l'auteur

en français

The articles
digital sensitometry (part 1)
 digital sensitometry (part 2)

Digital sensitometry, 
a 3-part article

Introduction by Emmanuel Bigler

Digital image capture with silicon detectors has been widely used for scientific applications in the last 20 years, but it is only since the end of the 1990's that photographers have started to know and use digital image sensors. At a first glance, photographers were not happy with mediocre results in terms of very limited resolution (not enough pixels in the image) and they did not realise at first the considerable improvement of silicon detectors over film in terms of noise, low light-level sensitivity, linearity, colour management and contrast control. In other words, the limited available number of pixels in digital sensors offered around year 2000 (with the exception of scanning large format backs, for static objects only) when compared to any image recorded on film, has prevented photographers to recognise immediately the true revolution of silicon image detectors. This era is now over.

In this series of 3 articles, Henri Gaud will start by what was clearly overlooked by photographers, namely photometric characteristic curves and contrast and ISO sensitivity management issues in a silicon image detector associated with a camera-level proprietary software. In a second article, Henri Gaud will examine on real images different possibilities offered by various ISO settings in the field followed various different post-processing in the "digital darkroom". It will be shown how digital sensors bring an incredible improvement over film in terms of dynamic range and image noise. In a last article comparative image resolution tests will presented, the competitors being the Canon 1 Ds MkII, a 6x8cm film camera and a 4"x5" view camera.

Preliminary Technical Note

In these articles, a full-frame 24x36mm sensor (Canon 1 Ds Mk II digital SLR camera) was tested in combination with its proprietary software delivering RAW format images. The only data to which the photographer has access is this RAW file which is already the result of some internal computing, no access to the physical sensor level is allowed.

What is actually stored inside a RAW image is actually unknown since those format are proprietary and not publicly disclosed. Very probably, RAW imaga data are very close to the physical level of the sensor output, which is linear versus the number of incident photons per second and per unit area or per pixel in a wide range of incident brigthness. However, as soon as the RAW format is converted to be post-processed, data are expressed in a non linear scale (power law) and no longer a linear scale vs incident brightness. It is interesting to represent those post-processed image levels exactly like characteristic curves of a film (traditionnally in a log-log scale), not that we need some nostalgic reference to the optical densities of a black and white film, but simply because the human eyes requires a non linear scale in the image so that grey levels look properly spaced and balanced.

In the first article, separate RVB colour management will not be taken into account, only the global brightness level will be considered exactly like with a black and white panchromatic film. In reality, the real output signal, the physical signal on output of a silicon photodetector, is a number of photo-electrons per second for a given input incident photon flux. In this input/output model, a silicon photodetector is perfectly linear in a range of 1:10000 between the weakest photon flux and the brightest one. Taking this into account it is hard to understand at a first glance why a non linear scale would be needed if the sensor is so extraordinary linear. In fact the human eye is mostly sensitive to the logarithm of the number of incident photons, so digital image post-processors actually compute a non linear power-law scale derived from the number of output photoelectrons.

Doing so, the comparison with classical film is much easier, photographers who are used to think in terms of optical densities on output and f-stop scales in input will feel at home with their familiar sensitometric curves. The goal of the article is to show that anything valid and well-known for film-based sensitometry can be immediately adapted to a silicon image sensor, with actual results that would have appeared incredible to photographers only five years ago : silicon goes far beyond what is known of film limits in terms of noise, sensitivity and contrast management.

A millimetre-size conventional photodetector operating in analog mode with a continuous input photon flux (this is not exactly the case with CCD image sensors where charges are trapped in a CCD register behind the photo cell) has a dynamic range of 1:10000, equivalent to 13 f-stops. A dynamic range of 1:1000 is equivalent to 10 f-stops, corresponding to the 10 or 11 black-to-white zones of Ansel Adams' Zone System (in fact a 1:1000 ratio corresponds to 11 zones ; to many authors this is too wide, 9 zones or 1:256 being considered more reasonable for conventional B&W prints). The optimistic dynamic range of 1:10000 in a conventional silicon photodetector is probaly not valid for a very small sub-pixel silicon detector element, about 3 microns size in the curent 2005 "bayer" pattern of silicon colour image sensors. But let us forget for a while what we know about conventional silicon photodetectors and let the photographer speak and share his experiments with us in photographic terms.

Reference on RAW image coding and visualization http://www.normankoren.com/digital_tonality.html 
(thanks to Yves Colombe for this useful information)

dernière modification de cet article : 2005