Welcome, Guest. Please Login or Register
Home Help Search Login Register
Page Index Toggle Pages: 1
Send Topic Print
Differences in EXR creation (Read 12962 times)
Eric Miner
Ex Member


Differences in EXR creation
12/26/09 at 19:59:30
 
Happy New Year!!

I'm n the process of writing an article on HDR and have a question on something I've noticed while doing my research.

I'm using the same .tiff files to create OpenEXR files in Photoshop CS4, Photomatix 3.1 and of course Picturnaut 3. Each of the resulting EXR files has very slightly different color balance but very different brightness and contrast. For example Photomatix tends to darken the highlights thus showing more detail in the EXR than either Photoshop or Picturenaut. Picturenaut displays blown-out highlights in the EXR (of course tone-mapping brings out the detail). My question is why this very wide difference between the final EXR files? Is it the method of file creation each app uses? Is it deliberate? What are the mechanics of this bit of EXR creation?
You can see sample images at:
http://www.flickr.com/photos/14800199@N08/
Thanks,
Eric
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Differences in EXR creation
Reply #1 - 12/26/09 at 21:19:34
 
Indeed, I made the same observation. To answer this question, it would be good to know if you have the HDRI-Handbook there, because I'd like to reference some specific pages.
Back to top
 
WWW  
IP Logged
 
Eric Miner
Ex Member


Re: Differences in EXR creation
Reply #2 - 01/03/10 at 18:42:05
 
Yes, I have the HDRI Handbook....it's the best one on the subject. Please let me know what pages I sould read. Thanks
Back to top
 
 
IP Logged
 
Eric Miner
Ex Member


Re: Differences in EXR creation
Reply #3 - 01/10/10 at 23:04:09
 
Hi Blochi, what are those pages you were going to reference for me?

Thanks,
Eric
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Differences in EXR creation
Reply #4 - 01/11/10 at 06:18:15
 
I'm really sorry for the delay, I got sidetracked.

There are three main reasons for differences between HDR generators. They all use different camera response curves, different weighting, and most importantly, there is no real standard for what numbers in an HDR file represent.

Let me tackle these reasons one by one.

1. Camera Response Curves


Since TIFF is a gamma-encoded format, it needs to taken into linear space prior to merging into an HDR image. How exactly that works is described on pages 108 ("What the camera response curve is about..."), and further elaborated until page 114. The point you can take from this is, that every software maker actually has a different approach: either self-calibrating from the source images themselves, pre-calibration with a reference sequence, or from the color profile.
That explains the difference in color balance.

2. Weighting


It has shown that just adding pixels across all exposures is a bad idea. Actually, early versions of Artizen did exactly that, and you can imagine that an image portion that is blown out to white in 50% of the exposures will mess up the average. In reality, an HDR should be made only from the well exposed pixels of each exposure. And this is where the software must make a decision: What is considered a worthy pixel to be taken into consideration?

Obviously, we  drop the black and white pixels.
Methods already shear apart when looking at the almost-white pixels. Are they considered valid when they are clipped in a single color channel? How about clipped in R and G, but not in Blue? Photomatix, for example, will still take them into consideration (which is prone to produce halos around the sun), Photoshop will apply some recovery magic (which can be just as faulty, Picturenaut lets you choose to either take it or leave it, and the next program down the line will do it differently.
The same can be said about the darkest parts of each exposure. Where do you consider the noise to be dominant, and rather look in the next exposure for a good pixel value?
Under the hood each program uses a weighting curve for this. Imagine a histogram, where you overlay the reliability of each possible brightness value. It will mostly look like a bell-shape: Most reliable color values are in the middle, or right of the middle, and then it falls down on both sides until it reaches the bottom for black and white.
The difference is here in the details: Does that look like a gaussian bell? Or is it a triangle (using a bit more of the darker and brighter portions)? Again, Picturenaut lets you choose the weighting scheme here, because we found that in a night shot the close-to-dark portions may become important, whereas a daylight shot gets a cleaner result when you use more of the center of the histograms.
Different weighting schemes explain the difference in Contrast, most notably on the darkest and brightest end of the exposure scale. I guess a good reference for this can be pages 128 to 130, where I did a similar cross-software test like you just did.


