Processing the 255 hour M51 collaboration

Image:

Full-quality image for free download here: https://live.staticflickr.com/65535/53102830146_dcf5e6941c_o.png

You can view all the details and interesting features of the image on astrobin here: https://www.astrobin.com/7hwtz0/

For a great video on this collaboration, look here: https://www.youtube.com/watch?v=zkES2ltGSoc

For a zoomable image, look here: https://elveteek.ch/en/m51-dsc

Processing

The processing steps here are a result of an iterative processes where I determined the best methods for this dataset. I processed the entire dataset a total of 7 times, and this was the best result.

Background Flattening

Multi-scale Gradient Removal was used to remove any widefield gradients from L, R, G, B, and Ha. Additional DBE was done on H-alpha to remove background blotches created after denoise due to the low snr.

Noise Reduction

In order to effectively noise-reduce the image, deepsnr was used. Because this noise reduction algorithm requires an rgb image with independant chrominance profiles, the datasets for L and Ha were split into three equal chunks and integrated to create the RGB image. This was then denoised. The RGB image was noise reduced using the typical TGV and MMT apporach.

H-alpha Continuum Subtraction

To isolate the pure H-alpha components of the image, an HRR image was created. The H-alpha component was isolated from this image using the pixelmath: $T[0[ – ($T[1]-med($T[1])). This process was done 2 times, each with differing amounts of noise reduction, in order to preserve details in the galactic nebulosity while also revealing the faint outer structure.

Other linear processing

I worked quite hard on Luminance to balance the amount of noise reduction and detail that I could bring out. The following steps were used to processes the Luminance image in the linear state:

• Deconvolution on RGB image using the Regularized Van-Cittart algorithm
• BlurX only targeting stars to shrink the stars
• Deepsnr
• Extract grayscale
• MMT noise reduction

Because the RGB image is only used for color information, there was a strong bias towards reducing chrominance noise. The following steps were used to processes the RGB image in the linear state:

• TGV denoise targeting Chrominance
• DeepSNR with a star mask
• NoiseX with a star mask
• Spectral photometric color calibration fit to a average spiral galaxy
• BlurX only targeting stars to shrink the stars

Non-Linear Processing

The first steps to do when processing an LRGB image is to stretch and combine the L and RGB images. It’s very important that neither image be stretched too strongly, and that the stretch of L matches the stretch of RGB. This is the result of the initial LRGB stretch and combination:

While linear, HDR is applied to the core of the galaxy to equalize the contrast with the spiral arms and the color was changed using curves transformation:

Next, the goal is to show the extent of the tidal tails without blowing out the core. This is done with a careful iterative stretch using MMT masks and histogram transformation, as well as LHE:

Then, the H-alpha image was non-linearly added using the linearization technique. GAME masks were used to add differing amounts of H-alpha to the core and outskirts.

Small adjustments were made to saturation and color using curves transformation and histogram transformation together with luminance masks

Finally, stars were added back to the image using the same relinearization technique.