Posted by Patrick Nixon on April 30, 2009 at 2:41 AM
Is it really necessary to resort to advanced formats (RAW or TIFF) or – moreover -- combine images taken at different exposures into high dynamic range (HDR) images to faithfully record the scenes we all seek to capture?
Under the best conditions, the most healthy and perceptive human eye can distinguish about a million (1048576=2 to the 20th or 20 EV or stops) different levels of luminance – and that requires time for adaptation to light or darkness and may vary depending upon individual sensitivity to color. The comparable value for a quick glance under bright sunlight is 131,072=2 to the 17th or 17 EV or stops.
The JPEG format is equivalent to 256=2 to the 8th or 7 EV (plus/minus 3 from pattern-metered exposure) EV or stops. At best, 14-bit RAW=16,384=2 to the 14th or 13 (plus/minus 6) EV or stops. The most common 12-bit RAW=4096=2 to the 12th or 11 (plus/minus 5) EV or stops. Theoretically the 14-RAW, 12-RAW and JPEG formats can record only 1/8th, 1/32nd and 1/512th of what the eye can typically see in bright sunlight, respectively. Proving that even the best cameras can record only a small percentage of what the eye can see or distinguish.
The real question seems to be: How often do actual scenes exhibit more than the plus/minus 3 stops (7 stop range) of luminance we have come to accept with the JPEG format. Or the plus/minus 5 stop range equivalent to the 12-bit RAW format. What kinds of scenes typically exhibit a greater range? Sunsets? Green mountains or prairies against blue sky at mid-day? Or sandy beaches or snow against blue sky?
If the answer is “Not often.” there would seem to be little need to resort to the RAW format or HDR post-processing of bracketed exposures. Yet we have all seen stunning improvements using RAW or HDR in photo books.
What say you all? What kinds of scenes or light conditions typically demand bracketing, RAW, or HDR? Does anyone know of an authoritative reference setting forth the TYPICAL luminance range for different scenes or light conditions?
The answer is a lot simpler than you think: any scene where contrast range is larger than the dynamic range of the sensor will require combining exposures *if* you want to have detail in every zone (not always the case). What makes "typical" is really up to you - where and when you prefer to work.
File formats and bit depths are not related to dynamic range. Dynamic range is the number of EVs (aka "stops") between the darkest and lightest values the sensor is capable of detecting detail in. The bit depth represents the number of discreet tones within that range.
Beyond that - detail tapers off towards the extremes so within the dynamic range there's only a portion that is usable (i.e. has enough detail to sufficiently separate visual elements).
Then there is noise. The sensor measures the amount of photons reflected towards it. If there are not enough of them, the measurement will be inaccurate and errors will show as noise in dark areas of the frames. Even if there's sufficient dynamic range, it may still make sense to combine exposures to gather more accurate data from the shadows.
At the end it's not just about the sensor but also depends on the specific image - how important extreme shadow/highlight detail is to its success. In many cases the dynamic range in a single exposure is more than sufficient, but not always.
Remember many of the most memorable photographs made over the years used films with usable dynamic range as narrow as 4 stops. Modern DSLRs are generally capable of about twice that.
Guy
Posted by Dan Borzynski on April 30, 2009 at 10:48 AM
Quote:
"The JPEG format is equivalent to 256=2 to the 8th or 7 EV (plus/minus 3 from pattern-metered exposure) EV or stops. At best, 14-bit RAW=16,384=2 to the 14th or 13 (plus/minus 6) EV or stops. The most common 12-bit RAW=4096=2 to the 12th or 11 (plus/minus 5) EV or stops. Theoretically the 14-RAW, 12-RAW and JPEG formats can record only 1/8th, 1/32nd and 1/512th of what the eye can typically see in bright sunlight, respectively. Proving that even the best cameras can record only a small percentage of what the eye can see or distinguish. "
No.
The issue is sensor and electronics limitations, not file format limitations. You could build a 16 bit sensor that captured one stop of dynamic range. What you would wind up with is is 2^15 different levels of sensitivity in that one stop of range. You could also set things up so that a sensor that captured 8 stops of dynamic range did it in 4 bits. Niether would be very useful, but that is not the issue.
What the bit depth does is give you more steps between the brightest and darkest areas that the sensor captures. The ideal case would be a sensor that captures a large dynamic range *and* a high bit depth to give very fine differences between tones.
As you point out, but for the wrong reasons, we currently do not have a sensor which is capable of capturing the dynamic range which the human eye can take in. If we ever do, we will probably have to go to 24 bit or 32 bit color to avoid posterization, but there is no technical reason for going to this bit depth. It would be an aesthetic reason.
Posted by Dan Borzynski on April 30, 2009 at 12:11 PM
This might clarify the relationship between dynamic range and bit depth (the above may confuse the issue)
High bit depth, high dynamic range: A Fahrenheit themometer that reads between 0 and the boiling point of water.
Low bit depth, high dynamic range: A Celsius thermometer that reads between 0 and the boiling point of water.
High bit depth and low dynamic range: A Fahrenheit thermometer that reads between 0 and 100.
Low bit depth and low dynamic range: A Celsius thermometer that reads between 0 and 38.
Note that the scale is independent of the device. This is the way to think about sensor dynamic range and the bit depth. Mathematically, the problems are identically and if thermometers were being invented today, the practical issues would be the same as well (how do we build a useful thermometer that reads to 400 degrees? or to -40 degrees).
