Chasing Faint Objects

Transcription

1 Chasing Faint Objects Image Processing Tips and Tricks Linz CEDIC 2015 Fabian Neyer 7. March

2 Small Objects Large Objects RAW Data: Robert Pölzl usually around 1 usually > 1 Fabian Neyer ǀ CEDIC 2015 ǀ 2

3 Small Objects Exposure time Optimal stacking Drizzling Deconvolution RAW Data: Robert Pölzl usually around 1 Fabian Neyer ǀ CEDIC 2015 ǀ 3

4 Concept of Drizzling Linear Reconstruction of undersampled images (originally developed for HST) Works best with a large set of dithered images Coarse input pixel grid Geometric transformation Fine output pixel grid Fabian Neyer ǀ CEDIC 2015 ǀ 4

5 Drizzling with PixInsight Add and update drizzle data with statistical properties determined during image integration Drizzle Integration - Add drizzle data files - links original files with pixel statistics and position Activate Drizzling data generation Fabian Neyer ǀ CEDIC 2015 ǀ 5

6 Drizzling and Deconvolution Original stack Drizzle stack Drizzle improves FWHM (here ~12%) Deconvolution improves FWHM (here ~20%) A significant improvement of small and also faint objects Drizzle stack, deconvolved Fabian Neyer ǀ CEDIC 2015 ǀ 6

7 Large Objects Exposure time Proper Calibration (Flats!) Optimal stacking Star removal Masked stretch usually > 1 Fabian Neyer ǀ CEDIC 2015 ǀ 7

8 Mihos et al Dezember August Fabian Neyer ǀ CEDIC 2015 ǀ 8

9 PacMan Nebula NGC h Halpha Fabian Neyer ǀ CEDIC 2015 ǀ 9

10 PacMan Nebula NGC h LRGB Fabian Neyer ǀ CEDIC 2015 ǀ 10

11 PacMan Nebula NGC h LRGB h Halpha Fabian Neyer ǀ CEDIC 2015 ǀ 11

12 Flatfielding The Problem Level of contrast: < 0.01% Fabian Neyer ǀ CEDIC 2015 ǀ 12

13 Flatfielding The Problem Idea behind Flatfielding: Correction of vignetting, stray light, and dust particle effects This would work if the flat fielding light source (panels, twilight sky, LEDs, etc.) would have the same spectrum as the night sky! Why? Many black surfaces (telescope, focuser, etc.) are not black (depending on spectra of the illumination source) Lights reflects off the bond wire glints at the edges of the CCD and reflects again at the CCD cover glas (large spectral dependency) Black objects in infrared light (940nm) Fabian Neyer ǀ CEDIC 2015 ǀ 13

14 Flatfielding The Problem Aurora Flatfield Panel Spectrum Spectrum of natural night sky (blue) and light polluted night sky (red) F. Patat, IAU/ESO Fabian Neyer ǀ CEDIC 2015 ǀ 14

15 Flatfielding How to deal with this Combination of Standard Flats and Night Flats : Standard Flats: - Update regularly - For mobile imaging: update every session - Check filter wheel positioning accuracy Night Flats: - Acquire images of empty sky areas - E.g. during each session, take 3-4 images (5min) with the filter in use - Suitable Standard Flats must be available - Assumption: More stable over time Combine both to measure the flat field error Correct standard flat and use this for correct flat fielding Always use the same light source for the Standard Flats Fabian Neyer ǀ CEDIC 2015 ǀ 15

16 Additional notes regarding Night Flats For each individual Night Flat image, point the telescope to different (not busy ) areas of the sky. Tracking (or even guiding) during the exposures should be enabled to guarantee tight stars (which helps a lot for rejecting them later stars are considered as outliers). The idea behind large FOV offsets between the single Night Flats images is not only to reject all the stars but also to not include large scale nebula structures (which could by chance be present and which could remain if you chose small dithering amounts it would destroy your goal of a proper Night Flat). The background ADU of the Night Flats must not be very high (around is usually enough) because a median filter is used to reduce the noise after stacking. The filtering is important in order to not introduce additional noise during calibration with the corrected Masterflat. Fabian Neyer ǀ CEDIC 2015 ǀ 16

