Anaglyph (Red-Cyan 3D Pictures) / Discuss

ByteMyzer    says:

In Photoshop, to apply the Optimized Anaglyph method described in the article to which you refer, you would need to do the following:

Channel Mixer - Left Chip:
* Red Output Channel: Red - 0%, Green - 70%, Blue - 30%
* Green Output Channel: Red - 0%, Green - 0%, Blue - 0%
* Blue Output Channel: Red - 0%, Green - 0%, Blue - 0%

Channel Mixer - Right Chip:
* Red Output Channel: Red - 0%, Green - 0%, Blue - 0%
* Green Output Channel: Red - 0%, Green - 100%, Blue - 0%
* Blue Output Channel: Red - 0%, Green - 0%, Blue - 100%

Paste the Left Red Channel into the Right Red Channel, then:

Levels Adjustment (Red Channel):
* Input: 0 | 1.5 | 255
* Output: 0 | 255
Originally posted 47 months ago. (permalink)
ByteMyzer edited this topic 47 months ago.

SeanGleeson says:

I went and tried it on an image that had particularly bad retinal rivalry because of bold primary colors. The Optimized Anaglyph certainly cured the retinal rivalry, and made the image easier to fuse. But of course the colors underwent a big change too.
Posted 47 months ago. (permalink)

Jeff Cushner    says:

I have to believe that curing the retinal rivalry has to change the image colors unless we use different viewing filter coloration or crossed-eye or flicker techniques.
Posted 47 months ago. (permalink)

crookedtool says:

"Paste the Left Red Channel into the Right Red Channel"

Please explain this step further. Thanks again!

Jeff: With color filters over your eyes, you can't see the true image colors anyway. For me, curing retinal rivalry produces a much more solid and convincing 3d image, and I feel that's worth much more than a large color depth, as far as anaglyphs are concerned.
Originally posted 47 months ago. (permalink)
crookedtool edited this topic 47 months ago.

DanD3Dman    says:

Hi Matthew, thanks for the process steps.
Crooked had a good question and I want to ask about
"Levels Adjustment (Red Channel):"
Which chip does this refer to, both?
Posted 47 months ago. (permalink)

ByteMyzer    says:

crookedtool420:

After you channel mix both chips, copy the Left Chip, select the Red Channel of the Right Chip, and paste. (Consult the help file for your particular version of Photoshop)


DanD3Dman:

This refers to the anaglyph after you've combined the chips. Go to Image > Adjustments > Levels:

* Channel: Red
* Input Levels: 0 | 1.5 | 255
* Output Levels: 0 | 255
Posted 47 months ago. (permalink)

crookedtool says:

ByteMyzer: I appreciate all the help. Just one more question, though. I'm sorry, I'm new to this, but what's a chip?
Posted 47 months ago. (permalink)

ByteMyzer    says:

The Left and Right images, or "chips" represent the Left and Right eye views of a subject, respectively.
Posted 47 months ago. (permalink)

Louis Carlsson 3D    says:

hey, look at all these players..

The channelmixer is the key allright. This seems a bit extreme though. I allways mix the red with the blue in the left image to reduce rivals. I never ever mess around with the green channel because I want to fix the problem where it is, not create new problems in the proces. I use Hue/Saturation in the right image to control the colors. This is ofcourse what I do, it does not mean it is the best way. I´ve uploaded my version of Seans 2D>3D conversion as evidence material here.. www.flickr.com/photos/42224310@N00/2680073501/

So now there is the original plain anaglyph, the optimized version and my compromize aiming to get a little bit of both. Please help us by telling us in which direction you would like to see your 1st choice of anaglyph go.

