Well said, Godfrey. This goes in my keepers file. Paul On Jul 4, 2006, at 12:18 PM, Godfrey DiGiorgi wrote:
> Yes, of course you don't want to go beyond the sensor's saturation > point in areas where you want detail. But "protecting" an exposure by > underexposure is the wrong methodology. It's difficult to explain > without some graphs to make it more intelligible, that's why I > recommend reading Bruce's first two chapters. He did a great job. But > I'll try. > > --- > A digital sensor is a photon counter. It simply counts up the amount > of light falling on a photosite's unit area in the time of exposure > and reports that number. As such, it is a linear gamma device ... > unlike the human eye or film, intensity values in a scene are simply > represented by that linear count of photons falling on the photosite > array. > > The saturation limit happens when the sensor runs out of numbers ... > in the case of the sensor used in the Pentax DSLRs, it can count from > zero to 4095, 12bits of quantization ... which depends upon how much > light energy is falling on the sensor, how much loss is embedded in > the photosite's design, and how much time the photosite is exposed. > This all conspires together to place saturation on a hard edge ... > the number 4090 is not saturated, the number 4095 is because you > cannot record the number 4096. > > By that we've established what "saturation" means. The next thing to > understand is how RAW data relates to a tonal scale that makes sense > to our eyes. > > One of the basic operations performed by RAW Conversion is to do a > gamma correction on the captured data ... By this is meant that the > linear capture of the digital sensor is transformed to re-place the > light values in the characteristic curve of human vision's > sensitivity and perception. The human eye perceives fewer differences > between bright values and greater differences between dark values > than the sensor by comparison. So the high values recorded by the > sensor are compressed together ... values that are insignificantly > different in perception are thrown out ... where the low values are > expanded ... values that are crowded too close together are > interpolated/stretched to fit the range required. > > Now, if you consider a linear scale of numbers recorded by the sensor > as a binary representation, you'll see that the half the total amount > of exposure falling on the sensor is stored in the last bit of > representable numbers. Half again is stored in the next bit down. > Half again is stored in the next bit down after that. What this means > is that, in the range of the sensor's linear number scale, the > midpoint of exposure to our eye (call it Zone V) is NOT in the middle > of the scale, it's actually down around the 1/8 point in the linear > scale. So all the tonal values that make up the important range from > Zone II to Zone V are smashed together in the bottom end of the > sensor curve, and most of the data values that take up more than 3/4 > of the scale are insignificant to human perception. If you > underexpose the scene, more interpolation and expansion of that small > part of the scale must be performed to fit the data to the proper > perceivable range, which has as its byproduct noise and ambiguity in > the Zone II to V range through round-off error. > > What this means to a photographer making an exposure evaluation is > that the photographer should consider the linear capture qualities of > the sensor. If you pick the brightest points in a scene, the points > of specular reflection for instance, you want to place your maximum > exposure so that they are just AT the 4095 data value threshold ... > this is hard because you can't see when you go over with them. So you > look at the Zone IX values, the brightest parts of the exposure where > you want to retain detail, and try to keep the values in a capture of > those points to somewhere in the range of value 3686 (around 210-220 > in 8bit data), and leave the other values to fall where they may. If > you look at a scene after capture with the histogram display on the > camera, this kind of exposure will "crowd the right" ... The goal in > doing it is to capture as much distinct data as possible without > saturating the important detail areas. > > In processing, you place the gamma correction curve to handle a given > scene's exposure by adjusting the white clippint point (exposure), > the brightness and contrast (essentially, it moves the nodal point of > the gamma correction and the angular relationship of the resulting > curve inflection) and then the black clipping point (the point at > which you decide where the differences between low values are purely > ambiguous and insignificant). Exposing as much as possible without > saturation means you have more data at the low end to expand through > interpolation with the least round-off error and noise. > > Regarding the current Pentax *ist D series cameras currently > available, I'll use the *ist DS as my specific example but I believe > the same is true for all of them: > > The *ist DS is set by default for Auto Picture exposure automation > with JPEG *** fine quality, using a Bright color tone and intended to > produce a snappy, pleasing image for a 4x6 inch print. What this > means is that the in-camera RAW conversion algorithm is tuned to that > output, and the meter is calibrated to produce results compatible > with that algorithm. The *ist DS does NOT change the meter > calibration curve when the user takes control of the rendering engine > and requests a RAW file as output. The difference in output > requirements is critical: the default JPEG *** rendering and Bright > color tone means that meter calibration has to be set optimistic to > suppress highlight saturation with the embedded RAW conversion. > Switching to RAW capture and using the meter's default calibration > results underexposure because RAW format data has more stops of > overhead before saturation values are reached at the sensor. With a > customized RAW calibration curve, you can obtain better data with > more exposure on the 12bit capture. So I find that my average > exposure compensation when capturing RAW format runs +0.3 to +0.7 EV, > without saturating highlights, and allows much cleaner, lower noise > data in the critical Zone II to Zone V range. > ---- > > I hope that helps. I'm not as good at explaining this stuff as Bruce. > > Godfrey > > > On Jul 4, 2006, at 8:23 AM, Bob Sullivan wrote: > >> Godfrey, >> You've got to explain this. >> Digital sensors can't give any detail in overexposed highlights. >> You can recover details in underexposed areas with post processing. >> So don't you want to avoid blown highlights at all costs? >> Regards, Bob S. >> >> On 7/4/06, Godfrey DiGiorgi <[EMAIL PROTECTED]> wrote: >>> Sensors respond to light differently compared to film. Chapters one >>> and two of Bruce Fraser's "Real World Camera Raw with Photoshop CS2" >>> explains why there is a difference. As a result, exposure evaluation >>> requires a different mindset and different settings. JPEG and slide >>> film, although they are different, generally end up taking about the >>> same exposure. >>> >>> However, underexposing in RAW by 0.3-0.5 EV is exactly the wrong way >>> to go. In general, with the *ist DS, I find my average exposure for >>> RAW capture requires +0.3-0.7 EV additional exposure compared to JPEG >>> or slide film. > > > -- > PDML Pentax-Discuss Mail List > [email protected] > http://pdml.net/mailman/listinfo/pdml_pdml.net > -- PDML Pentax-Discuss Mail List [email protected] http://pdml.net/mailman/listinfo/pdml_pdml.net
