One of the problems of films and even more of digital sensors is their capacity to reproduce scenes with high contrast. This is characterized by blowned high-lights and/or completely dark shadows, whereas in the field, the eye discern much more tonalities. Sensors have indeed a limited dynamic range (a capacity to record variations of luminosity), inherent in technology employed. It is thus necessary in such scenes to choose between an exposure preserving high-lights, or an exposure preserving shadows. There are however solutions, but which impose certain constraints in term of shooting. We will try here to understand HDR photography and its usability.
  HDR, definition
Application to photography
HDR photography, realization (shooting)
HDR with Photoshop CS2
Conclusion
HDR, definition
First, let us start by defining term “HDR”, or High Dynamic Range. I use myself such images since the end of nineties, when it appeared thanks to Paul Debevec studies, in 3D images and animations. This format is now more used in openEXR (developed by ILM, the famous studio of George Lucas) in rendering and compositing.
A pixel of an image is composed of three layers of colors, Red Blue Green, each layer being able to contain a certain number of tones. In 8bits, each layer is thus coded on 8bits (or 8bpc, for Bits Per Channel), to say 256 tones of red, green, and blue. Indeed, data processing being binary, one has 8 “0” or “1” to code information, so 28 = 256 possibilities. On three layers, one thus obtains 2563 = 16 777 216 colors, in other words 24bpp (Bit Per Pixel).
A 32bits file is in fact a 8bpc file (so 24bpp) but with a fourth layer, called alpha layer, containing information of transparency. For example, a pixel 100,100,100,128 will be a dark gray pixel, at 50% of transparency. This just in order to clarify things, as we won't be interested here in the alpha layer.
Well, any scene should be divisible in 256 levels, but that depends directly on the dynamic range that the sensor is able to stand, and thus on the technology of the current sensors. For example, if you have shadows in a fully sunny scene, the sensors (even best films) won't be able to "divide” such high variation of contrast into 256 levels. From a technical point of view, while recording the very low number of photons emitted by the shadows of the scene, the photosites of the sensor (or silver grains of a film) should be able to record at the same time an enormous quantity of photons in the high-lights. Today, such a quantity of photons “overflows” of the photosites, they saturate, and thus records a pure white. It's even more obvious on ultra compact cameras, which have very tiny photosites. Maybe one day we will have some photosites able to be emptied and record continuously during one exposure time, but that's still far … Thus, in such a scene, on one hand you increase exposure time to capture the shadows, saturating high-lights, on the other hand you decrease exposure time to preserve high-lights, but then shadows will be black, because too little (or no) photons emitted by the shadows will hit the sensor in such a short time !
You should now glimpse a solution : to divide the capture of the scene to once for shadows and another once for high-lghts, and then to merge these two parts to make only one larger image ! That's indeed the principle of shooting HDR.
HDRI images (High Dynamic Range Image) are in fact a combination of several shots with different exposures, authorizing levels for brightness and chrominance higher than 255. For example, where a white wall will have an intensity of 1.0 (255,255,255 in RVB space),, the sun, also white in space RVB, will have an intensity of 4.5.
The raw, recorded in 12bits, now in 14bits on last Canon DSLRs (and 16bits on the digital backs) makes it possible to get the complete dynamic range of the sensor, already higher than 8bits, but does not completely solve the problem. HDR Formats are indeed much wider, reaching usually 32bpc. Still some figures :
8bpc = 28 = 256 tones per channel.
12bpc = 212 = 4096 tones per channel.
14bpc = 214 = 16384 tones per channel.
16bpc = 216 = 65536 tones per channel.
32bpc = 232 = 4.2 billion tones per channel.
Even our 16bits raw file is all at sea by an HDR file, including 4.2 billion tones per channel. Do we have to still refer to the 256 tones of our jpgs files ?
Particular case of the human vision.
We could argue that an HDR file approach human vision and the capacity of the eye to distinguish details in the shadows even in full sun. It would a lot simplifying things, because the eye dynamic range, even if wider than any sensor or film, is far from being infitite, and doesn't work in the same way. Indeed, when we show an image file (screen or prints), all informations is contained in the file and do not vary.
But the eye doesn't see the entire scene, or rather see it, but don't make out details apart in a restricted aera, where you really look at , corresponding to the fovea, central zone of macula, zone of the retina where the vision of details is most precise. And where the grace of nature resides, is that the eye adapts in nearly real time to the average brightness present in the fovea, on the one hand by instinctive reaction by adapting the diameter of the pupil, but also by adapting the sensitivity of the retina, which translates in a nonlinear way this brightness into luminosity sensation (in electric signal to the brain). Thus, like our sensor, you won't be able to tell apart details in shadows if the sun is directly in your field of vision (in the fovea), on the other hand by shifting your field of vision a bit, even if the sun is still “visible”, the eye adapts and tells apart details.
It will thus be necessary in our HDR file, which is complete, to have an even wider dynamic range than the eye.
|
Top