If you really want some fun, then join in by making your version of the challenge. I´m sure Sean will provide you with the chips too.
Posted 47 months ago. (permalink)

SeanGleeson says:

Yes, absolutely I will. Anyone who wants to try to optimize the Club Penguin pic for comparison, just send me a message and I will send you the left and right images.
Posted 47 months ago. (permalink)

SeanGleeson says:

All of the contenders are posted here, and also here on my site, for easy comparison.
Posted 47 months ago. (permalink)

Mike Bittner    says:

This topic pretty much explains why I use AnaBuilder if I am going to post anaglyphs online. Once you have modified the config files for speed on your computer, the program works quite well for a java application. There is a lot that you can do in the filters area. There are numerous presets for strong colors in your images that work very well. Perhaps I'm a bit lazy not wanting to use Photoshop or PSP (okay, not perhaps ... usually), but I don't see why you would want to reinvent the wheel when someone else has put so much time and effort into the cause.
Posted 47 months ago. (permalink)

Louis Carlsson 3D    says:

Okay Mike, let´s see that Anabuilder version please.
Posted 47 months ago. (permalink)

Mike Bittner    says:

Okay ... AnaBuilder and Cosima anaglyphs should be posted soon. Even I was surprised by the results.
Posted 47 months ago. (permalink)

e.dubois    says:

Here is my shot with the least-squares algorithm.
www.flickr.com/photos/e_dubois/2688787053/
Ghosting is rather strong.
Posted 47 months ago. (permalink)

Dan (aka firrs)    says:

Yes...if you take that last one and split the RGB channels you can see quite a lot of "left eye" in the "right eye" and vice versa - much more than you get caused by jpeg compression. Guess that is a result of the strong colours in the orginal?
Originally posted 47 months ago. (permalink)
Dan (aka firrs) edited this topic 47 months ago.

SeanGleeson says:

There, I have all nine anaglyphs by Mike Bittner on the page now. Some of them are very good, I just might become a convert to AnaBuilder myself.

I do find it strange that the AnaBuilder "Dubois" anaglyph is not identical to Eric Dubois's anaglyph. Are there variations of the Dubois algorithm?
Posted 47 months ago. (permalink)

e.dubois    says:

I have a couple of comments on the last point.
1. The least-square algorithm that I developed uses the spectral transmission curves of the glasses and the spectral density of the RGB display primaries. The version I use is for red-cyan glasses, specifically ones from American Paper Optics, and for a CRT display. Thus there is in principle a different transformation for each glasses - display pair. I did notice that ghosting was less visible on a CRT display than on an LCD display for my image. I have not implemented the algorithm for LCD spectral responses yet.
2. I have no idea how the algorithm is implemented in AnaBuilder. In my own implementation, I assume that the left and right images are gamma corrected, so I apply gamma, do the transformation in the linear domain, then apply gamma correction to the final anaglyph. If AnaBuilder does not do that, it might explain the slightly different tone scale. My own implementation uses MATLAB and is not portable.
Posted 47 months ago. (permalink)

jaysdesk    says:

Here's my shot at optimization:
farm4.static.flickr.com/3103/2699328439_b556ee104f_o.jpg
I desperately tried to preserve some red in this anaglyph.
In Photoshop CS3, I desaturated the blues and cyans by about 50% in the left chip, and the reds by about the same amount in the right chip. I did switch to Lab mode before the adjustments.
Originally posted 47 months ago. (permalink)
jaysdesk edited this topic 47 months ago.

SeanGleeson says:

I posted your adjustment to the page. Good job of preserving the red, and the colors generally.
Posted 47 months ago. (permalink)

Dan (aka firrs)    says:

Here is my "secret sauce" method cooked up really quickly, but specifically for this image.

farm4.static.flickr.com/3177/2701629835_a163386a0e_o.png
Posted 47 months ago. (permalink)

SeanGleeson says:

Dan, it looks good. But could you give at least a general description of the steps you took to get it?
Posted 47 months ago. (permalink)

Dan (aka firrs)    says:

Not easily but it involved changing the shape of the R, G and B histograms of the L and R views and then using using the saturation blend mode to overlay the modified image over the original. The intended effect was to clip the highest R, G and B values, muddy the purest R, G and B colous, and yet leave the the rest of the colours more or less intact. It was literally five minutes of tinkering while wearing the glasses until I got an image that had tolerable retinal rivalry. Then I made the anaglyph.

No idea if any of that is logical, but I think it worked nevertheless.
Originally posted 47 months ago. (permalink)
Dan (aka firrs) edited this topic 47 months ago.

jaysdesk    says:

I uploaded yet another version of club_Penguin that I believe compares favorably to my earlier posting, with a difference that this one is stupefyingly easy to do and describe. Best of all it can be done in Photoshop Elements.
Here is the Link : www.flickr.com/photos/fromjaysdesk/2742210997/
Originally posted 46 months ago. (permalink)
jaysdesk edited this topic 46 months ago.