Right now we have the third case, and we want the first case.
Posted by David Bostedo on April 30, 2009 at 12:45 PM
Dan and Guy are both right in theory, but in practice, sensors usually do show a relationship between bit depth and dynamic range. This is because of the fact that they are linear devices, and the fact that they are noise limited. So for a given pixel :
-- Double the amount of light doubles the value read from the pixel. So in binary 1111 is double 111 - you've added one bit. This is equivalent to part of the scene having twice as much light reflecting into the camera. So if you assume a 10 stop scene, the lowest and highest values will cover about 10 bits worth of data; 1 stop per bit.
-- Now you could divide these stops in smaller and smaller steps. This would add extra bits arbitrarily to the lower end, and you would get finer grained numbers. But these finer grained numbers would be random unless the noise inherent in the value was smaller that the finest step you read off the pixel. Since the lower end values are already your darkest exposure and are noise limited, going finer than that single bit will produce random values.
These things are also the reason a camera like the D3 still can't really use 14-bits - it would need to be able to capture 14 stops in a single shot. And the noise won't let it do that.
I don't know that I've explained all that well, so let me know if I've just confused anyone...I can try again.
If you really want to know what the brightness range in your scene is you will need to use a very accurate spotmeter, such as those manufactured by Pentax, Minolta (used) or Gossen, Spectra (new). These will allow you to sample reflected light with a 1 degree (more or less) angle of view, and allow you to make intelligent decisions about your exposure. I used to own a Minolta and found it to be amazingly sensitive, accurate to .3 f stop or better.
The question is how to you interpret the information and apply it. This depends on your intent and the scene before you. Knowing your camera's dynamic range when shooting raw is key. If you know ahead of time that your gear provides 8 stops of dynamic range between blown highlights and blacked-out shadows, you can set your exposure somewhere between those two extremes the majority of the time.
Specular highlights, such as sunlight reflections on water will confuse many in-camera metering systems into exposing to keep the highlights from being overexposed. Creative intent may drive you to expose to keep the dark side intact, preserving shadows and avoiding silhouettes, or you might want to do the opposite.
A thorough understanding of the Zone System, an excellent scene brightness evaluation tool, will be key here. A light meter will always measure light and give you a series of combinations of shutter speed and lens openings to expose whatever it was that you measured to appear as middle gray. The Zone System will allow you to interpret what the meter tells you allowing you to make precise adjustments in your exposures. Here is an example of how it works.
You are going to measure your high, low and middle parts of a scene, then extrapolate to your camera. For instance, if you find your scene has a brightness range of 14 stops and the highlights are important to you and your camera will only record 8 stops (zone 1 through 8), you will need to set your camera to expose for those highlights without blowing them out. You do this by measuring the brightest part of your scene in which you wish to retain detail. Let's say that you get a reading from your spotmeter that tells you that you should use F16 and 1/1000 for that highlight. However, that reading will ensure that your highlight will be rendered as middle gray - right in the middle between zone 4 and 5. If you wish to record that highlight as bright as it appears before your eyes, then you have some choices to make. The easiest is to increase the exposure by 3-4 stops. If you have important information in the shadow areas, then you should consider the HDR option, with 5 exposures or more at least 1 stop apart. Five exposures will give you a dynamic range expansion to 13 stops, more or less. You can then manually merge these or use merge to HDR in Photoshop, or a third party application such as Photomatix. Using automated HDR processes can result in higher noise levels with the current technology, but the software is getting better in this respect.
As far as using RAW is concerned, it is the best way to start out with the most information in your scene with a single exposure. In camera processing to produce a compressed jpg image will most certainly result in colors and highlights/shadows being clipped in those extreme circumstances where brightness range > sensor capability. Using RAW will give you possibly one more stop headroom before clipping hightlights, shadows and individual colors, giving you a little more "adjustability" in post-processing.
Posted by Dan Borzynski on April 30, 2009 at 1:44 PM
David, I understand where you are coming from and in terms of the real world, I think you are right.
My concern in reading the original post and having seen very muddled thinking elsewhere was a possible belief that just because you may record as a 8, 12, or whatever bit, it means that you have a maximum of x stops of dynamic range. Theoretically, bit depth is unrelated to dynamic range but in the real world, it is at least loosely tied by the stop per bit that you mention.
The simple answer is: you should bracket whenever the range is high. In this age of cheap memory and powerful post processing software, why not? When interesting light happens on an interesting scene, and more importantly you are there to capture it, I really dont see why you should not bracket for insurance
Posted by Patrick Nixon on April 30, 2009 at 4:19 PM
Yep, my original post assumed that the sensor had a degree of latitude as wide as that of the format of which the camera is capable. Maybe not, but it seems restricting to put into the camera a sensor capable of recording only a luminance range of 9 stops (1024 values between black and white) and then record the information in a format capable of expressing a wider range (14-bit RAW, 16384 values between black and white). All this would do is expand the gradation between darkest and lightest pixels captured into finer divisions. There is an awful lot of hype these days about RAW bit-depth and HDR, Nikon's inclusion of 14-bit in addition to 12-bit RAW in its cameras being an example -- although most books and photo forums say that 99 of 100 people are incapable of distinguishing any significant difference between properly exposed 12 and 14-bit RAW images. It is this latter fact suggesting "techie" overkill, that led to my original post.
