While working on a camera comparison for a RASC Journal column I stumbled across some strange things with Nikon cameras. It seems that filtering is applied to their raw images, producing some strangely clipped histograms in both bias and dark frames.
Bias frame histogram
While trying to explain the clipped histograms and the very quiet light frames I came up with a simple class of statistical filter that reproduces all the histogram features and gives great noise reduction with very little blurring of the image. One of the nicest features is that the blur is so well controlled that masks are not needed and the filter can be applied even before any stretches are applied to an image.
The filter is a version of a simple average kernel. The difference is that the central pixel of the kernel is replaced only if it is less than the average of the pixels within the kernel. This means that for images like darks and bias frames, the histogram will be clipped at the average which is usually the peak of the histogram. Any data below the average likely contains noise, while data above the average tends to hold the detail in the image. In order to accomplish this with standard filtering tools available in most image processors all we need to do is to apply a small kernel averaging or box filter to the image on another layer. Set the blend mode of this new layer to lighten and presto – a great noise reduction filter.
Image without filter
Image with filter applied
As you can see from the M8 shots above there is very little blurring of the of the image while the noise in the dark areas of the image is reduced. The filter applied to the filtered image uses a 7 by 7 kernel averaging filter and the lighten blend preserves detail throughout the shot.
Using this noise reduction technique and a LMS combine of a longer exposure and a shorter barlowed shot to get core detail the M42 image below gives an HDR rendition of one of my favorite deep sky objects.
