AIC Narrowband Imaging Things That Make a Difference Saturday, October 27, 2007 Neil Fleming. (

Transcription

1 AIC 2007 Narrowband Imaging Things That Make a Difference Saturday, October 27, 2007 Neil Fleming (

2 Agenda and Assumptions Agenda: Light pollution? Why even try? RGB and Narrowband Differences Processing Workflow CCDStack Processing Workflow Photoshop Things that Make a Difference are in red Assumptions: You have a working knowledge of image processing techniques and tools like CCDStack For a deeper tutorial on CCDStack, see Adam Block s DVD on the topic I ll go at a fairly quick pace, you can view the presentation online after this seminar My goal in this example is not strictly scientific, it is more the presentation of an enjoyable result

3 Imaging from Boston Why Even Try!?!?

4 What is Narrowband Imaging, Anyhow? Most CCD imaging is done with a black and white camera, and you use color filters to emulate the RGB results so you can obtain a color image: The wavelengths captured are across the entire visible spectrum

5 What Does RGB From Boston Look Like? Light pollution typically causes horrible gradients Here is a sample of a stack of 3-8 minute exposures: A little processing in Photoshop helps, but still, not great results Bright objects are better

6 Try Narrowband Filters Instead Narrowband filters let through only the tiniest bandpass of light, light that is associated with emission lines in nebulae. Typically 3 to 13 nm in width: Doubly oxidized oxygen (OIII) has its main emission line at nm Hydrogen-alpha (Ha) emission wavelength is nm Sulfur (SII) emits at 672 nm

7 NGC6992 The Veil Nebula

8 IC1396 The Elephant Trunk Nebula

9 DWB-111 The Propeller Nebula

10 Processing Workflow: Image Capture Go deep Go for a greater sub duration and overall exposure time than for RGB imaging Instead of 3-8, to perhaps 10 minute subs, I shoot at least 30 minute subs I get hours of usable data for an object Get lotsa data I aim for at least 12 subs per channel, more if possible, especially depending on the object This allows for both depth, and enough subs in the stack to do efficient data rejection of outliers (satellites, airplanes, and cosmic ray hits) Gather ample calibration frames: ~30 bias, darks, 1,000,000 ADU of flats (~30 for me) I get automated dawn flats

11 Processing Workflow: CCDStack Prepare calibration masters bias/darks/flats: Use some sort of sigma rejection or clip min/max, don't use mean This eliminates the impact of outliers like cosmic ray hits (all three types) and stars (for the flats) Bias-subtract your flats when you create your flat master to accommodate differing temperatures Subexposure processing Load the individual subexposures (subs) I start with Ha data Calibrate them (apply your darks, flats, and bias if time or RMS scaling) Check DDP and apply to all

12 Uncalibrated Sub

13 The Same Calibrated and DDP d Sub:

14 Processing Workflow: CCDStack (continued ) Subexposure processing (continued ) Rotate all of the subs to the same orientation Carefully evaluate your data! CCDInspector for contrast, aspect ratio, and FWHM evaluation Mark I eyeball as a second step, especially if your data is undersampled, for gradients and star aspect ratio Good / Marginal / Bad I discard the Bad subs, and keep the Good along with a few of the Marginal The larger the stack of good subs, the more of the marginal I can include

15 Bad, Marginal, Good Bad Marginal Good

16 All Versus Best Plus a Few Marginal (Combined) All subs Selected subs

17 Rejection of Large Blooms If working with an object that imposes large blooms, I will take shorter subs to include with the longer subs so as to get to the real data: I impute the blooms on the shorter exposures, and Set to MissingVal (make transparent) the blooms of the longer exposures See: for a full write-up on this approach

18 Processing Workflow: CCDStack (continued ) Registration I image the same object over multiple nights Sometimes I see a little offset or a slight extra rotation So for speed sake, I do a dual pass at registration The first pass is a Star Snap with two fairly bright, isolated, closely-spaced stars, selected as the registration stars This gets the registration, close :