The zone system has been around a long time. Isn't there someone out there (perhaps a camera manufacturer) who has done the research to record the true luminance difference (in EV or stops) between darkest and lightest in typical scenes? If that difference is often above what the sensor can truly record, use of HDR to more realistically portray what the eye originally saw would be warranted. If not, single-exposure capture in JPEG or 12-bit RAW would most often be good enough.
All this being said, it seems hard to deny the stunning improvements one sees in photo books comparing single shot and multiple-exposure posy-processed HDR images. Or to deny the comparisons of JPEG and RAW images in other books.
Posted by David Bostedo on April 30, 2009 at 5:33 PM
A couple notes Patrick :
...a format capable of expressing a wider range (14-bit RAW, 16384 values between black and white)
Well, the way it ends up working if you push to the noise floor is that 14-bits can cover more stops. But if you had no noise floor, you could use 14-bits to express the exact same 9-stop data with finer gradations. This isn't usually done because the finer gradations become random; But with better noise control, you could get useful data in terms of fine details and smooth gradations by using 14-bits to cover only 9 stops. We're a ways from having a sensor that's that low in noise though.
All this would do is expand the gradation between darkest and lightest pixels captured into finer divisions. There is an awful lot of hype these days about RAW bit-depth and HDR, Nikon's inclusion of 14-bit in addition to 12-bit RAW in its cameras being an example...
Finer divisions can be very useful in post processing - mainly for avoiding posterization. That you can definitely see in prints. And in subtle smooth gradients when taking a photo, higher bit depth could avoid posterization within the RAW file itself, and capture finer detail (if it's very subtle). Depending on the print and size, you could definitely see it. I'm not sure how common it would be however.
Nikon's and Canon's use of 14-bit is pretty specious though. There are very few cases where the extra gradations don't show up purely as randomness. That's because of what I said before about the noise being greater than the gradations. Here's a terrific paper you might like if you've never seen it :
It's by Emil Martinec, U. of Chicago physicist and avid nature photographer.
...the true luminance difference (in EV or stops) between darkest and lightest in typical scenes...
Your histogram will largely tell you if the scene is going beyond the capabilities of your sensor. Assuming that the camera handles it perfectly (and none do), If a scene with optimum exposure and white balance fits entirely within the histogram, the scene contains less stops than the camera can record. Otherwise, it's beyond the sensor's range. So there are lots of scenes that are beyond most sensor's range.
The individual color histogram will tell you if the image you just recorded has individual colors are being clipped as well. In camera processing, required for jpg output, can and often does "pop" color saturation to the brink of clipping. Shooting RAW eliminates the jpg effect (but not necessarily the clipping if it is really bad), and allows you greater precision as far as saturation is concerned.
No it's not
111 binary is 7 decimal. Double it and you get 14, which in binary is 1110.
Guy
Posted by David Bostedo on April 30, 2009 at 9:39 PM
111 binary is 7 decimal. Double it and you get 14, which in binary is 1110.
Oops...rounding error. I should have known better (and used 100 and 1000!) Thanks Guy - my binary math, much like my photography, leaves a lot to be desired.
I've seen lots of answers, but no real simple ones concerning the RAW vs TIFF vs JPG. How are you going to use your images once you have them? How many are you going to shoot? Are you going to attempt to process 500 images a pop taken in RAW? You better have a lot of time on your hands for post-processing (if that is what you need to do for your work, then that is all good - if not, why are you wasting so much time?). Are you a professional or amateur? Does that even matter depending on the number of images you will take in a pop and what you will be doing with them later? How much time do you have for downloads? How up to date is your computer and software? Do you want 15 minute downloads or an hour of downloads? As far as the visual differences depending on your output, I dare you to figure out how the image was originally captured - depending on output I said. Will an original RAW file outsell the same image taken in JPG when both have been post processed by someone who knows what they are doing for both cases? Oh yeah, let''s not forget about what you are shooting to begin with - images for sale to museums or snapshots of the kids at play or something in-between. Even then the argument remains. There is too much math going on in the original question. Are we photographers or mathematicians? Unless you have an awful good reason for RAW and are trying to ink out every bit (pun) out of an HDR multiple exposure combination, you just might not need RAW. Horrors I am sure! Everything has its time and place. I can take a Concorde from here to France (well used to) and get there super fast. I could take a normal jet and get there, too. I still arrive at the same destination; just a little bit differently. Some people might be able to tell how I got to France. Most will not.
JPEGs definitely have their place. I will use them for shooting family and in situations where I know that I won't be needing to get the most out of an image - the latter when marketing the image won't be an issue.
For 99% of my workflow, though, it is RAW all the way. I view most of my shooting as an opportunity to produce potentially marketable images and RAW is the only good way to do this.
And yes, I am a mathematician (Ph.D and all that fun) so I kind of drool over these types of things, but I do understand where you are coming from. Film and 4x5 is the ultimate anti-digital, has a lot of appeal and I shoot it to get away from the computer.
I can take a Concorde from here to France (well used to) and get there super fast. I could take a normal jet and get there, too. I still arrive at the same destination; just a little bit differently. Some people might be able to tell how I got to France. Most will not.
