Adding a Background Noise Floor During Post-Processing

December 26, 2017

Over the past couple of years I’ve found that running PixInsight’s Dynamic Background Extraction (DBE) process after removing noise from a stacked image can result in a blotchy, mottled background especially in galaxy images with large dark backgrounds. This is likely due to large variations in background levels across multiple night’s of shooting using our fast scopes. Denoising can result in medium-sized structures with varying brightness. While running DBE before noise reduction reduced the issue, it didn’t disappear completely.

While researching new techniques for using DBE, I found this discussion about adding a floor of gaussian noise to fill in the dark gaps:

Using our cropped Luminance (L) stack of NGC 660, I came up with a workflow which did a great job of evening out the image’s background seemingly without losing any signal from the foreground or background galaxies.

Here is the original stacked image with autostretch:

Even with proper flat fielding, you can easily see the background gradients. Applying DBE to this image resulted in a mottled background. Even though 69 8-minute subs had been stacked, gradient removal revealed the remaining noise (no noise reduction had been done at this point, and flats which were shot at the end of several days of shooting did not process out one traveling dust mote):

In Photoshop I created a new image of matching size and filled it with 99% black. In PixInsight, I used the NoiseGenerator process to add gaussian noise to this gray image, making sure to keep the noise level no higher than the noise level in the original DBE image:

Finally, I used PixelMath’s Max() function to apply the noise floor image to the L DBE image. I added a multiplier to the noise floor image to allow scaling of the amount of noise added, carefully checking the results so that no signal was lost. The max function only adds the noise floor to areas of the background which are the darker parts of the mottling, darker than the noise floor image:

The results are fantastic. The background has been evened out nicely, and there has been no degradation in the signal. The even background will allow more aggressive processing in later stages. Here is a a before/after comparison showing the effects of the noise floor:

Finally, here is the L image after processing:

Related Posts