I have recently been imaging the M101 Pinwheel Galaxy in Hydrogen-Alpha, with a 3nm filter. Since this is going to be a serious image with tons of exposure time, I decided to do 50 exposures of 15 minutes each at 1x1. The result of stacking all these exposures together has a very, very clear effect on the Signal-to-Noise Ratio (SNR) of the overall result. The following is a GIF animation showing the result of stacking exposures from 5 to 50 in steps of 5 (click the animated GIF to see a larger version):
The individually-stacked images were pre-processed in exactly the same manner, using Linear Fit Clipping as the rejection algorithm in ImageIntegration in PixInsight. Once that process was ran, DrizzleIntegration was used to produce these individually-stacked images presented above. Each one was cropped identically using DynamicCrop and non-linear stretched with HistogramTransformation (using the same stretch on all the images). No noise reduction measures or further post-processing was carried out.
Above, we can see how quickly the SNR improves as we approach 20 exposures in the stack. Beyond it however, there are only minor SNR improvements visible, particularly on the very faint spiral arm fine details (appearing above the noise floor). The result of diminishing returns is clearly presented above.
Everything is better with numbers so I thought it would be a good idea to measure the SNR of these images. I did this by using the NoiseEvaluation script in PixInsight and multiplying the resulting values by 65,535 to convert to 16-bit pixel value. I then divided the Mean of the images (found using the Statistics process) by their corresponding noise values to find an estimate for SNR. Below is a graph displaying the results:
Above, we can see how quickly the SNR improves as we approach 20 exposures in the stack. Beyond it however, there are only minor SNR improvements visible, particularly on the very faint spiral arm fine details (appearing above the noise floor). The result of diminishing returns is clearly presented above.
Everything is better with numbers so I thought it would be a good idea to measure the SNR of these images. I did this by using the NoiseEvaluation script in PixInsight and multiplying the resulting values by 65,535 to convert to 16-bit pixel value. I then divided the Mean of the images (found using the Statistics process) by their corresponding noise values to find an estimate for SNR. Below is a graph displaying the results:
Above each data point is the SNR calculated to three decimal places, using the aforementioned method. To the top-right of the graph is the equation of the exponential curve of best fit, which was plotted to R = 0.9999. This equation demonstrates that the SNR of the image is indeed approximately proportional to the square root of the number of exposures used in stacking.
Interesting as this may seem, I wonder if the rule holds as closely for more expansive targets (large nebulae that cover most of the image) and for other filters that are more permissive of light (e.g. Luminance). Definitely something to test!
Interesting as this may seem, I wonder if the rule holds as closely for more expansive targets (large nebulae that cover most of the image) and for other filters that are more permissive of light (e.g. Luminance). Definitely something to test!