I would like to thank everyone for the vigorous response to this post. This has certainly been a learning exercise for me. Now I proceed to summarize what I believe I have learned from your responses and also responses from a similar original post on the Landscape forum at Nikonians. Hopefully, this summary will prove helpful (and not misleading) to others.
First, let’s take a look at the human eye. The line below represents the various degrees of tonality that can be perceived by the human eye. Various references put this at about 1 million gradations or intervals. (The eye doesn’t really see in intervals but adjusts smoothly over the 1,000,000:1 range. But I have chosen to express the range as intervals because we are talking about transition from the natural to digital world.) At any given point in time, the eye does not see across the entire spectrum because (1) the scene itself does not contain all 1 million intervals or (2) the eye is adapted by adjustment of the pupil to see only a narrower range of gradations.
Scene 1 Scene 2 Scene 3
-------- -------- -------
---------------------------------------------------------------------
Darkest the eye can ever see Brightest the eye can ever see
Now let’s move on to what a digital sensor can record. Within any single scene there is also a range of tonalities. There is a level of dark tonality at which electronic noise overwhelms and becomes indistinguishable from the actual signal from the sensel (individual pixel within a sensor). There is also a level of bright tonality at which the electrical charge reaches the maximum capacity that can be stored. These factors determine the tonal range of the sensor. Larger sensels (pixels) can store larger amounts of electrical energy, and thus sensors containing such sensels have a wider tonal range.
APS-C Sensor
-------------------------
Full Frame Sensor
------------------------------------------
---------------------------------------------------------------------
Darkest tonality of the scene Brightest tonality of the scene
Now let’s move on to actually capturing the scene, which introduces the key concept of exposure. As I understand it, what the camera actually does is shift the recordable range of tonalities for different exposures (by adding or subtracting base voltages to the signals from the sensels). The ideal, of course, would be an exposure that centers the recordable range of tonality upon the center of the tonality of the scene (or the tonality of the key subject within the scene, if any). There comes a point where exposure can be shifted so far in one direction or the other that clipping of the actual tonalities of the scene occurs,
Ideal Exposure
--------------
Under Exposure Over Exposure
-------------- --------------
Extreme Under Exposure Extreme Over Exposure
(Clipping) (Clipping)
-------------------- -------------------
Range of Sensor
-----------------------------------------------
---------------------------------------------------------------------Darkest tonality of the scene Brightest tonality of the scene
There are, of course, scenes and light conditions under which the actual range of tonality falls wholly within the capability of the sensor. But the responses and comments suggest that these are rare, or at least uncommon.
Now let’s move on to what happens AFTER the image is captured by the sensor itself. This is where compression and image format become important.. First, it’s important to understand that all the information captured by the sensor may not be recorded by the camera. The manufacturer’s firmware may, in the interest of conserving memory (card) space, truncate the information recorded. For example, the software of my former Nikon D70s truncated some of the brighter tonalities because, although they could be captured by the sensor, differences between them cannot be distinguished by the human eye. (I’m not sure whether this also occurs in my current D300.)
Finally, let’s consider the concept of compression. Many commenters emphasized that the belief that the degree of compression (JPEG, 12-bit RAW, 14-bit RAW, etc.) determines the limits of the range of tonality that will be recorded is a MYTH. As I now understand it, all that higher bit depth does is to subdivide the range of tonality that is recorded into finer increments, thus allowing expression and discrimination of finer variations in tonality (provided the final means of expression such as computer monitor, paper print, etc. can do so).
For example, suppose a rock in a landscape exhibits a range of tonalities represented by the numbers 3028 to 4028, 100 different tonalities (able to be captured by the sensor, and later separately recorded in 12-bit RAW). An 8-bit depth (JPEG) might cause those 100 tonalities to be compressed (and recorded) to something smaller, say 6 tonalities. If this were to occur, the range of tonalities (and probably color) portraying the rock would be narrowed, and detail would be lost. The upper and lower tonalities recorded are determined by the capabilities of the SENSOR, not by the format. All the format determines is the number of increments between them.
Now back to the point of my original post. I had been concerned that the value of higher-bit formats and multiple-exposure post-processing to produce higher dynamic range (HDR) images was being exaggerated because the actual dynamic range of most scenes was within that of most camera sensors, even with JPEG format. I have become convinced of the opposite, that most cameras for most scenes record (in a single exposure) only a small fraction of the true dynamic range that the eye sees. While I still believe that the merits of RAW and HDR post-processing can be exaggerated, I believe that there is more value in those formats and techniques than I had at first realized.
One more important point. Several commenters suggested that I just simply rely upon the histogram on the back of my camera to determine whether clipping occurred and adjust exposure accordingly. The histogram displayed on the back of cameras and in photo-editing software such as Photoshop corresponds to 256 intervals of tonality between black and white, i.e. the JPEG format. If the histogram shifts too far to the left or right, true clipping of what was seen has occurred. If the histogram appears between black and white, it portrays tonalities of the scene in increments of 1/256. At this degree of crudeness, there may be a slight degree of imprecision at the left and right edges of the histogram. Had the histogram been based upon 4096 increments, there would be a smoother transition between nothing recorded and the first recorded tonality (or between the last recorded tonality and zero).