3. Lack of scene-referrend standards


That's actually the most puzzling one of them all, and it explains the difference in overall brightness. See, a traditional color value like 255,0,0 is very well defined. Not so with HDR. We are dealing here with floating point values, and they can literally be anything. 10000, 3, 0.7, 0.0000000012233. Literally, anything. So where do we put them?
The original idea behind HDR was to make our image scene-referred instead of screen-referred. That would allow a direct link between luminance of a scene item (like a light bulb) and the color value in out HDR file. So far Photosphere is the only one that actually lets you calibrate an HDR image this way (p 127).
For all practical means, the advent of Open EXR brought us a de-facto standard. What you see in the default exposure as white is 1.0, and medium 18% grey is the 0.5 mark. And from there it extends up and down. Just a matter of what the program chooses as default exposure. The one in the center? The first one loaded? What if the first 2 exposures are so dark that it uses only 5 pixels from it for the final HDR merge, would it still align it to the original center exposure? And where do you put the threshold of actual contribution to not change the middle exposure?

As useful as the White 1.0, Grey 0.5 rule of thumb may sound, I would actually rather prefer HDR would have sticked to the original idea of scene-referrend imagery and aligned everything to absolute EV values. I wrote on that right on the first three pages 15, 16, 17 (which is probably too early and cost me thousands of readers).

Hope this clears things up a bit.
Best,
Christian
Back to top
 
WWW  
IP Logged
 
Eric Miner
Ex Member


Re: Differences in EXR creation
Reply #5 - 01/16/10 at 21:10:38
 
Thank you very much for this excellent info. One more question on this one:
Do you beleive that there should be a standard for this area of HDR creation? If so, what should it be?
Thanks again,
Eric
Back to top
 
 
IP Logged
 
Blochi
Administrator
*****
Offline



Posts: 1726
Hollywood
Re: Differences in EXR creation
Reply #6 - 01/19/10 at 16:18:15
 
Well, the color values in an HDR file should be aligned to the absolute EV values. Or at least, there should be a factor built in that puts it in relation to that.
Back to top
 
WWW  
IP Logged
 
Gerardo
Senior Member
****
Offline



Posts: 74
Re: Differences in EXR creation
Reply #7 - 01/30/10 at 12:28:24
 
Quote:
...My question is why this very wide difference between the final EXR files?
 
 
I've been able to match Photomatix and Picturenaut results in Photoshop by just managing colors and by adjusting exposures.   
 
I think this has to do with how gamma and gamut are managed (or not) within these apps in the HDR creation and in its displaying.  Photoshop uses the chomaticities of your current working color space to create an HDRI, even when the TIF file has a color profile embedded on it, Photoshop doesn't take this gamut into account in the HDR file creation, though it's able to read the gamma curve of an ICC profile (as Picturenaut does). Photoshop will also apply your working color space gamma for displaying the linear image in a log space, as a kind of LUT, seeing that it doesn't bake it in the HDR file.      
   
Picturenaut also doesn't take the chromaticities of embedded color profiles into account, but since it doesn't manage colors, it seems it uses sRGB color space chromaticities when creating the HDR image. At least with TIFs and contrary to Photoshop, Picturenaut bakes the gamma in the HDR file if we haven't used linear option when generating the HDR image. This is apparent when we open this file in Photoshop or a 3D package.         
Last versions of Photomatix seems to work in similar way than Photoshop, I say it seems because it uses your monitor's color space chromaticities to show you a linear result, but when you tonemap/DRmap the image, it uses the chromaticities embedded in the LDR images. Photomatix doesn't bake the gamma in the HDR file.    
   
 
 
Gerardo 
 
Back to top
 
 
IP Logged
 
FlashPhoto
Junior Member
**
Offline


HDR Boy!

Posts: 6
Re: Differences in EXR creation
Reply #8 - 04/01/11 at 17:35:48
 
I was very unaware of these issues because i work primarily in CS4, but this is very interesting to know. I am currently working on some vehicle wrap designs, and this has not affected my progress. Have any of you heard of b2bmedia? I was hoping they might accept some of my car wrap designs. Hopefully this issue hasn't cause anybody too much trouble as it seems like minor issue.
Back to top
« Last Edit: 04/09/11 at 16:14:12 by FlashPhoto »  

Why did you have to go and blink your red eye?
 
IP Logged
 
Page Index Toggle Pages: 1
Send Topic Print