If you'd like to see a larger view of any image, please click the thumbnails and then your browser's 'back' button to return here. Because this site is intended as a photography portfolio, the images are rather large and will take a few minutes to download.
I work a great deal at high ISO on the Canon 300D/Digital Rebel (800ISO and 1600ISO) and am very comfortable in doing so. The question arises in forums regularly and I get asked often, about how good or usable the higher ISOs are in practice. I'd say very usuable indeed. It saddens me to often read people make misguided statements like "if you use noise reduction software, they can be made passable" - I'm afraid I consider such comments to be prejuduced twaddle.
About half of my portfolio is taken at high ISOs and in many cases, I bet you'd never guess it. The noise 'grain' will increase at higher ISOs, an inevitable side effect of increasing the gain in the digital sensor to make it more sensitive to light and images do require some special handling in order to minimise the potential problems and get the very best from them.
The question often arises about expected results when printing and on screen viewing and people inevitably want to see 1:1 pixel crops of images to inspect the noise and there are others that claim re-sizing an image on screen to a manageable size is cheating and some deliberate ploy to disguise the noise levels and make false claims of its acceptability.
Unless you plan on printing each image at 40" x 30" and viewing it from 20" away, 1:1 pixel screen viewing has limited value for noise appraisal. What really counts is what it looks like when published - by whatever means that image is intended for. Most of my high ISO work is web published, with a small percentage printed, my most common sizes being 8" x 10" and 10" x 15". Considerably smaller than 1:1 pixel viewing.
Printing is resampling too, just as resizing images for a decent screenful on-line. Your 8" x 10" print is similar in viewing resolution terms to seeing your full frame image fill the screen with pixels. Resampling is a very powerful noise reduction technique, especially when using a good algorithm like Lanczos3 and in many cases is enough to give acceptable results without further intervention, see the sample in the bottom section.
Your efforts in respect of noise management start before you take the shot. The big secret to minimising the potential creation of high ISO noise, is exposure - as soon as you under-expose even a modest amount the visible noise increases and this is exhaggerated if you then need to lighten it in post processing to taste. If you can expose to the right of the histogram - slightly over-expose (not much, just 1/3 or 1/2 stop) and prefereably work in RAW to nurse the data along a little better without in-camera compression, sharpening etc., you can get excellent results.
If you work in RAW, which I do and would advocate for any extreme work of this nature, even if you're a usual JPEG shooter, your image can be developed without any sharpening and whilst initially this makes the images on screen look rather flat and soft, there is much you can do later to bring the images back to life, but the great advantage is the minimal noise as your starting point. If you prefer to shoot in JPEG, using a set of custom Processing Parameters specially for high ISO shots, to use the lowest possible sharpening and contrast settings (-2) will also help, there's a study at the bottom of the page.
If you have over-exposed a fraction and can use negative exposure compensation to bring the image appearance within expected visual parameters, the noise can be reduced tangibly with only a modest tweak, as you can see in the uppermost screenshot right.
Conversely, if you've underexposed and need to apply positive EC, you will clearly see an increase in the noise on screen, it doesn't need much of a lightening adjustment to make it much more visible, especially in shadow and out of focus regions. Underexposure is your biggest single enemy in the battle with digital noise. The two samples shown are the same scene taken a stop apart, one is slightly over-exposed and has had 0.4 stop -EC applied, the other was slightly under-exposed and has had 0.4 stop +EC applied.
So, once you've worried about exposing to the right of the histogram, developed it without sharpening, your low contrast, soft and rather flat looking image is now in your editor and your workflow from here on in can make a big difference to the results too, they still need gentle handling!
I'm nowhere near as slavish about noise as most people, it doesn't visually offend me the way it clearly does some, so I tend to reduce it with caution myself as that process has to be a compromise between removing unwanted noise and killing texture and detail - I'd personally rather keep noise than lose other detail. I don't like to see plastic overly smooth looking images where you can clearly see NR has been done, if you can see it has been done, you got it wrong - like many enhancement techniques. Skin can stand to have it done, but it's often best to leave it alone in places like hair and clothing where texture can too easily be lost. Many textured areas will simply hide the noise at the finished published size and simply don't need to have noise reduction done. The example above right demonstrates this with minimal softening done to background areas only - none at all in the foreground, it's simply lost in the colours and details.
My own approach is to only apply noise reduction where it's clearly needed and likewise with sharpening - in many cases, much of the image can be simply left insharpened and without noise reduction if it isn't a key component of the image details. Sharpening a high ISO image globally will inevitably increase the appearance of noise in plain areas, like out of focus backgrounds, as there is nothing else there to sharpen, so don't do it if it's not actually required. Experience has taught me that the global application of techniques like sharpening and noise reduction do not give the best results with high ISO images, they need a little individual attention.
In many images there is a clear focal point for your attention - the eyes in a portait for example - and this is often surrounded by areas of differing degrees of visual importance. If you sharpen the focal point details carefully, it simply won't be obvious that other areas are less sharp, so medium interest areas like the surrounding face in your portrait can be left untouched and background areas which are likely to be out of focus (possibly even more so with high ISO shots taken in low light with wider apertures) may show more noise, but not enough detail to require sharpening.
There are several ways to reduce the appearance of noise and using a noise reduction application is only one. My own advice would be to use them sparingly. Some are better than others, but invariably, if used at default settings and globally across the image, are too aggressive and reducing noise can also kill texture and details and should be used very sparingly clothing, hair and foliage. I use a combination of technqiues to disguise or remove noise and now that NR applications can be used as plug-ins with editors, it makes it easier to apply NR in localised spots - the quickest way being to lassoo the required area (with a feathered selection) and just apply the filter to that area only. I also use the soft focus tool, gaussian blur for backgrounds (only a very small amount, too much is very obvious) and the softening retouch tool for small areas that can be painted over, this works well on skin because you have such delicate and variable control.