Whether all this matters depends upon whether one seeks to exactly and precisely record, or instead only produce a reasonable representation, of what the eye sees in the scene. Sometimes a reasonable impression is good enough. Think of the wonderful paintings of the Impressionist artists. And we commonly consider prints from film good enough although it is indisputable as a technical matter that they do not completely accurately portray what the eye saw. On the other hand, advances in photography and computer science are steadily moving in the direction of more realistic, often more impressive, and accurate photographs. Thanks to all for your comments and help.
NOTE: Upon viewing this post on-screen, I see that the horizontal lines do not display as typed in the Comment/Reply dialog box. If desired, send me an e-mail and I will send you a copy in Microsoft Word .doc format.
Finally, let’s consider the concept of compression. Many commenters emphasized that the belief that the degree of compression (JPEG, 12-bit RAW, 14-bit RAW, etc.) determines the limits of the range of tonality that will be recorded is a MYTH. As I now understand it, all that higher bit depth does is to subdivide the range of tonality that is recorded into finer increments, thus allowing expression and discrimination of finer variations in tonality (provided the final means of expression such as computer monitor, paper print, etc. can do so).
I think you're inaccurately conflating RAW and JPG Patrick. They aren't simply different file formats, and they aren't only different in terms of compression. RAW actually is just a data storage file, not an image file like JPG, in that it doesn't inherently provide a means for telling you computer what to display. It's just data from the sensor, which is why RAW developing programs exists.
A JPG, on the other hand, contains pixel information with everything already embedded for display. It simply needs to be read (which does involve some algorithms) and output to the screen. If you were to output a RAW directly to the screen in the same way, you'd probably see the bayer array and a very flat, linear image.
Now when that translation into a JPG happens, either in camera, or in a RAW developer, the steps taken could certainly (and often do) lead to clipping in an effort to get better contrast. So while it might be possible to show the same DR in a JPG, you usually don't find that to be the case in practice.
And as for compression, JPG is always a lossy compression format. RAW files aren't always compressed, and when they are, almost always use a lossless compression algorithm, so that they can be uncompressed precisely.
A lot of time, JPG and RAW are interchangable as "image formats". But when discussing these kinds of details, the differences need to be taken into account.
Had the histogram been based upon 4096 increments, there would be a smoother transition between nothing recorded and the first recorded tonality (or between the last recorded tonality and zero).
The histogram is definitely not perfect, and may show some clipping when there isn't any or vice-versa. There are any number of things wrong with it, because it has to do some sort of conversion of the RAW data to really figure out what the histogram should be. That's why individual color channels and white balance matter.
I think you'll find that the further you push into all this, the more you'll realize that despite being very similar, there are major difference between the eye and a camera, and a real scene and a print. Getting the two to even a reasonable approximation is really doing an amazing thing.
So all of that being said, I don't know that there's anything too useful about knowing all this - in terms of taking better photos. That's almost secondary. (I just enjoy this stuff myself, but I don't think it helps much in terms of improving my photography - but it does make this hobby more interesting.)
Even if the dynamic range of the sensor was the same as the human eye, I would still say that it would be a good idea to shoot raw and bracket exposures so you don't have a botched photo in the case of human error or camera error in metering the exposure. Sometimes it is difficult to see if you exposed it correctly on the LCD screen in bright sunlight and some cameras LCD screens are not accurate. When I get my exposures on to the computer they always look darker than they did on the 40d's screen.
Discussion
18 repliesUnder the best conditions, the most healthy and perceptive human eye can distinguish about a million (1048576=2 to the 20th or 20 EV or stops) different levels of luminance – and that requires time for adaptation to light or darkness and may vary depending upon individual sensitivity to color. The comparable value for a quick glance under bright sunlight is 131,072=2 to the 17th or 17 EV or stops.
The JPEG format is equivalent to 256=2 to the 8th or 7 EV (plus/minus 3 from pattern-metered exposure) EV or stops. At best, 14-bit RAW=16,384=2 to the 14th or 13 (plus/minus 6) EV or stops. The most common 12-bit RAW=4096=2 to the 12th or 11 (plus/minus 5) EV or stops. Theoretically the 14-RAW, 12-RAW and JPEG formats can record only 1/8th, 1/32nd and 1/512th of what the eye can typically see in bright sunlight, respectively. Proving that even the best cameras can record only a small percentage of what the eye can see or distinguish.
The real question seems to be: How often do actual scenes exhibit more than the plus/minus 3 stops (7 stop range) of luminance we have come to accept with the JPEG format. Or the plus/minus 5 stop range equivalent to the 12-bit RAW format. What kinds of scenes typically exhibit a greater range? Sunsets? Green mountains or prairies against blue sky at mid-day? Or sandy beaches or snow against blue sky?
If the answer is “Not often.” there would seem to be little need to resort to the RAW format or HDR post-processing of bracketed exposures. Yet we have all seen stunning improvements using RAW or HDR in photo books.
What say you all? What kinds of scenes or light conditions typically demand bracketing, RAW, or HDR? Does anyone know of an authoritative reference setting forth the TYPICAL luminance range for different scenes or light conditions?
File formats and bit depths are not related to dynamic range. Dynamic range is the number of EVs (aka "stops") between the darkest and lightest values the sensor is capable of detecting detail in. The bit depth represents the number of discreet tones within that range.
Beyond that - detail tapers off towards the extremes so within the dynamic range there's only a portion that is usable (i.e. has enough detail to sufficiently separate visual elements).
