Which approach is best in low light?

Harveylevine

When shooting in light that is inadequate for a hand-held picture, is it better to increase the ISO so that the picture can be taken at the minimum shutter speed needed to record the picture with relatively little camera shake, or is it better to shoot at a low ISO (but the same shutter speed so that the camera shake is the same as the high ISO image) and use image editing software to increase the exposure of the resulting underexposed shot? Which way produces more noise? Is there too much image degradation caused by the exposure adjustment of the underexposed shot? Will noise reduction software, such as Neat Image, work better on pictures taken using one of these two approaches rather than the other?



I ran a test. I took two pictures of a room lit by existing lamps with my Casio EX-Z750. One was shot at the minimum ISO of 50 and was underexposed by perhaps 2-3 stops. The other was shot at ISO 400 (the maximum). I adjusted the low ISO (underexposed) image in Photoshop by creating a duplicate layer and changing the blend mode to “Screen”. This lightened the image somewhat. I continued to create duplicate layers with “Screen” blend modes until the image was similar to the high ISO (properly exposed) image. I then saved this image.



Next I ran both images through Neat Image to reduce the noise and saved both images. I have uploaded all four images (high/low ISO, with/without noise reduction) to a Smugmug gallery with this link (so that you can examine the full size images):

http://harveylevine.smugmug.com/gallery/887258/1/40177323




My observation is that the noise reduction was quite effective in both cases. When I looked at the histograms in Photoshop, the high ISO images looked somewhat smoother than the low ISO (originally underexposed) images, which displayed more spikes. Does this indicate that the high ISO images are in a better state for additional editing than the low ISO images? Visually I didn’t see a large difference between the high and low ISO images.



A speculation: If I had performed the same test with my Nikon D70, rather than the point and shoot Casio, would the outcome be different? I suspect that the DSLRs display less noise than point and shoot cameras when the ISO is cranked up. So perhaps in this case the noise from the underexposed, low ISO image would be greater that the noise in the high ISO image. What do you think?



Has anyone looked into this question?



Thanks,



Harvey

Shay Stephens

Without a doubt, shooting at higher ISO provides better results. Not only do you get a better exposure, you save all that time and effort spent trying to do the same thing in post. You also get more dynamic range.

By shooting at lower ISO and purposefully underexposing, you risk losing your shadow detail. So get the exposure as close as you can in camera for best results. Don't worry about the noise. I periodically shoot at ISO 3200, and never use noise reduction software aside from chroma noise reduction that is available through the Photoshop RAW converter.

Back in the day, when high ISO film was used in low light, it was understood that there would be heavy grain in the resulting photos. Some would say it even added character to the photos. Today, the noise in digital shots taken at high ISO can be less than with film. So I like to say, embrace the noise, incorporate it, and benefit from higher quality photos ;-)

Harveylevine

Shay Stephens wrote:

Without a doubt, shooting at higher ISO provides better results. Not only do you get a better exposure, you save all that time and effort spent trying to do the same thing in post. You also get more dynamic range.

By shooting at lower ISO and purposefully underexposing, you risk losing your shadow detail. So get the exposure as close as you can in camera for best results. Don't worry about the noise. I periodically shoot at ISO 3200, and never use noise reduction software aside from chroma noise reduction that is available through the Photoshop RAW converter.

Shay:
Thank you for responding with such a clear analysis and recommendation. I suspected that getting the exposure right in the first place was the best course of action, but as I lack your experience as a professional photographer, I couldn't be sure. From now on, I'll adjust the ISO to whatever is required to get good exposure and not get hung up on the question of the resultant noise.

Harvey

Steve Cavigliano

Hiya Harvey


I'm with Shay on this one. Besides the potentail loss of shadow detail, I can tell you from experience that purposely under exposing then pushing during conversion winds up giving you as much, sometimes more, image noise than you would see with a properly exposed pic shot with one stop higher ISO.

I tried this with some indoor action shots and under exposed a stop using ISO800 instead of ISO1600. I switched halfway through the game and when I got home and did comparisons, I found out what I stated above. Noise is always worse in the under exposed areas of a pic. When the whole pic is under exposed and you were using a high ISO to begin with you can see even more noise in areas like skin once the image is pushed during conversion

Steve

Harveylevine

Steve Cavigliano wrote:

I tried this with some indoor action shots and under exposed a stop using ISO800 instead of ISO1600. I switched halfway through the game and when I got home and did comparisons, I found out what I stated above. Noise is always worse in the under exposed areas of a pic. When the whole pic is under exposed and you were using a high ISO to begin with you can see even more noise in areas like skin once the image is pushed during conversion

Steve

Steve:
Thanks for responding. Your observations are very useful since they are based on photographs actually taken under marginal lighting conditions. Between your comments and Shay's, it has become clear that getting the exposure correct in the first place is the highest priority.

