When capturing a high-contrast scene by bracketing multiple exposures, I've noticed that in certain situations the exposure for the shadows (+2 stops) has poor detail in the shadows. None of the channels clipped on the shadow end of the histogram. (plenty of clipping on the highlights, which will be blended away.) However, the shadows will be low contrast, low detail and a little noisy even at ISO 100. It's sometimes severe enough that I don't really want to open up the shadows very much since it's not particularly attractive.
I most often run into this situation when shooting just after the sun sets, in the direction of the sunset. I don't have an example RAW on hand -- if necessary I can find one. I'm currently shooting with a Canon 5D, but I had the same issue with my old 350D.
My theory is that this is caused by the lower density of luminosity levels on the shadows end of the sensor. I.e. the exact opposite of "expose to the right" method. If that's the case a 16-bit image processor on the camera might improve the situation, on a 5D mk2 for example. I haven't heard any 5d2 owners say that this issue is dramatically improved on the new system, however.
Does anyone have any recommendations for how to best handle this? Either in the field or in Photoshop. GND filters are an option, but I'd like to be able to manage this digitally as well.
Matt Anderson had a video tutorial he shared with us a while back on increasing dynamic range and using smart objects. It had a step in there where he uses the over exposed file to create a mask for the shadow areas and then decreases the exposure on that file by 2 stops to match the exposure of the properly exposed file. The result is the overexposed file which has been darkened by 2 stops shows through in the shadow areas and the image has a lot less noise in the shadows. Here is a link to his post with the video. http://www.naturephotographers.net/imagecritique/bbs.cgi?a=vm&mr=41538&CGISESSID=8553dfaa3325880ef0120b5c041b7ce6&u=21200
Thanks for the reply, Austin. As a matter of fact I actually found that video while researching this subject last week.
Correct me if I'm wrong, but I don't think it helps. He would take the +2 stop RAW and darken it by 2 stops. Yeah the noise is reduced but the shadows are back to being damn near pure black, and I'm trying to pull out some detail there. I think Anderson's intention is to actually do tonality adjustments within PS, instead of ACR. And when he got to that point he'd then run into the same problem I have (depending on the scene and RAW files of course).
Mat--I just watched his video and he does a raw blend of 3 bracketed images--essentially a manual HDR but also does alot of processing along the way that brings out the details in the dark areas. You might want to take a closer look. I think I'll need to watch it several times jsut to get more than a basic overview of what he really does though as it is a long and complicated process--Larry
-- The noise benefit of exposing to the right has nothing to do with the number of levels (despite anything on Luminous Landscape). It's due to the signal to noise ratio of that part of the sensor. Exposures further to the right maximize the SNR at the sensor. The extra levels available may be a benefit to post processing, but they don't inherently lower the noise in the RAW file.
-- There are two different but related issues here - lowering the noise, and pulling out detail. They can both be helped by exposing for the shadows, so that they are brighter, and have more detail. Then when you bring them back down, you can work on increasing contrast, and bringing down to a level you like. If you have completely black shadows at a normal exposure, and you overexpose them by, say, 4 stops; Nothing says you have to lower them by 4 during RAW conversion. That's only if you want to preserve the original exposure without the noise.
David, thinking about it turns of S/N makes more sense to me than luminosity levels. I'll have to try exposing +4 stops the next time I encounter this type of scene. Unfortunately it's not very convenient since the 5D can only bracket +-2 stops, but it's something to work with.
You can switch the dial to M and calculate the +4 exposure manually (and probably shoot faster than setting up the auto bracketing in the process). The exposure scale will just blink at you, but it will let you do it.
I don't think a 16 bit or 32 bit processor would help at all with the problem you are describing. What is needed is a sensor with a truly expanded dynamic range so that the shadows can get up off the noise floor.
I'm not a big fan of DXO mark. I think the results are somewhat dubious in many areas. But I do think it shows this one thing well.