3dbeadyeyes2    says:

Just tried on an image making the left chip black and white, and the right chip colour to see what happened, I was ammazed to see that it gave a similar effect to the optimization as at the top of this article.......
Originally posted 44 months ago. (permalink)
3dbeadyeyes2 edited this topic 44 months ago.

ByteMyzer    says:

3dbeadyeyes2,

That's why the principle behind the Half-Color Anaglyph works.

Because the Red filter only passes one color, the most that the viewer can perceive about the image is its brightness, not its color. Therefore you might as well grayscale the Left chip before rendering the anaglyph.

This by itself does not accommodate all colors. You may still encounter extreme retinal rivalry for purely Red color components because, unaltered, the Right chip contains no Red information in the Green or Blue channels. Thus, any Red objects will appear extremely dark through the Cyan filter in contrast to the Red filter view.
Posted 44 months ago. (permalink)

santisan says:

hi, any one has advise for printing anaglyphs?
Posted 44 months ago. (permalink)

ByteMyzer    says:

Hello, santisan,

You're more likely to get attention from your question as a separate question if you post it as a NEW topic, not as a reply to this one.
Posted 44 months ago. (permalink)

santisan says:

thanks ByteMyzer
Posted 44 months ago. (permalink)

D.Broberg    says:

Hello,
I'm still a 3D-Noob and just produced my first anaglyph here:
http://www.flickr.com/photos/dbroberg/3218707563/sizes/l/
While I understand how all this color and filter optimization can be helpful, I always considered anaglyph to be a very sloppy process by default since the cheap glasses vary so widely in quality and the monitors and printing processes also vary widely in colorimetry.

Some of the images I've seen look great with one pair of glasses but have a lot of cross-talk when another pair are used. All the glasses I have are described as Red-Cyan but are they really?

Are there some standards for the glasses and or the image processing that describe an accurate anaglyph transformation?
Posted 41 months ago. (permalink)

jaysdesk    says:

Roughly speaking, optimization for colored anaglyphs, define a triangle, with ghosting, retina rivalry, and problematic colors at the corners. One can only aim at one point within this triangle, regardless of standardizations and calibrations.
This said, the Danes, Louis Carlsson, and B. Allan Andersen have come closest to optimum, in some of their anaglyphs, I suspect by targeting individual colors in Photoshop. This requires a skill and time expenditure that is probably beyond the reach of most of us.
I have heard the advice that to make a decent color anaglyph, certain situations should be avoided, such as bold colors, large tone scales, abrupt tonal change........ Not a solution really, but an informed conformity to the limitations of the format.
Originally posted 41 months ago. (permalink)
jaysdesk edited this topic 41 months ago.

Dan (aka firrs)    says:

Dan (aka firrs)    says:

On the CRT monitor I am currently using fig 1 looks quite like fig 2 through the red filter. But that's not because my red filter is passing blue or green light...its because the monitor is emitting some red light when it displays the cyan colour. I'm able to tell this is the case by viewing through a diffraction grating spectroscope. If, for example, I set the monitor to display pure RBG green I can see through the spectroscope that the green colour contains a small amount of blue (frequencies at the green end) and a fair amount of red (across almost all the red frequencies). If I put the cyan filter of the glasses between the monitor and the spectroscope that cuts out all the red out of the displayed green. The conclusion is that the glasses are very effective colour filters...the ghosting problem lies with the monitor. It's not a fancy monitor but I did calibrate the colour, contrast and brightness against a printer's standard.

Dan (aka firrs)    says:

I was very careful to say nothing general - just about the glasses and monitor combination I was using at the time.

The spectoscope does not lie ...and there is nothing in that link that contradicts what I said - it fully supports it. The reason the red filter of my glasses shows some lightness of the cyan square in your figure is because the light contains red wavelengths - not because the filter passes (visible amounts of) blue or green wavelengths.

