Here's a quick tutorial on how I work with high ISO images. I use a very similar workflow for my regular images, but I sometimes skip the noise reduction step. This tutorial assumes that you have a copy of Photoshop 6.
All text and images are © 2001 by Ronald Parr.
This shot was taken at ISO 400. The noise would probably be a lot worse were it not for the cold temperatures; I took this shot in Germany in early November and it was close to freezing.
Now we'll see how we can go from something that's pretty ugly to something that's quite salvageable, especially for web use.
The standard thing to try with a shot like this is to reduce the shot and then apply a sharpening filter. I reduced to 640x480 using bicubic resampling and applied the standard Photoshop sharpening filter, yielding the following result:
It's not bad, but you'll notice a lot of colored splotches in the vertical branches. This is the result of noise that was not completely averaged out when we reduced the image size and then enhanced by the sharpening filter. We can do better:
This time we're going to start with Jes's color grain reducer. I discovered this in a thread on dpreview. I used the heavy smoothing option for this one:
|
|
| Original | After noise reduction |
Notice that the random color noise is cleaned up but essentially no detail is lost. What's also interesting here is what we have not done. Noise in the brightness channel remains, so even though the color is smooth, we still have some random fluctuations in brightness. At this point, you can do a number of things: There are aftermarket programs for reducing noise, or you could try one of the filters built in to Photoshop. I find that despeckle does a reasonable job, but it does extract a penalty in sharpness if you plan to use the image at full size. In my fall gallery I used despeckle and I still got a very sharp image after reducing and sharpening. However, I realized in preparing this tutorial that no further noise reduction was necessary for this shot. Reducing to 640x480 melts away most of the luminance noise:
Finally, we need to apply a sharpening filter to account for the softening introduced by the bicubic resampling when we reduced. The default sharpening filter will tend to amplify any noise left in the image, so we'll want to use something a little better. In this case, we apply this excellent sharpening action from Fred Miranda. This does a great job of pulling out detail without enhancing noise. In this case I used the 8-bit AUTO setting:
|
|
 |
| First attempt | Noise reduction | Noise reduction and sharpening |
Notice the way the edges of the leaves have sharpened, how the leaves have a certain crispness and glow to them, and how the branch between the two bottom leaves gets a more three-dimensional feel to it. At the same time, the remaining noise in the branches and background is largely unaffected.
We're done! The image is now ready to share with the world and nobody will know that it was shot at ISO 400.