If you look at what the DXO Mark application is showing for dynamic range (or color spectrum for that matter), it shows that there has been very little improvement in the dynamic range of sensors from the first CMOS sensors (like the 20D and 350D) to the latest CMOS sensors (like the Sony A900 and 5DII). This leads me to believe that the properties of the CMOS sensor are the limiting factor in dynamic range. They get better at handling the noise floor through various software methods, but the EV range the camera can capture is still stuck at X stops.
The CMOS sensors still don't match the usable dynamic range of color negative films but they do exceed the usable dynamic range of most slide films. Digital seems to be stuck here waiting for some hardware breakthrough with HDR software filling in for the need for wider latitude. Of course we don't always want the exposure latitude of color negative film, so when the advancement comes along, I'm not sure how they are going to tackle that problem (possibly with more bit depth in the A/D converter).
The CMOS sensors still don't match the usable dynamic range of color negative films but they do exceed the usable dynamic range of most slide films.
Mel - From what I've read, it's true that there's a lot less exposure latitude with current sensors (because of the lack of roll off in shadows or highlights), but for real captured dynamic range, they beat pretty much every film except maybe some high ISO B&W. (Caveat - I've never shot film.)
And the +4 is just a number I've used before. Depending on the scene, that may also have to be adjusted. I find it easiest to do in manual mode myself. If you want to bracket, 2 stops may still accomplish what you want - depending on what the initial exposure is in the first place.
I shoot pretty freely across digital and color negative and slides and also some traditional and chromogenic B&W. For all of my film photography, I scan the film, so in the end, I'm still doing digital photography. I'm not trying to convert anyone to a color negative scanning workflow. Scanning is time consuming. I'm not coming down on digital either. And I'm definitely not trying to start a war over digital or film (I'm stuck in the middle).
Based on my real world experience, the DR numbers they've stated for DSLRs are either wildly trumped up or they are measuring against a completely different scale or the color negative values have been significantly understated. Digital CMOS DR does not match normal color negative dynamic range right now. Unfortunately, I don't have the step wedges or other tools to demonstrate this quantitatively, but my practical experience has shown this to be true.
Exposure latitude and dynamic range are in effect the same thing even if the meanings are quite different. Exposure latitude is simply a subjective measurement of how much over or underexposure will still result in an acceptable print, dynamic range is a quantitative measurement of light values the media can capture. If a media has a higher exposure latitude, by its nature it must have a higher dynamic range as well if it is to produce an acceptable print.
My point was only that I think digital capture at this time is waiting for the next big thing. Faster processing and more pixels are not fundamentally changing what a digital image looks like. The changes in image quality are iterative, not exponential. I suspect a new technology on the hardware end will be needed to provide this breakthrough, but I don't know what that would be. Right now, the two commercially viable photo sensitive options for a sensor are CCD and CMOS and both seem to offer similar performance. There is talk of an alternative to the Bayer array on one of the mentioned image sensors, still using 4 pixels to generate the color information, but leaving one of them unfiltered with a possible goal being a bump in DR or lower noise in low light.
Thanks Mel - You know, what I'm curious about is given all the DxO Mark, DP Review, PopPhoto type labs out there, why can't I find a nice article that did formal testing and compares the DR of films vs. sensors. You'd think someone would have done it - measured what the media can actually capture (which might be slightly different than what's useful or pleasing). I can't find one though...anyone know of one?
Discussion
9 repliesI most often run into this situation when shooting just after the sun sets, in the direction of the sunset. I don't have an example RAW on hand -- if necessary I can find one. I'm currently shooting with a Canon 5D, but I had the same issue with my old 350D.
My theory is that this is caused by the lower density of luminosity levels on the shadows end of the sensor. I.e. the exact opposite of "expose to the right" method. If that's the case a 16-bit image processor on the camera might improve the situation, on a 5D mk2 for example. I haven't heard any 5d2 owners say that this issue is dramatically improved on the new system, however.
Does anyone have any recommendations for how to best handle this? Either in the field or in Photoshop. GND filters are an option, but I'd like to be able to manage this digitally as well.
Thanks for any help or insight you can provide!