The filters of my glasses do indeed pass light at all wavelengths but only when the light is very bright - brighter than the light from a normal monitor.
Originally posted 41 months ago. (permalink)
Dan (aka firrs) edited this topic 41 months ago.

Dan (aka firrs)    says:

I think I've just worked out what you've done ByteMyzer.

In figure 2 you coloured the box corresponding to the cyan box in figure 1:
Red 1
Green 50
Blue 20

...and then equated that with the red filter passing 50/255=20% Green light and 20/255=8% Blue light. The reality is more complicated than that because monitors do not create 256 uniformly placed brightness levels of 3 discrete wavelengths.



Another thing you can try to check your glasses is to overlay the red and cyan filters so that light has to pass through both. Any light that does (and of course some will) is being passed through one filter in error and one filter correctly. It is also being attenuated a certain amount by the filter it is supposed to pass through.

I can see my monitor through this combined filter of my glasses - but only just and my eyes have to adjust for the darkness. The passing light looks purple. By shining a bright white light through and viewing on the spectroscope I can tell that the cyan filter is passing frequencies from yellow higher (ie all the red) and some of the very low visible frequencies (ie the far blue end). I have no doubt it is also passing the high blue and green frequencies as well, but in relative terms they are not significant. It could either be that green is being blocked by the red filter successfully or that green is being attenuated by the cyan filter...but the point is, there are no particular frequencies that are coming through in any great amount.
Originally posted 41 months ago. (permalink)
Dan (aka firrs) edited this topic 41 months ago.

ByteMyzer    says:

THIS was the link I intended to post:

www.cmst.curtin.edu.au/publicat/2004-08.pdf

The studies referenced in this publication show that wavelength overlaps exist both in different displays AND in different glasses, and that combinations of both are a factor in the amount of crosstalk, or ghosting, that is perceived by the viewer.
Posted 41 months ago. (permalink)

Dan (aka firrs)    says:

Good reference ByteMyzer. The link you sent previously also said that.

There are two potential problems with the glasses and ghosting - any overlap in the frequencies passed and the point of crossover (as it says in the Woods & Rourke paper)

The problem is, that unless you happen to have a spectroscope* or very bright rainbow to hand you can't tell how much ghosting is down to the screen and how much is down to the glasses.

I was actually very surprised how effective the filters in my cheap glasses are...and how bad my monitor is (especially as it is the one I use that works best with anaglyphs). The main problem for me is green. When the monitor displays it, it throws in a lot of blue (not a problem) and a fair amount of red (a problem). I guess I have the option of making anaglyphs with a reduced green channel (making them look a bit purple and less bright) or adjusting the controls of the monitor and video card (and messing up my print calibration).

The Woods & Rourke paper is very interesting and is reassuringly consistent with what I had seen on my CRT monitor - even down to the spikes in the red colour channel. It would be even better if they had done a volunteer trial to see whether people pick the same screen-glasses combination as the (partially qualitative) theoretical approach.

The conclusion of Woods & Rourke is that the glasses are generally pretty good and that for CRT and some LCD the lion's share of the problem lies with the display screen. You can't do much about the glasses you have so fixing your effort on adjusting the output of the screen is probably sensible anyway. The good news is that 3-chip LCD projectors are pretty good for showing anaglyphs so if you want to have a 3D cinema party it'll probably work quite well.

The idea, sometimes mentioned, of deliberately designing one filter to pass some of the light it is supposed to block to combat brightness imbalance makes no logical sense to me. If it helped brightness imbalance it would also cause ghosting. The paper says that the human eye is less sensitive to low levels of blue light so I guess you could allow the red filter to pass a small amount of blue light...but if you can't see that light I don't understand how it could help.

*spectrocopes are great fun but quite expensive (£50 upwards)...although you can make a cheap one with a prism.

Here is another quick idea for a screen calibration image. I'm keen to know if anyone can make a better one (I'm betting someone can).

ByteMyzer    says:

> I was actually very surprised how effective the filters in my cheap
> glasses are...

It's true, the cheap, cardboard-frame anaglyph glasses are typically equipped with better filters than those found in most plastic-frame anaglyph glasses, especially where the Red-cancellation factor in the Cyan filter is concerned. The only plastic-frame anaglyph glasses I've found that even come close to the cardboard-frame standard are from Photo 3-D (www.Photo3-D.com), but even those have noticeable Red-leakage.


