I read with much interest the thread about the comparison of various HDR modes (Comparison of Various HDR Modes: Newbie Seeks Advice). To me, this is not at all a newbie question. On the contrary, it expresses a concern with a real understanding of what is going on in the use of the tone mapping programs available. I myself have pushed the sliders around repeatedly and experienced major difficulties in trying to put words on the effect this has on the image. I belong to that group of people, who want to go beyond the gross qualification of the "I like / I don't like" type.
The basic principle of tone mapping HDR shots sounds simple: reduce the overall contrast and enhance the local contrast. But even if one limits oneself to one single program, like Photomatix, the variety of results, which can be obtained, is tremendous. This makes it actually difficult to compare any program with any other: to document the comparison with precision, you should take notice of all values entered and of the position of all sliders.
What we lack is a perceptive / conceptual framework against which we could qualify the visual effects obtained.
The software editors do actually not help us in any way to build such a framework. The explanation about what the program does behind the scenes is quasi inexistent. I expected some help of several books I bought or went through, but most of the time, the authors copy/paste the tutorial of the program and conclude writing "try yourself", which at the end means "help yourself"... This is not a fifty bucks worth.
Tone mapping works best when there is a clear delimitation between large scale zones with clearly distinct groups of tones; The image of Westminster discussed in the cited thread complies largely with this requisite. I yet agree with the contributor, who suggested that the picture would eventually be best rendered by an appropriate edition of a single shot.
Tone mapping works much less well when the tones from the whole 14 EV DR mix up on a small scale, which will be the case for an image taken in a forest with very large tonal differences between bright rocks in the sun and wet woods in the shadow.
The worst thing, which may happen, is an inversion of tonal values, where the brightest point in the original picture turns gray, sometimes a fairly pronounced Grey, just because it is large enough to be treated by the program as a bright gross zone, which has to be mapped down. In some programs, the radius of the details, which have to be enhanced on the local level, is user defined. On the opposite, the scale of the gross zones, to be compressed, is out of the reach of the user. And in any case, a real image involves usually zones of various scales, a variety which is not best handled by software, which will apply a uniform filter for defining the picture zones.
So, we get artifacts not only due to the detail enhancing technique used (the halos) but also due to a mismatch between the presuppositions the algorithm works with and the scale of the gross zones and smaller details in the picture.
In my view, the gross zones to be globally remapped should be user defined, by the way of a selection operator (the spline curves of Lightzone would be perfect for that), because the user only knows which mapping really makes sense. The 'universally applicable' technical criteria predefined in the software MUST mismatch with a majority of pictures. A software cannot Analise a picture as does the human eye 'n brain.
Presently, I try to build a two stage process in which I have more control. In a first stage, I compress the overall range to fit most of it into a tiff file. In a second stage, I use a picture editing program to enhance local detail at various radii and on selected parts of the picture. The main difficulty is to find a program, which really just compresses the overall range without adding local contrast by itself. As I discovered HDR Exposure, I first believed it made this possible; experimenting more, it appeared that all tone mapping procedures it proposes include local contrast enhancement, which produces the unwanted tone scale inversions in the highlights.
At the moment, I feel that the Compression tab in Photomatix is the best option, because it preserves the high value of the highlights although it seems to apply local contrast in the mid tones and shadows (Contrast Adaptation). Of course, if some of you have made better experiences with other programs, I would be happy to read about it.
As of HDR in itself, I wonder if the scope of the processing is not somehow ill-defined. Of course, the luminosity may span over 16 to 24 EV's in some scenes and thanks to its multiple adaptive mechanisms (physical, physiological and neural), the human vision can capture the whole span with detail, provided one gives it the time to adapt to specific viewing conditions. Even the human eye cannot differentiate bright colors in the sun and subtle shades in a deep shade AT THE SAME TIME. In reality, for given viewing conditions, it may well be that the effective dynamic range covered at a glance goes not beyond 8 to 12 EVs... This may be a reason why images covering about that range feel realistic, on screen as on print.
I tried to experiment this the other day while hiking in a bright rocky landscape with many shades from very bright to deep shadows. Looking in various directions and concentrating either on bright sunlit stones or on dark pieces of wood in the shadows, I certainly covered a wide range, say 16 EV. But obviously, I did not see any detail in shades while concentrating on bright rocks and the latter were totally 'overexposed' while I concentrated on shades. So, human vision itself seems not to cover the same range if one considers panning around, giving the adaptive mechanisms time to adapt, or a snapshot, where human vision seems to be far more limited.
If the description above fits, then HDR tone mapping consists in squeezing the panning view over time into a snapshot view. It makes it possible to see in a snapshot 'as much' as panning around. It is a play with various kinds of visions rather than an extension of the photographic capabilities to the capabilities of human vision.
To me, this means that straight pictures limited to 8-12 EV's should not only be viewed as a result of technical limitations of the screen and print but also as the most 'truthful' reproduction of a snapshot a human being takes of a situation. On the other hand, tone mapped pictures squeezing overall contrast into those 8-12 EV's and blowing up local contrast feel more artificial, because we never really see things like this. This does not mean that they should not be made. To me, they have a status similar to the views astronomers take of distant galaxies in false colors: the manipulation is aimed at making something visible, which otherwise would be overseen. The question which we have to answer as photographers concerns the sense of the manipulation. One can consider documentation, stylistic, artistic aspects. Here I come back to the discussion launched by John T. in the other thread: how far is the HDR to 'grunge style' merely a consequence of the technique applied, to what extension is it a choice which makes it the signature of an artist?
There is nothing like 'photorealism'. Photographic pictures are always transposition. Solarisation, toning, contrast enhancing, low or high key, blurring, exaggerated grain, many manipulations have been used with best results. HDR tone mapping can also produce 'artistic' results (whatever this means...). My feeling is yet that many tone mapped pictures are obtained more or less 'by chance' and that the technical constraints of the programs presently available are raised to the status of 'style'. There is still much to be done to have real control on a high dynamic range.
Reto