Then there is noise. The sensor measures the amount of photons reflected towards it. If there are not enough of them, the measurement will be inaccurate and errors will show as noise in dark areas of the frames. Even if there's sufficient dynamic range, it may still make sense to combine exposures to gather more accurate data from the shadows.
At the end it's not just about the sensor but also depends on the specific image - how important extreme shadow/highlight detail is to its success. In many cases the dynamic range in a single exposure is more than sufficient, but not always.
Remember many of the most memorable photographs made over the years used films with usable dynamic range as narrow as 4 stops. Modern DSLRs are generally capable of about twice that.
Guy
"The JPEG format is equivalent to 256=2 to the 8th or 7 EV (plus/minus 3 from pattern-metered exposure) EV or stops. At best, 14-bit RAW=16,384=2 to the 14th or 13 (plus/minus 6) EV or stops. The most common 12-bit RAW=4096=2 to the 12th or 11 (plus/minus 5) EV or stops. Theoretically the 14-RAW, 12-RAW and JPEG formats can record only 1/8th, 1/32nd and 1/512th of what the eye can typically see in bright sunlight, respectively. Proving that even the best cameras can record only a small percentage of what the eye can see or distinguish. "
No.
The issue is sensor and electronics limitations, not file format limitations. You could build a 16 bit sensor that captured one stop of dynamic range. What you would wind up with is is 2^15 different levels of sensitivity in that one stop of range. You could also set things up so that a sensor that captured 8 stops of dynamic range did it in 4 bits. Niether would be very useful, but that is not the issue.
What the bit depth does is give you more steps between the brightest and darkest areas that the sensor captures. The ideal case would be a sensor that captures a large dynamic range *and* a high bit depth to give very fine differences between tones.
As you point out, but for the wrong reasons, we currently do not have a sensor which is capable of capturing the dynamic range which the human eye can take in. If we ever do, we will probably have to go to 24 bit or 32 bit color to avoid posterization, but there is no technical reason for going to this bit depth. It would be an aesthetic reason.
High bit depth, high dynamic range: A Fahrenheit themometer that reads between 0 and the boiling point of water.
Low bit depth, high dynamic range: A Celsius thermometer that reads between 0 and the boiling point of water.
High bit depth and low dynamic range: A Fahrenheit thermometer that reads between 0 and 100.
Low bit depth and low dynamic range: A Celsius thermometer that reads between 0 and 38.
Note that the scale is independent of the device. This is the way to think about sensor dynamic range and the bit depth. Mathematically, the problems are identically and if thermometers were being invented today, the practical issues would be the same as well (how do we build a useful thermometer that reads to 400 degrees? or to -40 degrees).
Right now we have the third case, and we want the first case.
-- Double the amount of light doubles the value read from the pixel. So in binary 1111 is double 111 - you've added one bit. This is equivalent to part of the scene having twice as much light reflecting into the camera. So if you assume a 10 stop scene, the lowest and highest values will cover about 10 bits worth of data; 1 stop per bit.
-- Now you could divide these stops in smaller and smaller steps. This would add extra bits arbitrarily to the lower end, and you would get finer grained numbers. But these finer grained numbers would be random unless the noise inherent in the value was smaller that the finest step you read off the pixel. Since the lower end values are already your darkest exposure and are noise limited, going finer than that single bit will produce random values.
These things are also the reason a camera like the D3 still can't really use 14-bits - it would need to be able to capture 14 stops in a single shot. And the noise won't let it do that.
I don't know that I've explained all that well, so let me know if I've just confused anyone...I can try again.
The question is how to you interpret the information and apply it. This depends on your intent and the scene before you. Knowing your camera's dynamic range when shooting raw is key. If you know ahead of time that your gear provides 8 stops of dynamic range between blown highlights and blacked-out shadows, you can set your exposure somewhere between those two extremes the majority of the time.
Specular highlights, such as sunlight reflections on water will confuse many in-camera metering systems into exposing to keep the highlights from being overexposed. Creative intent may drive you to expose to keep the dark side intact, preserving shadows and avoiding silhouettes, or you might want to do the opposite.
A thorough understanding of the Zone System, an excellent scene brightness evaluation tool, will be key here. A light meter will always measure light and give you a series of combinations of shutter speed and lens openings to expose whatever it was that you measured to appear as middle gray. The Zone System will allow you to interpret what the meter tells you allowing you to make precise adjustments in your exposures. Here is an example of how it works.
You are going to measure your high, low and middle parts of a scene, then extrapolate to your camera. For instance, if you find your scene has a brightness range of 14 stops and the highlights are important to you and your camera will only record 8 stops (zone 1 through 8), you will need to set your camera to expose for those highlights without blowing them out. You do this by measuring the brightest part of your scene in which you wish to retain detail. Let's say that you get a reading from your spotmeter that tells you that you should use F16 and 1/1000 for that highlight. However, that reading will ensure that your highlight will be rendered as middle gray - right in the middle between zone 4 and 5. If you wish to record that highlight as bright as it appears before your eyes, then you have some choices to make. The easiest is to increase the exposure by 3-4 stops. If you have important information in the shadow areas, then you should consider the HDR option, with 5 exposures or more at least 1 stop apart. Five exposures will give you a dynamic range expansion to 13 stops, more or less. You can then manually merge these or use merge to HDR in Photoshop, or a third party application such as Photomatix. Using automated HDR processes can result in higher noise levels with the current technology, but the software is getting better in this respect.
