Noise Control – The Upsizer
I know that last tutorial was also a post processing tutorial, and I will put together something more camera oriented for next time. For now, let’s bust open a mistake I made the other day, and see where I was able to recover the image.
Today I am going to share a little trick to reduce and control noise with a very simple procedure available in most graphic editing programs.
I was snapping some shots of the Meta bikes at Joyride 150, and realized after I had lined up this shot, that the camera’s iso setting was up at 3200 iso! Oh no, had I been shooting Raw, I would easily have recovered this shot, however instead I had shot this in fine jpeg, which lacks the latitude to recover the broad range of tones in the image.
This image also lends itself well to illustrating a few points on high iso shooting to begin with. With proper exposure, high iso is usually workable when it comes time to print and the artifacts and luminance noise that you see on a computer are usually only a problem in the extremely magnified image you see on your computer screen when you zoom in to 100%, and are not visible in a print.
Noise shows itself in two ways in digital images, and prominently at certain tones. We see Luminance noise where the grain of the image becomes blocky and “sandy”, and we see Chroma (colour) noise in the appearance of red or blue flecks in the image. Many people can live with the luminance while the colour noise particularly bothers me. In many cases, a simple “despeckle” filter in photoshop will clean up most noise, or for many people, “res-ing down” an image to a smaller size will discard some of the noise in the process. But what happens when you want a clean image of the size that your camera shoots at?
There are a few filters that can be used in photoshop in the process of resizing an image, and to work best they need to be able to identify certain variables. In this case I am benefitting from one of the perks of shooting with the 25 megapixel A900. The photosites on this camera are big, gathering lots of clean light, and the noise is relatively small when compared to the detail in the image. This helps me when printing from noisy files, however it makes the noise itself difficult for the filters to identify. So what I did here was counterintuitive, I up-resed the image. That’s right I made it bigger using the resize image option. Some programs call it “resampling”.
I usually work with the longest edge as a guide. In this case I simply doubled the number from 6048 to 12096 pixels (97 1/2 mp). This is almost 4 times the information for the filters to work with.
Next I used a semi aggressive noise reduction filter called median. There are other filters you can use, like despeckle, though I usually end up having to do it more than once. You can also use the noise reduction filter if it’s in your version of photoshop.
Bicubic sharper seems like the obvious choice for images when down-resing blurry images that you want to sharpen. It works against you with noise, unless you’ve dealt with it already. In this case it works very well at cleaning up the image after you’ve blurred all that noise apart.
As in most cases, the proof is in the pudding, and to show the result at a highly compressed web version would not show anything, so I’ve included a couple of 100% crops to show the detail retention and noise reduction before and after (click for larger).
*Note that these images are web compressed, and also would not be the end of my editing.*
Have fun here, and give this a try. Compare it if you like to other noise treatments that retain the same image size, and see how much detail can be retained. In every case, I only use noise treatment with an end size of print in mind, and that allows you to aim for an “Acceptably sharp” image at the intended size. Remember that most of the noise complaints you may have are based on examining 1″ of an image, blown up to the size of your monitor.