Global Chroma Noise Reduction in Photoshop
By: Andrew Masur
The latest generation of DSLRs are now capable of producing high ISO images of remarkable quality. Anyone who used to shoot film is constantly amazed at being able to do photography at ISO 1600 and higher with hardly any noise.
But once we get above 1600 noise starts to rear its head, and even the current low noise king, the Nikon D3, suffers from noise at ISO 12,800 and 25,600. This noise is mostly chroma noise, which means that it's in the colour information more than in the luminance, and while luminance noise looks a lot like traditional film grain, chroma noise is simply nasty looking.
In this Photoshop technique tutorial by Andrew Masur we learn one of several techniques that are available for reducing chroma noise in high ISO files.
With the increasing dominance of digital SLRs, photographers find themselves shooting in more varied situations than ever before. It would be ideal if we could shoot at low ISO all the time, however this is rarely the case. While the noise characteristics of modern cameras are generally quite good, shooting at high ISO values still introduces unwanted noise into an image.
As seen in digital photographs, noise manifests itself in two different forms, chroma noise and luma noise. While having no noise at all would be ideal, that is simply not possible at high ISO values. Luma noise tends to resemble the type of grain that would be seen in high speed film stock, while chroma noise characterizes itself as random color splotching across the image; especially in shadow areas. In general, people are not overly bothered by luma noise; however chroma noise has an unappealing effect on an image. Therefore while we can’t get rid of all noise from an image, removing the chroma noise improves things a great deal. The goal of this article is to present a quick and easy way to greatly reduce or eliminate chroma noise from an image using some simple layer techniques in Photoshop.
Let’s start with an example photo taken at a hockey game. This was shot with a Canon Digital Rebel (300D) at ISO 1600 and purposely underexposed by one stop in order to increase the shutter speed. The image was then pushed one stop in post processing, normalizing the exposure at the expense of introducing noise. This, coupled with the fact that the original Rebel is quite noisy at ISO 1600, provides us with an image that is about as noisy as you would reasonably see. We can see the image opened in Photoshop at 100% below. All cropping, contrast, and colour adjustments should be completed prior to starting this procedure.
The first step is to create a greyscale version of the picture that will act as a luminance layer in the final image. Select the entire image (select all) and then copy it. Create a new blank canvas with the same dimensions as the current image (Photoshop should do this automatically) and paste the copied image into it.
Convert the new copy of the image to greyscale by clicking Image / Mode/ Grayscale. Perform all final sharpening to this greyscale version using whatever method you prefer. You may also wish to perform some minor noise reduction to the greyscale image in order to reduce grain. Once all of these adjustments are completed you should have something that looks like the following.
Now we’ll return to the original image in order to clean up the chroma noise. The first step is to run the Despeckle filter as shown below.
Next run the Dust and Scratches filter. A good starting point is a Radius of 5 pixels and a Threshold of 0 levels. Depending on the amount of chroma noise you may have to raise or lower the Radius; however 5 should suffice for most images.
Now run a Gaussian Blur filter on the image. A starting Radius is 3.4 pixels should be perfect for most images; however this can be increased or decreased depending on the amount of chroma noise present in the image. This is shown below.
Once the Gaussian Blur is complete, create a third canvas with the same dimensions as the original image and set the colour mode to RGB colour with a depth of 16-bits. Return to the greyscale image you created earlier and copy and paste it into the new canvas. Flatten the image to make the greyscale image the background layer. Next, copy and paste the blurred colour image into this canvas so that it occupies the layer directly above the background (typically Layer 1), as shown below.
Finally, select the layer with the blurred colour image and change the blend mode to Color as shown in the following screenshot.
Comparing final image shown below to our original image we can see that most of the chroma noise has been eliminated. The only noise that remains clearly visible is graininess caused by luma noise; however for a print this is far less objectionable. Flatten the image and save it to whatever format you require.
Andrew Masur is a space systems engineer working in Brampton, Ontario. When he's not busy building hardware for the Space Shuttle, he can be found shooting photos of things more down to Earth.