> The main problem for me is green. When the monitor displays it, it
> throws in a lot of blue (not a problem) and a fair amount of red (a
> problem).

That's not surprising. In standard RGB colorspace, where Luminosity (brightness) factor is concerned, Green is the most dominant component (87.7 L), with Red coming in second (53.2 L) and Blue in third (32.3 L). Where a given display does not have more discrete emissions, it is quite easy for the Green channel to "bleed" into the other two.


> I guess I have the option of making anaglyphs with a reduced green
> channel (making them look a bit purple and less bright) or adjusting
> the controls of the monitor and video card (and messing up my print
> calibration).

That's a fairly common solution used to combat ghosting, adjusting the histogram of the Green channel to reduce the ratio. As you pointed out, however, this results in significant color distortion and reduced brightness.

Another means of reducing ghosts can be found in Stereo Photo Maker, using the Ghost-Reduced Anaglyph option. The ghosts are reduced by decreasing overall image contrast in LAB colorspace, then resaturating the contrast in RGB colorspace. This works in most cases because it decreases the Red channel output where the Cyan-Red ratio is greater, thereby balancing the overall output. Also, altering the contrast in LAB colorspace alters the image's Luminosity, but does not change the Delta A/B color ratios, thus preventing additional color distortion.

Yet another means of ghost-reduction is available in the program AnaBuilder: a tool called Ghost Buster. The tool employs a more targeted approach to attenuating the output ratios to neutralize the ghosts, and is more effective than SPM's method with some of the more pronounced ghosts.


> The idea, sometimes mentioned, of deliberately designing one filter
> to pass some of the light it is supposed to block to combat
> brightness imbalance makes no logical sense to me. If it helped
> brightness imbalance it would also cause ghosting.

It actually makes perfect sense. If the Red filter is not lightened, the overall light passivity is not balanced between the two filters, thus the left and right eyes do not receive a balanced image.

Yes, it does cause ghosting. Manufacturers of Red/Cyan anaglyph glasses are aware of this, as they are aware of the different means of neutralizing them in the images themselves (you notice, however, that they aren't exactly forthcoming with this information).


> The paper says that the human eye is less sensitive to low levels of
> blue light so I guess you could allow the red filter to pass a small
> amount of blue light...but if you can't see that light I don't
> understand how it could help.

It's not that you can't see that light; it's that your brain does not consciously register it on the same level as the Red when viewing it through the Red filter. Your brain does register it, however.

> *spectrocopes are great fun but quite expensive (£50 upwards)...
> although you can make a cheap one with a prism.

Making your own is more fun, anyway :)
Posted 41 months ago. (permalink)

Dan (aka firrs)    says:

There are a few candidates for messing with the colour space between software and screen (especially for video), but for CRTs to be consistently bad I think the phosphors are the more likely culprit.

I tested this on my own screen by using a microscope to examine the dots on the screen with and without the filters. It looks like RGB red, green and blue do correspond pretty well to the red, green and blue phosphors being lit (meaning the screen settings are OK), but that the green phosphor creates a fair bit of red light as well as green. It's quite hard to record, but I've tried to show it in this video. When the screen is showing green there is red light passing through the red filter - but these are the green phosphor dots not the red ones.

ByteMyzer    says:

> Re the brightness imbalance thing. Rather than letting the red
> filter pass some blue or green light it seems more sensible to make
> the cyan filter block some blue or green. That would darken the
> image overall but not cause ghosting. If you reduce one of the
> colour channels to combat ghosting caused by a lightened red filter
> you're stopping that extra light going through by not making it in
> the first place...so you getting no balancing effect and darkening.
> You either have a more balanced image or you have no ghosting.
> Otherwise you're having your cake and eating it. (ie typical
> salesman's science)

I'm simply passing on what I've learned from various suppliers of Red/Cyan anaglyph glasses. If you wish to argue the point, I suggest you take it up with them.
Originally posted 41 months ago. (permalink)
ByteMyzer edited this topic 41 months ago.

jraaronson says:

Does anyone know how to achieve this in Final Cut? I am editing a 3-D movie now, but using simple color filters over the left and right image is producing some ghosting that I can't easily get rid of. I am sure the solution is in the colorizing, but I can't translate the instructions (which I followed in Photoshop on a still) to Final Cut Pro.