Correct me if I'm wrong, but I don't think it helps. He would take the +2 stop RAW and darken it by 2 stops. Yeah the noise is reduced but the shadows are back to being damn near pure black, and I'm trying to pull out some detail there. I think Anderson's intention is to actually do tonality adjustments within PS, instead of ACR. And when he got to that point he'd then run into the same problem I have (depending on the scene and RAW files of course).
-- The noise benefit of exposing to the right has nothing to do with the number of levels (despite anything on Luminous Landscape). It's due to the signal to noise ratio of that part of the sensor. Exposures further to the right maximize the SNR at the sensor. The extra levels available may be a benefit to post processing, but they don't inherently lower the noise in the RAW file.
-- There are two different but related issues here - lowering the noise, and pulling out detail. They can both be helped by exposing for the shadows, so that they are brighter, and have more detail. Then when you bring them back down, you can work on increasing contrast, and bringing down to a level you like. If you have completely black shadows at a normal exposure, and you overexpose them by, say, 4 stops; Nothing says you have to lower them by 4 during RAW conversion. That's only if you want to preserve the original exposure without the noise.
You can switch the dial to M and calculate the +4 exposure manually (and probably shoot faster than setting up the auto bracketing in the process). The exposure scale will just blink at you, but it will let you do it.
I don't think a 16 bit or 32 bit processor would help at all with the problem you are describing. What is needed is a sensor with a truly expanded dynamic range so that the shadows can get up off the noise floor.
I'm not a big fan of DXO mark. I think the results are somewhat dubious in many areas. But I do think it shows this one thing well.
If you look at what the DXO Mark application is showing for dynamic range (or color spectrum for that matter), it shows that there has been very little improvement in the dynamic range of sensors from the first CMOS sensors (like the 20D and 350D) to the latest CMOS sensors (like the Sony A900 and 5DII). This leads me to believe that the properties of the CMOS sensor are the limiting factor in dynamic range. They get better at handling the noise floor through various software methods, but the EV range the camera can capture is still stuck at X stops.
The CMOS sensors still don't match the usable dynamic range of color negative films but they do exceed the usable dynamic range of most slide films. Digital seems to be stuck here waiting for some hardware breakthrough with HDR software filling in for the need for wider latitude. Of course we don't always want the exposure latitude of color negative film, so when the advancement comes along, I'm not sure how they are going to tackle that problem (possibly with more bit depth in the A/D converter).
Mel - From what I've read, it's true that there's a lot less exposure latitude with current sensors (because of the lack of roll off in shadows or highlights), but for real captured dynamic range, they beat pretty much every film except maybe some high ISO B&W. (Caveat - I've never shot film.)
And the +4 is just a number I've used before. Depending on the scene, that may also have to be adjusted. I find it easiest to do in manual mode myself. If you want to bracket, 2 stops may still accomplish what you want - depending on what the initial exposure is in the first place.
Based on my real world experience, the DR numbers they've stated for DSLRs are either wildly trumped up or they are measuring against a completely different scale or the color negative values have been significantly understated. Digital CMOS DR does not match normal color negative dynamic range right now. Unfortunately, I don't have the step wedges or other tools to demonstrate this quantitatively, but my practical experience has shown this to be true.
Exposure latitude and dynamic range are in effect the same thing even if the meanings are quite different. Exposure latitude is simply a subjective measurement of how much over or underexposure will still result in an acceptable print, dynamic range is a quantitative measurement of light values the media can capture. If a media has a higher exposure latitude, by its nature it must have a higher dynamic range as well if it is to produce an acceptable print.
My point was only that I think digital capture at this time is waiting for the next big thing. Faster processing and more pixels are not fundamentally changing what a digital image looks like. The changes in image quality are iterative, not exponential. I suspect a new technology on the hardware end will be needed to provide this breakthrough, but I don't know what that would be. Right now, the two commercially viable photo sensitive options for a sensor are CCD and CMOS and both seem to offer similar performance. There is talk of an alternative to the Bayer array on one of the mentioned image sensors, still using 4 pixels to generate the color information, but leaving one of them unfiltered with a possible goal being a bump in DR or lower noise in low light.