As far as using RAW is concerned, it is the best way to start out with the most information in your scene with a single exposure. In camera processing to produce a compressed jpg image will most certainly result in colors and highlights/shadows being clipped in those extreme circumstances where brightness range > sensor capability. Using RAW will give you possibly one more stop headroom before clipping hightlights, shadows and individual colors, giving you a little more "adjustability" in post-processing.
My concern in reading the original post and having seen very muddled thinking elsewhere was a possible belief that just because you may record as a 8, 12, or whatever bit, it means that you have a maximum of x stops of dynamic range. Theoretically, bit depth is unrelated to dynamic range but in the real world, it is at least loosely tied by the stop per bit that you mention.
The zone system has been around a long time. Isn't there someone out there (perhaps a camera manufacturer) who has done the research to record the true luminance difference (in EV or stops) between darkest and lightest in typical scenes? If that difference is often above what the sensor can truly record, use of HDR to more realistically portray what the eye originally saw would be warranted. If not, single-exposure capture in JPEG or 12-bit RAW would most often be good enough.
All this being said, it seems hard to deny the stunning improvements one sees in photo books comparing single shot and multiple-exposure posy-processed HDR images. Or to deny the comparisons of JPEG and RAW images in other books.
Well, the way it ends up working if you push to the noise floor is that 14-bits can cover more stops. But if you had no noise floor, you could use 14-bits to express the exact same 9-stop data with finer gradations. This isn't usually done because the finer gradations become random; But with better noise control, you could get useful data in terms of fine details and smooth gradations by using 14-bits to cover only 9 stops. We're a ways from having a sensor that's that low in noise though.
Finer divisions can be very useful in post processing - mainly for avoiding posterization. That you can definitely see in prints. And in subtle smooth gradients when taking a photo, higher bit depth could avoid posterization within the RAW file itself, and capture finer detail (if it's very subtle). Depending on the print and size, you could definitely see it. I'm not sure how common it would be however.
Nikon's and Canon's use of 14-bit is pretty specious though. There are very few cases where the extra gradations don't show up purely as randomness. That's because of what I said before about the noise being greater than the gradations. Here's a terrific paper you might like if you've never seen it :
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/
It's by Emil Martinec, U. of Chicago physicist and avid nature photographer.
Your histogram will largely tell you if the scene is going beyond the capabilities of your sensor. Assuming that the camera handles it perfectly (and none do), If a scene with optimum exposure and white balance fits entirely within the histogram, the scene contains less stops than the camera can record. Otherwise, it's beyond the sensor's range. So there are lots of scenes that are beyond most sensor's range.
No it's not
111 binary is 7 decimal. Double it and you get 14, which in binary is 1110.
Guy
Oops...rounding error.
Good luck and get out there and shoot something.
Troup
For 99% of my workflow, though, it is RAW all the way. I view most of my shooting as an opportunity to produce potentially marketable images and RAW is the only good way to do this.
And yes, I am a mathematician (Ph.D and all that fun) so I kind of drool over these types of things, but I do understand where you are coming from. Film and 4x5 is the ultimate anti-digital, has a lot of appeal and I shoot it to get away from the computer.
Excellent analogy!
First, let’s take a look at the human eye. The line below represents the various degrees of tonality that can be perceived by the human eye. Various references put this at about 1 million gradations or intervals. (The eye doesn’t really see in intervals but adjusts smoothly over the 1,000,000:1 range. But I have chosen to express the range as intervals because we are talking about transition from the natural to digital world.) At any given point in time, the eye does not see across the entire spectrum because (1) the scene itself does not contain all 1 million intervals or (2) the eye is adapted by adjustment of the pupil to see only a narrower range of gradations.
Scene 1 Scene 2 Scene 3
-------- -------- -------
---------------------------------------------------------------------
Darkest the eye can ever see Brightest the eye can ever see
Now let’s move on to what a digital sensor can record. Within any single scene there is also a range of tonalities. There is a level of dark tonality at which electronic noise overwhelms and becomes indistinguishable from the actual signal from the sensel (individual pixel within a sensor). There is also a level of bright tonality at which the electrical charge reaches the maximum capacity that can be stored. These factors determine the tonal range of the sensor. Larger sensels (pixels) can store larger amounts of electrical energy, and thus sensors containing such sensels have a wider tonal range.
APS-C Sensor
-------------------------
Full Frame Sensor
------------------------------------------
---------------------------------------------------------------------
Darkest tonality of the scene Brightest tonality of the scene
Now let’s move on to actually capturing the scene, which introduces the key concept of exposure. As I understand it, what the camera actually does is shift the recordable range of tonalities for different exposures (by adding or subtracting base voltages to the signals from the sensels). The ideal, of course, would be an exposure that centers the recordable range of tonality upon the center of the tonality of the scene (or the tonality of the key subject within the scene, if any). There comes a point where exposure can be shifted so far in one direction or the other that clipping of the actual tonalities of the scene occurs,
Ideal Exposure
--------------
Under Exposure Over Exposure
-------------- --------------
Extreme Under Exposure Extreme Over Exposure
(Clipping) (Clipping)
-------------------- -------------------
Range of Sensor
-----------------------------------------------
---------------------------------------------------------------------Darkest tonality of the scene Brightest tonality of the scene
There are, of course, scenes and light conditions under which the actual range of tonality falls wholly within the capability of the sensor. But the responses and comments suggest that these are rare, or at least uncommon.