17 Single (calibrated) Night Flat Measure Flatfield Error Preparing Night Flats Fabian Neyer ǀ CEDIC 2015 ǀ 17

18 Measure Flatfield Error Result ADU max ADU min = 37 Calibration as with normal lights Integration parameters as with standard flats Median combination (or narrow Sigma Clipping) Noise reduction by median filtering (most robust) Fabian Neyer ǀ CEDIC 2015 ǀ 18

19 Measure Flatfield Error Apply Multiply Standard Flat ($T) with measured Flatfield Error (flat_err_b) Calibrate lights with this corrected Masterflat Fabian Neyer ǀ CEDIC 2015 ǀ 19

20 Traditional vs Corrected Flat field correction before after Gradient removed by simple ABE (function degree 1) for both results No further background correction Same STF applied Fabian Neyer ǀ CEDIC 2015 ǀ 20

21 Digging out large faint objects Collaboration with Robert Pölzl 4.3h L h RGB Fabian Neyer ǀ CEDIC 2015 ǀ 21

22 Digging out large faint objects where we want to get Fabian Neyer ǀ CEDIC 2015 ǀ 22

23 Digging out large faint objects Simple non-linear stretch obviously fails Fabian Neyer ǀ CEDIC 2015 ǀ 23

24 Star Removal - Idea It is not only the obvious diameter of the stars that bother but their halos Point Spread Function (PSF) fitting (only an approximation to the real scenario) Saturated stars can produce extensive halos Straight forward filtering (i.e. Dust & Scratches...) is not suitable, especially in crowded areas My Approach: Equally bright stars (no matter if they are in front of a nebula or not), produce equal halos The joint appearance of star halos in various image areas gives an idea about their influence Fabian Neyer ǀ CEDIC 2015 ǀ 24

25 The Starless Image (1) Create star mask with background corrected star intensities (2) Iteratively reduce star halos in linear image using the star mask created above (3) Replace star pixels with background (4) Photoshop cleaning Fabian Neyer ǀ CEDIC 2015 ǀ 25

26 (1) Star Mask Approximating Star Brightness original (1) single step BG fill Fabian Neyer ǀ CEDIC 2015 ǀ 26

27 (1) Star Mask Approximating Star Brightness original (1) single step BG fill iterative BG fill (2) (1) (2) Fabian Neyer ǀ CEDIC 2015 ǀ 27

28 (2) Removing Star Halos Split the star image in a linear (L) and non-linear (NL) part, i.e., non-saturated and saturated stars (use PixelMath, MorphologicalTransformation) Use various convolution filters to blur the linear star image (L) and subtract it iteratively from the original image Fabian Neyer ǀ CEDIC 2015 ǀ 28

29 (2) Removing Star halos For the non-linear (NL) part, use convolution and histogram clipping More trial and error needed Fabian Neyer ǀ CEDIC 2015 ǀ 29

30 (3) Replace star pixels with background Binary starmask MorphologicalTransformation MultiscaleMedianTransform Fabian Neyer ǀ CEDIC 2015 ǀ 30

31 (4) Photoshop cleaning Make Non-linear (Masked Stretch, Curves, ) Dust & Scratches Careful for relevant small scale structures Real structures or flat field error? How good is your flat fielding? Δ ADU = 6.3 Fabian Neyer ǀ CEDIC 2015 ǀ 31

32 The Starless Image 22.5h f h f3.8 Fabian Neyer ǀ CEDIC 2015 ǀ 32

33 M27 LRGB 17.8h LRGB

34 Fabian Neyer ǀ CEDIC 2015 ǀ 34

35 Where to apply? Care must be taken to not amplify star residuals We cannot extract background where the field is too crowded

36 Whale Galaxies NGC 4631/56 R Coronae Australis region 4 h LRGB

37 Whale Galaxies NGC 4631/ h LHaRGB

38 Whale Galaxies NGC 4631/ h LHaRGB

39 Summary & Take-Home Message Small faint objects: Limited by: Aperture and Seeing (and exposure time) Drizzling Deconvolution Large faint objects: Limited by: Flat field correctness and influence of stars (and exposure time) Correct flat fields using Night flats Remove influence of stars by iterative subtractions of blurred star masks Check your flats, they tell you how deep you can get! Fabian Neyer ǀ CEDIC 2015 ǀ 39

40 Thank you! Image Credit: F.Neyer/R.Pölzl 79h HaOIII-LRGB