I looked at your galleries and particularly liked the sports pictures, which must have been taken under very difficult lighting conditions.

Thanks for you practical advice.
Harvey

pathfinder

Shay's and Steve answers are the traditional, expected answers. And I generally would agree with them. I wonder if they actually took the time to examine the images of the chair in the room lit by tungsten light you posted and compared them. You might post 100% crops of each here to other folks can examine them also.

There is, too my eye, much less noise in the underexposed low ISO image that is then processed and run through Neat Image, than the properly exposed higher ISO image, much to my surprise!! This is a P&S image sensor, not a DSLR's larger sensor, and hence tends to be much noiser at higher ISOs and this was not shot in RAW either.

The shadow detail is as good in the underexposed lower ISO after NI, than the higher ISO images. Look at the chair legs under the desk. The image is slightly less sharp though, and has taken on a green cast that I do not like as well. Interesting experiment that I would not have believed if I had not seen your results in the gallery you linked.

I wonder if this doen't have something to do with the quality of the image handling in your P&S as the ISO is raised. When you raise the ISO in a digital camera, what exactly are you doing?? And HOW EXACTLY does the data get processed. It may be that Neat Image is just a much better algorithm to reduce noise than the algorithm built into the camera. I'd love to hear from some engineers that really know the answer to these questions.

rutt

Hey, I couldn't see this post at all. Please consider editing and removing the white text color tag. That makes it unreadable for those who use the white dgrin theme. Thanks.

Shay Stephens

Some things to consider

There is, too my eye, much less noise in the underexposed low ISO image that is then processed and run through Neat Image, than the properly exposed higher ISO image, much to my surprise!!

The whole reason for being for noise reduction programs is to reduce noise, so it is no surprise that they are quite capable of reducing it to a level that has less noise than images that have not had such a treatment. However, what is missing in such images is a world of detail. All that noise reduction (specifically luminance noise) comes at a prices of fine detail. Instead of a photo, what you wind up with is an impressionistic painting that shows shape and color quite well but not much more.

When using noise reduction programs, typically the image gets sharpened more than a ginsu knife too. So when comparing, pay attention to sharpening artifacts.

Also, when you take an underexposed image and brighten it, you have to work with a compressed histogram, and the results are an image with a smaller dynamic range than a properly exposed image.

All of these things taken individually may not be a problem for a particular image, but considered together and over the course of many images can have a big impact on the quality of the imagery produced by a photographer.

Now I can hear the cries of "elitist" even now hehehe so I will only say this. You have to do what you think looks best. You have to make your own judgements and comparisons. Learn the skills of image analysis for yourself. What one person finds unacceptable, another finds much delight in, this I know well. So if noise reduction programs make an unacceptable image better for you, then you should feel no regret using it.

Andy

rutt wrote:

Hey, I couldn't see this post at all. Please consider editing and removing the white text color tag. That makes it unreadable for those who use the white dgrin theme. Thanks.

i agree that folks shouldn't use white text but geez rutt - y ou can simpy change the th=eme for this one post

Harveylevine

pathfinder wrote:

Shay's and Steve answers are the traditional, expected answers. And I generally would agree with them. I wonder if they actually took the time to examine the images of the chair in the room lit by tungsten light you posted and compared them. You might post 100% crops of each here to other folks can examine them also.

Pathfinder:
Thanks a lot for looking carefully at the pictures in my Smugmug gallery that show the results of my experiment in comparing high iso vs processed low iso images in poor light. I appreciate your observations and will post 100% crops over the weekend as you suggested.

pathfinder wrote:

There is, too my eye, much less noise in the underexposed low ISO image that is then processed and run through Neat Image, than the properly exposed higher ISO image, much to my surprise!! This is a P&S image sensor, not a DSLR's larger sensor, and hence tends to be much noiser at higher ISOs and this was not shot in RAW either.

