you might have to reduce the distance between the cameras (for this reason and some other factors too).
2) There shall not be front element cut by the edge
This one is really important. The brain does not like it at all, although people react differently, because it does not correspond to a real-world situation: when an object is in front of the window, it cannot be masked by it. This is called window violation.
So beware to not cut anything close with your frame (note that you can choose window placement in depth, which changes what elements are close).
This stereo picture from Randy Hester shows it well: black and yellow cuts are okay, white is bad. Also, the yellow one creates a pop-out effect.

To place the stereo window, we just have to [beware of the resulting edges]:
1) choose a point in the 3d scene which will be exactly at the screen plane
2) align the two images on top of each other so that this point matches
3) cut every parts of the images that do not overlap
And of course, merge the two resulting images with the format of your choice.
To avoid windows violations, the easiest way is to choose the closer point of the photograph; or at least the closer which is next to the edges (because closer point in the middle can pop out often safely).

Note that because of the way the cameras are placed and their own field of view, you always have to cut the edges of raw images from the sensors anyway, at least if you do not want quite bad stereo images to be shown.

A last note about pop-out effect (a.k.a. through-the-window/TTW effect)...
The elements that pop out cannot be placed anywhere. Actually, the 3D space of a stereo photography is like a truncated pyramid (we can see larger background than the window), so the popping elements shall not be too close to the edges or you might get a window violation.

Edit: fix one typo that changed meaning + add precision about closer point choice.

Re: Stereo window placement

Posted: Sun Jun 16, 2019 2:06 pm
by stereomaton
I hope I was clear. If it is not the case, feel free to ask. I phrased it the way I think of it the more naturally, but I am familiar enough with this notion to rephrase it in multiple ways if you need more explanations.

Re: Stereo window placement

Posted: Wed Jun 26, 2019 11:13 am
by Realizator
Stereomaton, I think I need to add this information to my next article about stereoscopic photo, thank you!
You see, actually we have a lot of information in this scope. May be we need to do something like "beginners course" with a lot of small step-by step articles, describing each feature and suggesting some simple experiments with the StereoPi? I'm thinking about this course structure now.

Re: Stereo window placement

Posted: Sat Jun 29, 2019 11:07 am
by stereomaton
I practice stereophotography for a few years now, and window violation is a very common mistake for beginners. Actually, the brain accommodate somehow, but the same photograph with stereo window placed to avoid those violations make it far far better.
The second most common mistake is possibly (mis)alignment.

In my personal learning journey, I started to deeply understand the hurt of window violation when I displayed my photograph on 3D screen which emphasize the phenomenon, mainly because the image is anchored to a physical object of the real world. With freeview for example, the brain has a tendency to adjust the stereo window placement by itself ignoring the resulting artifacts, and can do it without being disturbed much by other elements.

So yes, it deserves totally a place in your article. Maybe, some little drawings and examples to illustrate can help to understand better. You can ask for review if you wish.
Realizator wrote:
Wed Jun 26, 2019 11:13 am
You see, actually we have a lot of information in this scope.
Sorry, I am not sure to understand what you mean by this. Which scope do you talk about... stereography or window violation specifically ?

Concerning step-by-step mini-tutos, this is always a good way to introduce beginners into a new product, but it requires a lot of efforts. It is a wonderful way to show lots of possibilities of the product too, in a concrete manner. Also after the initial effort, users tend to be inspired and start to propose new applications so it might be worth it.
A difficulty is to choose which experiments might be described, but it can be a discussion in another thread.

Re: Stereo window placement

Posted: Tue Jul 02, 2019 2:37 pm
by stereomaton
To illustrate, let's work on this image:


This is the kind of images a stereopi can output, except that I pre-aligned the two images so that everyone can run its own tests on this pair if needed with immediate results.

Here is a quick-and-dirty sketch of the scene, seen from above:


We can recognize the two main roses in pink, the pole in gray-purpleish, the leaves in grayish-geen, the fence in vivid-green and two yellowish flowers.
The black drawings represent the two cameras and their respective field of view.

First, wee see two zones in the images:


The green one contains elements visible on the two images and are thus good candidates for stereoscopic 3D.
The orange one is composed of useless data for stereoscopy.
Notice that the tip of the green triangle is valid but there might be high discomfort if something is there.

Let's place the window (thick blue) in front of the scene so that everything is behind the screen/paper (screen plane in thin blue):



We can also place it a bit more deeper so that a part of the elements are in front of the screen (pop-out)



There is no window violation because in these examples, all the elements that are before the window are lying in the green zone.
If we had set the window at pole depth, a part of the content in front of it would have been in the red zone (e.g. the leaves and fence on the left) thus we would have had a window violation.

Note that you might need to watch these images on a 3D screen or in anaglyph to better understand the difference.

I hope these illustrations help to understand the concept a bit further.
In real-world photo, I would have cropped a little more to get better composition.