[HDRI] Re: NYT article

[Geraldine Joffre]

On Fri, 11 Aug 2006 10:42:38 -0400, Kirk Thibault wrote
> On Aug 11, 2006, at 4:36 AM, Geraldine Joffre wrote:
> <snip>
> >
> > Photographers in general do not wish to produce photorealistic images.
> > Geraldine Joffre
> >
> </snip>
> 
> This is a very interesting statement, especially in light of 3D  
> modelers and renderers who are trying to produce "photorealistic"  
> images by using HDR photographs as radiance maps and reflectance  
> maps.  Fascinating philosophical point of view!  I guess it comes  
> down to one's definition of what is "photorealistic".

You are right. I should have made it clear in what sense I used the term "photorealistic" here. By 
"photorealistic", I was referring to the attempt to mimic the effects of camera limitations in order to 
make an image (a computer generated one for instance) look as if it had been taken with a camera. 

If you ask a photographer what the biggest limitation of today's camera is, he will probably answer 
without any hesitation: dynamic range. So, because of this limitation, a photo of any high or medium 
dynamic range scene will end up with blown out highlights or undistinguishable shadows or both. 
Therefore, an image that does not show this limitation, i.e. that reproduces the details in highlights 
and shadows of the original HDR scene the same way a human observer had seen it, such image 
does not look like what people expect a photograph to look like. In this sense, such image is not 
photorealistic.

Photographers are not happy with the limitation of their camera, and this is what I meant by 
"photographers do not wish to produce photorealistic images". Photographers would like that their 
camera were able to reproduce that beautiful HDR landscape that captivated their eyes, but they 
know the camera will not collaborate... 

This is why the promises of HDR photography are so exciting for photographers. They are all looking 
forward for a camera that will give them directly an HDR picture. (Please note that I am intentionally 
using the term "picture" instead of "image" here, to make it clear that what photographers are 
interested in is the final tone mapped output and not the "raw" 32-bit HDR image itself.) 

By saying "HDR picture" a photographer would thus mean a tone mapped HDR image. Such image is 
technically speaking an LDR image, but for a photographer it reproduces the high dynamic range 
captured, and is thus referred to as HDR. A photographer will consider such tone mapped image as a 
good "HDR picture" if it succeeds in reproducing the highlights and shadows as he/she saw them at 
capture time. If the tone mapped image looks like a "photorealistic" blown-out image instead, it 
won't be considered "HDR" for a photographer. 

Let's take this example to illustrate that
http://www.hdrsoft.com/images/grandcanal/tm.html

Most people looking at the above tone mapped image (especially when they are not photographers 
themselves) will typically react this way: 
"This picture does not look natural. It looks more like a painting than a photograph". 

However, if you tell that to Jacques Joffre, the photographer who took the bracketed shots for this 
picture, he will reply: 
"But I was there and this is exactly what I saw! It is precisely because it looked so beautiful that I 
decided to shoot this scene. I would not have done it if it had looked like a dull photograph."

I won't claim that the tone mapper of Photomatix always succeeds in closely reproducing what the 
photographer saw. This depends on the image and the dynamic range, sometimes it works very well, 
sometimes it does not, and other tone mapping operators -or simply blending the original 
exposures- will work better. And of course, a user can always intentionally choose settings that 
make the image look "special" and far away from what it was in reality. But the same can be done 
with any TMO that offers sufficient user control, it is unfair to blame only Photomatix for that. 

My point is just that, in my opinion, we should not rate a tone mapping algorithm on how 
photorealistic its output is perceived to be, but rather on how well it reproduces what a human 
observer has seen.

Geraldine Joffre