19 First Pass at Registration

20 First Pass at Registration

21 Processing Workflow: CCDStack (continued ) Registration (continued ) The first pass was not perfect, just close: The second pass at registration is my fine pass I use Star Snap again, this time with 3-4 mediumsmall, widely-spaced, stars to tune up the registration:

22 Second Pass at Registration

23 Processing Workflow: CCDStack (continued ) Registration (continued ) Much better!: First Pass: Second Pass:

24 Processing Workflow: CCDStack (continued ) Registration (continued ) Alterations to the images for alignment purposes do not take place until you Apply the change, so feel free to tweak to your heart s content I find that I prefer the Quadratic B-Spline algorithm to apply the registration Normalize your stack of subs Normalization utilizes techniques to assess the weighting contribution that each individual subexposure should contribute to the final combine:

25 Processing Workflow: CCDStack (continued ) Normalization

26 Processing Workflow: CCDStack (continued ) Data Rejection CCDStack allows you to reject poor data like satellite trails, cosmic ray hits, and airplane trails independently of the combine method! You do not have to rely on mean, median, etc., to get rid of these pests! Options include: STD Sigma Poisson Sigma Linear Factor Clip Min/Max Each of these methods will throw out the outliers and average the remaining pixel values I often use the Poisson Sigma reject, with 1.5 to 2 standard deviations (sigma multiplier) Larger stacks can take tighter tolerances

27 Rejected Data a Good Subexposure

28 Rejected Data a Marginal Subexposure

29 Processing Workflow: CCDStack (continued ) Combine: Do a Mean combine Noise removal in CCDStack (Hot/Cold Pixels) You need to play with this each time I try not to let the upper limit get higher than the medium-to-faint stars, else they will get rejected too Impute at 0.2 pixels

30 Processing Workflow: CCDStack (continued ) Deconvolution I prefer the Positive Constraint algorithm Choose a medium star on a dark background Original Deconvolved

31 Ha is All Done

32 Processing Workflow: CCDStack (continued ) Now do the OIII and the SII (if you have it) via similar processing steps After calibration and rotation, when I start registration, I re-load the Ha master and use that one to register the OIII and SII subs This allows for only one destructive registration step to be applied, not two! I then do a final tweak with DDP to maximize the presentation of the object You now have your Ha, OIII, and SII masters Color time!

33 Processing Workflow: CCDStack (continued ) One approach is to do the initial color combine in CCDStack, MaxIM, etc.:

34 Processing Workflow: CCDStack (continued ) For narrowband work, unlike RGB, I prefer to "optimize" each channel in Photoshop (PS) before the color combine This helps to maximize the contribution from each data set, especially the OIII and SII So, at this point, I will save the FITS as a "scaled TIF" from CCDStack This squeezes or translates the umpteen million data values into the 16-bit color space I open each TIF in PS, and closely examine the histogram to make sure I have not clipped the data at either end I find the scaling process in CCDStack will usually clip just a bit on the dark end of the histogram So, I will go back into CCDStack, lower the dark value cutoff a bit, and re-save the scaled TIF until I am satisfied

35 Before and After DDP and Clipping Adjustments

36 Processing Workflow: Photoshop To "optimize" each channel in Photoshop (PS) before the color combine, I generally do: A contrast curve adjustment layer Noise reduction layer with an inverted layer mask Local contrast enhancement layer, created with Noel Carboni s Photoshop actions Shadow/Highlight adjustment layer (in PS-CS2 or higher)

37 Contrast Curve

38 Noise Reduction With an Inverted Layer Mask Do your favorite noise reduction process step: Noise reduction in PS NeatImage Noise Ninja, etc. I do this on a separate layer in PS because I will use a mask later I find that noise reduction is less necessary in the brighter areas of an image than the dark areas I use an inverted layer mask because that gives me the finest representation of the bright and dark areas of the image, as well as excellent control of how much de-noising to apply!

