Dear Thomas, > David Kay wrote: > >> Surely there's a relationship between bit depth and dynamic range when >> available bits are required to describe tonal variations in our subject? > > No. I really can't see your logic here Thomas. Underexposure with any bit depth device lowers all values by however many f-stops you underexpose. Therefore, this has the effect of nudging the lowest values off the scale where they FALL below the threshold of the device! The device stores all data before processing linearly. As you know, the unprocessed image looks anything but "photographic," until properly toned.
When you make exposures with a digital back, photons or light energy, are converted to electrons or electrical energy, PROPORTIONATE to the intensity of the light! Therefore there IS a relationship to exposure in f-stops! >> >> For example: >> Imaging we only had an 8-bit capture device and I set up a shot where the >> highlights were underexposed by say two f-stops. >> >> So instead of having all 256 8-bit levels available, I now have only 64. > > Ok Let's play this theory game. > You have all 256 levels available as you always do in 8 bit, but everything > from 64 to 0 will be pitch black, or filled with noise if we were to follow > you (flawed) logic. No! You missed the point completely. Everything from 64 to 1 will contain information, however, my point is simply this; when you reach the part of the subject represented by 0, the device can't describe tonal variations lower than this! And yes, the presence of noise could destroy detail because it's added to whatever signal we have been able to record. > If you underexposed the image two f-stops, your image would be underexposed > by two f-stops, which, depending on the dynamic range of the camera can mean > a great deal of things in regard to which bit will hold the last discernible > detail. Increasing the bit depth of the camera would allow you to draw out > the bits that you couldn't discern in 8 bit, but you couldn't pull out > details that are totally black (0 in any bit depth). Agreed > Where 10 bit is an advantage is when you in 8 bit, after 6-7 bits > (256,128,64,32,18,8,4,2) have to decide whether a dark colour will be > rendered with a value of 1,2, 3 or 4. If you have 10 bit, the choice of that > same pixel colour is anything between 32 and 1. This will lead to more > accuracy in correctly rendering shadow details. It doesn't lead to the > camera being able to see darker colours though. I believe it does. See first response. (proportionate) In subjects with very long exposure scales, the higher bit depth device, with correct exposure, can "see" more into the shadows. > >> >> My subject happens to have a tonal scale of 8 f-stops (measured with a spot >> meter) and the brightest part of the subject, thanks to my shoddy exposure >> technique, fell on 64, then 32 represents half that brightness or one f-stop >> less, 16 represents two f-stops less, 8 is 3-stops, 4 is 4-stops, 2 is >> 5-stops 1-is 6-stops until we run out of available bits with which to >> describe tonal variations in the subject. > > No! > Without knowing the Dynamic range of the capture device (2 f.stops, 6, 7, 8 > 9, 10, 11, 12 or umpteen) there is no way of saying which bit's will be > black! My understanding is that with every f-stop less exposure, the volume of stored electrons in the pixel bucket is halved, or doubled with one f-stop more exposure. It is this relationship I am referring to. > If we exclude any gamma tweak in the software (deviation from linear) any > curve adjustment (shoulder and toe contraction as on film) and colour > adjustment matrices (other than adjustments to the quantum efficiency after > colour filtration/bayer mosaic to adjust the spectral sensitivity of the > camera to something akin to what humans see - and removing IR and x-Ray > sensitivity) and a few other issues, AND we know that the camera has a > dynamic range of exactly 8 f-stops without introducing noise, then your > assumption is correct. (Sorry for the long sentence). > But it isn't. >> >> Had I "exposed for the highlights" by placing them as far up the scale as >> possible, I could have captured all 8 f-stops of information - without >> turning up the fill light. > > If your camera had a 10 f-stop dynamic range you would have anyway (assuming > linearity, no gamma/curve etc.). But if your sensor had had a 10 f-stop > dynamic range and you exposed for the highlights, then your absolute shadow > area would have been lighter than black so you would have needed post > shooting adjustment to fill the 8 bits fully. Agreed. Fortunately we have higher bit depth devices that give us a margin for error and allow us to "tone" the image to suit the look we're after. > Or you would have been able to > see detail that your spotmeter couldn't measure out of the shadows. > Had the camera had 12 f-stop dynamic range you image would probably look > pretty flat and uninteresting without post adjustment, as your wouldn't have > any good black anywhere in the image. If your camera had an 8 bit f-stop > tonal range, but a 16 bit bitdepth, and you exposed for the highlights your > image would have been perfectly captured, with accurate shadow detail, and > no post capture adjustments would be nessesary. Please explain how a 16-bit device can have only an 8-stop tonal range. The manufacturers of these devices are claiming 11-12 f-stops of useable range. And if the device is cooled, virtually all of it is useable. >> >> Many recent medium format based digital backs have 16-bit A/D converters, so >> there's a huge f-stop range available. > > These are separate issues. > It's like saying my car with 4 wheels can go 100 mph. Now if I get 12 wheels > I'd be able to go 300 mph... > Not exactly true, but more wheels would without doubt increase the road grip > and make sure the car will be more stable when you press it to the limits. > > If you have a 12 bit chip and a 12 bit A/D converter there is no way you > will fit all 12 bit unharmed through the A/D converter, and you'd be left > with less than 12 bits. Therefore most vendors fit an A/D convert of a > higher bit depth than their CCD/CMOS sensor. Agreed. However, the key to how many tones a given device can describe is related to things such as; how many electrons the CCD can store when fully saturated, whether the manufacturer uses an A/D converter of a capacity to preserve everything stored - they cost more, the quality of the wiring/shielding within the device and whether there's any active cooling been implemented, thus helping to keep the shadow end cleaner due to noise minimisation. > > If they fitted a 139 bit A/D converter it wouldn't increase the dynamic > range one bit though. Dynamic range ARE different from Bit depth. I agree, there's no point in fitting an over-specified A/D converter to a device that can't deliver that much range. >> Just make sure you don't overexpose, >> as it's very hard to put detail back once this happens. > > True. I always recommend "exposing for the highlights" to help preserve whatever range is available. This keeps the noise, which appears more in lower values, from being added to whatever image data is there. However, eventually all tones below the threshold of the device go to black. With most normal subjects, in a high bit depth device, there will be room at both ends to "render" these values as white and black - or first and last printing tones. I often use a linear tone curve to initially set the end points. (black and white) This effectively "cuts off" any values above the white point or below the black point. Once I have set this I can then add points to the tone curve, where needed, to selectively lighten or darken local areas. And there are times when I export two images, one with a steeper shadow end for better local shadow separation, (local contrast) the other with a curve that best suits the highlight end. Oftentimes it's impossible to use only one toning curve to get the best result. Naturally you always have the option to "light for a tone curve", rather than "light with a tone curve". I have found starting with a default contrast-increasing curve shape is often inappropriate with many subjects. Try copying some trannies with a digital back with a curve shape such as this. >> So isn't there a relationship between bit depth and subject contrast in >> f-stops? > > Nope, not one bit > <pun intended>. Pun received! As you can see from the above, I strongly disagree. Maybe in light of the above comments you'll reconsider? Best regards, David Kay (The Tone Ranger) =============================================================== GO TO http://www.prodig.org for ~ GUIDELINES ~ un/SUBSCRIBING ~ ITEMS for SALE