Now let’s move on to what happens AFTER the image is captured by the sensor itself. This is where compression and image format become important.. First, it’s important to understand that all the information captured by the sensor may not be recorded by the camera. The manufacturer’s firmware may, in the interest of conserving memory (card) space, truncate the information recorded. For example, the software of my former Nikon D70s truncated some of the brighter tonalities because, although they could be captured by the sensor, differences between them cannot be distinguished by the human eye. (I’m not sure whether this also occurs in my current D300.)
Finally, let’s consider the concept of compression. Many commenters emphasized that the belief that the degree of compression (JPEG, 12-bit RAW, 14-bit RAW, etc.) determines the limits of the range of tonality that will be recorded is a MYTH. As I now understand it, all that higher bit depth does is to subdivide the range of tonality that is recorded into finer increments, thus allowing expression and discrimination of finer variations in tonality (provided the final means of expression such as computer monitor, paper print, etc. can do so).
For example, suppose a rock in a landscape exhibits a range of tonalities represented by the numbers 3028 to 4028, 100 different tonalities (able to be captured by the sensor, and later separately recorded in 12-bit RAW). An 8-bit depth (JPEG) might cause those 100 tonalities to be compressed (and recorded) to something smaller, say 6 tonalities. If this were to occur, the range of tonalities (and probably color) portraying the rock would be narrowed, and detail would be lost. The upper and lower tonalities recorded are determined by the capabilities of the SENSOR, not by the format. All the format determines is the number of increments between them.
Now back to the point of my original post. I had been concerned that the value of higher-bit formats and multiple-exposure post-processing to produce higher dynamic range (HDR) images was being exaggerated because the actual dynamic range of most scenes was within that of most camera sensors, even with JPEG format. I have become convinced of the opposite, that most cameras for most scenes record (in a single exposure) only a small fraction of the true dynamic range that the eye sees. While I still believe that the merits of RAW and HDR post-processing can be exaggerated, I believe that there is more value in those formats and techniques than I had at first realized.
One more important point. Several commenters suggested that I just simply rely upon the histogram on the back of my camera to determine whether clipping occurred and adjust exposure accordingly. The histogram displayed on the back of cameras and in photo-editing software such as Photoshop corresponds to 256 intervals of tonality between black and white, i.e. the JPEG format. If the histogram shifts too far to the left or right, true clipping of what was seen has occurred. If the histogram appears between black and white, it portrays tonalities of the scene in increments of 1/256. At this degree of crudeness, there may be a slight degree of imprecision at the left and right edges of the histogram. Had the histogram been based upon 4096 increments, there would be a smoother transition between nothing recorded and the first recorded tonality (or between the last recorded tonality and zero).
Whether all this matters depends upon whether one seeks to exactly and precisely record, or instead only produce a reasonable representation, of what the eye sees in the scene. Sometimes a reasonable impression is good enough. Think of the wonderful paintings of the Impressionist artists. And we commonly consider prints from film good enough although it is indisputable as a technical matter that they do not completely accurately portray what the eye saw. On the other hand, advances in photography and computer science are steadily moving in the direction of more realistic, often more impressive, and accurate photographs. Thanks to all for your comments and help.
NOTE: Upon viewing this post on-screen, I see that the horizontal lines do not display as typed in the Comment/Reply dialog box. If desired, send me an e-mail and I will send you a copy in Microsoft Word .doc format.
I think you're inaccurately conflating RAW and JPG Patrick. They aren't simply different file formats, and they aren't only different in terms of compression. RAW actually is just a data storage file, not an image file like JPG, in that it doesn't inherently provide a means for telling you computer what to display. It's just data from the sensor, which is why RAW developing programs exists.
A JPG, on the other hand, contains pixel information with everything already embedded for display. It simply needs to be read (which does involve some algorithms) and output to the screen. If you were to output a RAW directly to the screen in the same way, you'd probably see the bayer array and a very flat, linear image.
Now when that translation into a JPG happens, either in camera, or in a RAW developer, the steps taken could certainly (and often do) lead to clipping in an effort to get better contrast. So while it might be possible to show the same DR in a JPG, you usually don't find that to be the case in practice.
And as for compression, JPG is always a lossy compression format. RAW files aren't always compressed, and when they are, almost always use a lossless compression algorithm, so that they can be uncompressed precisely.
A lot of time, JPG and RAW are interchangable as "image formats". But when discussing these kinds of details, the differences need to be taken into account.
The histogram is definitely not perfect, and may show some clipping when there isn't any or vice-versa. There are any number of things wrong with it, because it has to do some sort of conversion of the RAW data to really figure out what the histogram should be. That's why individual color channels and white balance matter.
I think you'll find that the further you push into all this, the more you'll realize that despite being very similar, there are major difference between the eye and a camera, and a real scene and a print. Getting the two to even a reasonable approximation is really doing an amazing thing.
So all of that being said, I don't know that there's anything too useful about knowing all this - in terms of taking better photos. That's almost secondary. (I just enjoy this stuff myself, but I don't think it helps much in terms of improving my photography - but it does make this hobby more interesting.)