Any help is greatly appreciated!

thanks,
jack
Posted 40 months ago. (permalink)

Dan (aka firrs)    says:

I don't use final cut, but could you give us a list of steps that you are doing? That might help identify the problem.
Posted 40 months ago. (permalink)

schillr says:

jraaronson: color filters will not work. You have to manipulate the color channels of the left-eye and the right-eye movie

Perhaps it would be helpful to create a parallel or cross-eye movie with final cut and then use "StereoMovieMaker" for creating the anaglyph movie?
Posted 40 months ago. (permalink)

Dan (aka firrs)    says:

This works in abobe flash to get your basic (non-colour optimised) anaglyph

Bottom layer = black
2nd layer = left video , colour levels R100%,G0,B0, blendmode = add
Top layer = right video , colour levels R0,G100%,B100%, blendmode = add

...and so I suppose it would work in video editing software if you had the required functions.

But StereoMovieMaker is by far the quickest way I've found to actually make the anaglyph for regular video.
Originally posted 40 months ago. (permalink)
Dan (aka firrs) edited this topic 40 months ago.

jraaronson says:

@dan and @schilly

thank you both for replying!

In final cut I am simply applying color filters to each video, (taking out blue and green on the left, and taking out red on the right). Then overlaying the left on top of the right via "screen" overlay. It works fine, but there is some ghosting I can't get rid of.

I was hoping there was some more advanced way in FCP. My workflow is Mac, so using StereoMovieMaker (which seems to be Windows only?) will certainly be more challenging, but I will take a look at it.

I was hoping there was some similar method to the Photoshop method posted above. FCP's filters aren't analagous (and there aren't the same idea of channels), so I can't directly map the instructions above to FCP.

Thanks!!
jack
Posted 40 months ago. (permalink)

Dan (aka firrs)    says:

Hmm. Can you post a screen grab of the ghosty video?
Posted 40 months ago. (permalink)

jraaronson says:

@dan

Here it is, look at it in full size:
farm4.static.flickr.com/3265/3257786669_40f77d9e86_o.jpg
Posted 40 months ago. (permalink)

jraaronson says:

@dan

If I align the images at my face, it obviously goes away, but I was aligning the blanket on the couch so everything else floated. Perhaps I just need to place the camera further away (it was about 6 feet away) to reduce the variance of the image, and it isn't a ghosting issue as much as shooting distance isssue?


but then again, taking the screen grabs into photoshop and apply the above technique seemed to remove the ghosting.


thanks again.
Originally posted 40 months ago. (permalink)
jraaronson edited this topic 40 months ago.

Dan (aka firrs)    says:

Looks like the video codec is the route of your problem. I don't know much about how they work...but try a few different ones and see if you can get better. I work with uncompressed avi for intermediate editing (yes really) and then compress the final with Sorenson3 Mov.
Posted 40 months ago. (permalink)

ByteMyzer    says:

jraaronson,

Does Final Cut Pro offer you the option to turn off Chroma Subsampling? (YCbCr ratio should be 1x1x1)
Posted 40 months ago. (permalink)

jraaronson says:

I know FCP has "Color smoothing" filters, but one is 4:1:1 and the other is 4:2:2.

Dan, you are definately on to something, as if I look through "QuickPreview", the issue isn't there, but it is in renders. I am experimenting with various Codecs now to try and find one that works the best.

I would prefer to output the files as .mov files. All my 2D movies are exported with H.264. I will keep experimenting unless anyone has any Codecs/Settings they think work best for anaglyph.

Thanks!!
jack
Posted 40 months ago. (permalink)

ByteMyzer    says:

jraaronson,

Is there no option for 4:4:4? Any subsampling, and you will see the ghosting artifacts. Unless you can set the ratio at 4:4:4, you will be able to eliminate these artifacts.
Posted 40 months ago. (permalink)

jraaronson says:

I'm having problems finding a codec that supports it.
Posted 40 months ago. (permalink)