High ISO images are ideally worked using layers, which is a very powerful and flexible practice to learn for any image adjustments, but particularly suited to the methods I would recommend for high ISO workflow. If you apply noise reduction to one copy of the image and lay the original untreated copy over the top of it, you can either adjust the transparency of the top layer to blend with the lower layer so the NR isn't too extreme, or you can erase or mask local areas to only allow it to show where required.
Masking on layers is my personal preference and I believe well worth the effort for the dividends - it's nowhere near as time consuming and complicated as people think it sounds to be. A mask creates a hole of transparency in the topmost layer to allow portions of an image below to show through where required. As a mask can be painted on with 256 degrees of transparency, it's very powerful for getting just the results you want where you want them - if you paint just short of your intended area and apply a gaussian blur, it gives it a nice soft edge to blend the areas. There are more details on creating masks in my blending exposures tutorial.
The same can be done with sharpening - only apply it where it's actually required and this might be a smaller area than you might suppose. If the focal point of the image is sharp, the viewers perception of the image as a whole will be that it was sharp. When you sharpen high ISO images, the small radiuses of USM normally used for image sharpening (i.e. less than 1) can sharpen and therefore brighten the noise, especially in dark areas, so my own practice is to use a combination of large radius which won't impact on the noise and a tiny radius for gentle overall sharpening.
A pass of a large radius USM will increase local contrast, putting back some of the punch minimised by the previous noise-friendly workflow, it will give the perception of sharpness without sharpening noise pixels. My own starting settings are radius 20 and amount 20 - which can be adjusted to taste from there. I then brighten the tiny details a little with a small pass of USM, usually radius 0.3 and amount 500% - this can be done several times without introducing halos or impacting on noise.
My own personal approach with high ISO images is to work the steps described at the size the image will be published, so for web publishing, I do any retouching or other global adjustments at full size, resample to the web publishing size, then work the noise reduction/sharpening routine as my last stage at the finished size so you can see exactly what you have to deal with and use appropriate levels of adjustment. If I want to print the same image, I crop and do any re-sizing required and treat that copy appropriately for print output. High ISO images need a little individual attention to get the best from them and I believe it's worth that effort.
Purely as an exercise, a series of 1600ISO shots were taken, half with zero Exposure Compensation in the camera and duplicates taken with -1 stop of EC, this has the effect of mimicing a higher ISO (1 stop = 3200ISO) and allows for either a tighter aperture or faster shutter speed at the time of taking. This may be necessary for your scene and despite everything I've said about ideally exposing well, sometimes underexposing at extremes is better than nothing and may well be preferable to movement blur or too shallow depth of field. Noise can be overcome, blur or lack of focus can not.
This experiment was to see how acceptable the results could me made with careful handling. The 1 stop underexposed image required +EC of 1.2 stops to bring it in line visually with the correctly exposed comparison image and to give a pleasing result.
Below left is a 1:1 pixel crop of the underexposed image after +1.2EC had been applied, the middle view shows the power of resampling in noise management. Nothing else at all has been done to this image, it has not been sharpened and has had no noise reduction done, just resampled with the Lanczos3 algorithm. The right hand view is the finished image from a mock 3200ISO photograph. The smooth areas to the top, left and bottom have received noise reduction, the middle region with the glass grapes is the only area sharpened (NO noise reduction!) and the bottom right corner with the wood grain is totally out of camera apart from the overall saturation boost I subsequently decided the entire image needed. I think the resulting image is more than acceptable.
I resampled this 1600ISO image and added a soft focus filter with a feathered selection to the plain areas inside the window bottom behind the cat that were a tad noisy and I added slightly less of the same soft focus on the far curtain half as that showed a little noise. I sharpened only her eyes, whiskers, nose and ear hairs, the rest is unsharpened and without noise reduction. A large radius USM pass was applied to the entire image to increase the contrast, this gives the impression of sharpness without sharpening details. Taken with the 50mm, 1/60 @ f1.8.
To illustrate the effect of in-camera Processing Parameters (set from the menu, page 55 and 56 of the manual) when shooting high ISO with JPEG, I set up a little test scene with some texture, contrast and colours and shot it in RAW using the same manual exposure and at 1600ISO. I took two sets of images, one each with in-camera Processing Parameters set to +2 and duplicate images with everything at -2.
The 1:1 pixel crops right are taken from the medium embedded JPEG with the RAW file. Although the -2 set is clearly less punchy visually (although I actually prefer that), the amount of noise is very much reduced too. A 1:1 pixel crop is also shown from the RAW file.
Although the in-camera Processing Parameters clearly have an effect on the embeded JPEG and also the .THM JPEG-type thumbnail file containing the EXIF and the difference this makes to the image is reflected in the camera LCD histogram, the Parameters do not influence the RAW image. The two RAW images captured for this test, when developed with the same workflow, were identical.
It is worth noting however that some RAW development tools do present an 'as shot' preview of the RAW file by taking the Processing Parameter choices into consideration.
To show the effect the in-camera Processing Parameters have on high ISO images, these three frames are the web published versions of the two medium sized embeded JPEGs and one of the RAW develops. All three images have had the same workflow (no additional noise reduction, selective sharpening only, to the same areas) although saturation, contrast and white balance were adjusted to taste when developing the RAW image. The images were taken a matter of seconds apart with the same exposure and no adjustment was made for Exposure Compensation, even though the +2 Parameter image (far left) clearly looks darker in the shadows, the effect of the increased contrast, which would be significantly detrimental to underexposed high ISO images.
If you found it useful . . .
Back to the Tutorials menu