Right on the mark! Much to my surprise too, the underexposed low ISO image does not show more noise or look substantially worse than the properly exposed high ISO images (especially after both had been run through Neat Image). That is why I decided to post my original question here. The smaller P&S sensor, as you pointed out, probably has a lot more trouble with high ISO images than would a DSLR sensor. My experience with my Nikon D70 is that high ISO raw images only show a small to moderate increase in noise compared to low ISO images. With the P&S camera, the image noise increases much more dramatically as ISO increases. So it would appear that setting the ISO to a high enough level to get a proper exposure is best with a DSLR, but it is not at all clear with a (or at least my) P&S.

pathfinder wrote:

The shadow detail is as good in the underexposed lower ISO after NI, than the higher ISO images. Look at the chair legs under the desk. The image is slightly less sharp though, and has taken on a green cast that I do not like as well. Interesting experiment that I would not have believed if I had not seen your results in the gallery you linked. .

The greenish-yellow cast you observed appears to be the major difference between the high and low ISO images. I stacked three duplicate layers on top of the background image and set the blend mode for each to "screen" so that the exposure was similar to the high ISO one. This processing probably increased the green cast or even created it. A more suble difference is that the histogram for the processed low ISO image shows spikes whereas the high ISO one is smoother. As I speculated in my original post, maybe the low ISO image would not stand up as well to further post processing.

pathfinder wrote:

I wonder if this doen't have something to do with the quality of the image handling in your P&S as the ISO is raised. When you raise the ISO in a digital camera, what exactly are you doing?? And HOW EXACTLY does the data get processed. It may be that Neat Image is just a much better algorithm to reduce noise than the algorithm built into the camera. I'd love to hear from some engineers that really know the answer to these questions.

My experience is that Neat Image does a great job of removing noise without excessively softening the image. Adobe Camera Raw is quite good too (as is Noise Ninja). The post processing noise reduction software is undoubtedly far superior to anything built into cameras. It takes about two minutes to process a 6 or 7 megapixel image with Neat Image, so it is not currently feasible to add hgih-quality noise reduction software to a camera.

Thanks again for taking the trouble to look carefully at my "noise tests". I'll try to find the time this weekend to post 100% crops of sections of those images here. If anyone else has done similar experiments with either a P&S or DSLR, I'd love to see the results.

Harvey

Harveylevine

rutt wrote:

Hey, I couldn't see this post at all. Please consider editing and removing the white text color tag. That makes it unreadable for those who use the white dgrin theme. Thanks.

Rutt:
Sorry, I hadn't realized I somehow changed the text to white! I've edited the initial post and I think I've removed the text tag . Hope you can read it now.
Harvey

Harveylevine

Shay Stephens wrote:

Also, when you take an underexposed image and brighten it, you have to work with a compressed histogram, and the results are an image with a smaller dynamic range than a properly exposed image.

Shay:
I have been thinking about your comments and I finally understand your point and the problem created by a small dynamic range due to underexposure. I've been performing a "thought experiment". Suppose we have an extreme example of an underexposed image. Instead of the usual dynamic range of 0 (black) to 255 (white), we instead have a very underexposed image that only has pixels between 0 - 7. In effect we have a 3 bit image rather than the normal 8 bits.

Now suppose we use some software to make the image brighter. Say this software results in an image that increases the brightness values to 0 -14. What would actually happen to the values of individual pixels? The pixels that are pure black or 0 would still be zero. The ones that were 1 now become 2; the ones that were 2 become 4 and so forth. There are no pixels with brightness of 1,3,5 etc. only 1,2,4,6 ... We don't have the intermediate values in the image; we still only have eight possible values. The histogram would be a series of eight spikes. If we continued to brighten the image using the software, eventually we would spread the histogram to cover the entire 0 -255 range, but we would stil only have eight spikes with no pixels between them. There is still only eight states of brightness that a pixel could have. There would not be small gradations in brightness between adjacent pixels, just crude jumps in brightness. The image would look like a child's crayon drawing rather than a photograph.

So, I see your point, poor dynamic range should be avoided. Noise becomes a secondary consideration. Thanks for pointing this out.

However, in the case of my posted images from a 7 megapixel P&S camera, as Pathfinder pointed out, the expected benefits from a properly exposed image at a higher ISO are not readily apparent (I'll try to post 100% crops from those images in the next few days). Undoubedly, for a DSLR with a larger sensor, the benefits of a wide dynamic range would be more apparent.

Harvey

Harveylevine

Here are the 100% crops that were requested of the noise comparison files. The results are for my Casio EZ750 P&S and is probably not indicative of the results from a high quality DSLR that can handle high ISO settings better than a P&S.

First, the entire image (high ISO) , the 100% crops will be taken from the area around the chair:



Here is the crop from the High ISO image:



Here is the crop from the Low ISO image that was brightened in Photoshop using duplicate layers set to Screen blend mode:



Here is the High ISO image after processing with Neat Image:




Here is the processed Low ISO image after processing with Neat Image:



Harvey