39 Noise Reduction With an Inverted Layer Mask I do my noise reduction on that separate layer Select the noise reduction layer CTL+A to select the contents of the layer CTL+C to copy those contents Layer / Layer Mask / Reveal All adds in the layer mask I click on the little chain link to unlink the mask from the layer This allows for adjustments to the layer mask to not be irretrievably linked to the layer contents

40 Noise Reduction With an Inverted Layer Mask I then ATL/Click on the layer mask This selects the layer in such a fashion that you can work with it as an image in itself! CTL+V to copy the noise-reduced layer onto the layer mask CTL+I to invert the image on the layer mask The areas that were dark are now light, and vice versa I then do Image / Adjustments / Levels to darken the image overall, to ensure that the brighter areas of the source image receive less emphasis than the darker areas

41 Noise Reduction With an Inverted Layer Mask Now we have an inverted layer mask that allows the noise reduction layer to impact the image more efficiently

42 Original, De-Noised, And De-Noised with a Mask Original Too noisy De-Noised Too much Masked Just right

43 Processing Workflow: Photoshop (continued ) Local Contrast Enhancement with Noel Carboni s Photoshop Actions This works to increase the contrast in the mid-range of the image:

44 Processing Workflow: Photoshop (continued ) A little taste of a Shadow/Highlight adjustment, then I am done the pre-combine work: Original Ha data Final, after tweaking

45 Processing Workflow: Photoshop (continued ) Same process for OIII and SII:

46 Processing Workflow: Photoshop (continued ) Color combine; early in CCDStack vs. later in PS:

47 Photoshop: Color Enhancements Perhaps a bit of a Channel Mixer adjustment layer (I may add a bit of the Ha to the blue channel, to reflect the H-beta component):

48 Photoshop: Color Enhancements (continued ) A Selective Color adjustment layer in this case, to reduce the overall yellow cast in certain colors and to reduce the magenta stars:

49 Photoshop: Color Enhancements (continued ) Curves adjustment layers to make the reds redder and the blues bluer I sometimes work by selecting a color range to pick up the desired reds or blues

50 Photoshop: Color Enhancements (continued ) So color adjustments; before and after:

51 Photoshop: Another LCE Tweak Another subtle go at LCE, this time with a positive layer mask:

52 Photoshop: Remove the Magenta in the Stars I select the star cores, and work outwards by expanding and feathering the selection I then apply a Hue/Saturation adjustment layer to reduce the magenta cast

53 Photoshop: Process the RGB Stars Collect RGB data, align to the original Ha data to avoid the second, destructive, alignment process

54 Photoshop: Now add in the RGB Stars Bring in the scaled TIF, prepare a mask for just the stars, and change the blend mode to color

55 Photoshop: Sharpen to Taste PS sharpening, NIK Sharpener Pro, others

56 Works Better on Undersampled Data

57 Photoshop: Crop for Presentation Voila!

58 Dumbledore Version

59 Summary Get lotsa data: Shoot at least 30 minute subs Aim for at least 12 subs per channel Gather ample calibration frames: ~30 bias, darks, 1,000,000 ADU of flats Use some sort of sigma rejection or clip min/max, don't use mean Carefully evaluate your data, reject the bad For extensive blooming, take two sets of data: Impute the blooms in the short-exposure subs Make the blooms transparent in the long-exposure subs

60 Summary (continued ) Apply advanced data rejection techniques independently of the combine step Poisson Sigma rejection, Clip Min/Max After combining, do one final run at hot/cold pixel rejection Use DDP and Deconvolution For subsequent data sets, always register to the initial master that you produced Ensure that you do not clip the data as you save your masters as scaled TIFs and import into Photoshop

61 Summary (continued ) Do your color combine in Photoshop, after you have optimized the presentation of the individual channel data Use an inverted layer mask to carefully control the noise reduction applied For mid-tone contrast enhancements, I like the Local Contrast Enhancement action from Noel Carboni For handling extremes in the image, I like the Shadow/Highlight adjustment in PS

62 Summary (continued ) Color adjustments in adjustment layers: Channel Mixer Selective Color Curves (in concert with masks) to bring up the reds and blues Reduce the magenta in the stars by selecting the stars and working out, in combination with a Hue/Saturation adjustment layer If you have time, collect RGB data to be used for RGB-colored stars Bring in via color mode, not normal

63

64 Questions? Narrowband Imaging Things That Make a Difference Saturday, October 27, 2007 Neil Fleming (