UNIVERSITY COLLEGE LONDON Department of Physics and Astronomy. An Introduction to Image Processing

Transcription

1 UNIVERSITY COLLEGE LONDON Department of Physics and Astronomy UCL Observatory PHAS An Introduction to Image Processing 1 Introduction Students will have submitted imaging requests to the RoboScope queuing system. These notes should help you produce colour images from your data; alternatively, if your observations have not yet been obtained, you can use archival material. In either case, the data are obtained from the digital archive in the same way. You first need to log in to the ChRIS system just go to and log in with your UCL username and password (not your UCL Live or UCLO login details). Be aware that this is still a beta system, under development, so not everything will work, some bits are out of date, some bits won t do what you might expect. One particular feature to note is that ChRIS doesn t always play nicely with the back button in a browser; if in doubt, force a reload of any pages reached in this way (the F5 key does this on most browsers). Any specific bugs should be reported ( details to idh@star.ucl.ac.uk, copied to sjf@star.ucl.ac.uk). 2 De-archiving data You can check to see if your personal observation request has been fulfilled by going to My Scheduler from the ChRIS main menu; this is the easiest way to check which of your observations have been taken, and which have not. At the time of writing it is not recommended that you actually access your data through My Scheduler; instead, choose Image Archive. If you select your name from the Observer list, you ll be taken to your personal results. (Optionally, you could make use of other observers results.) Click Submit to actually interrogate the archive. previews is ticked. Make sure that the check-box to Display You should now be presented with a list of archive images of your selected target, with corresponding preview thumbnails. You can click on the thumbnails for more detailed views, but just the postagestamps are generally sufficient to identify obviously flawed or mis-centred frames. [If you end up with very large numbers of preview frames, you might restrict yourself to one filter at a time, or limit the range of observation dates. For the purposes of this training exercise, something like a dozen or so reasonable images should be enough for a satisfactory result.]

2 2 Image Processing Identifying frames with which to work For critical or quantitative work, one would normally carry out a series of processing steps on every image, including bias subtraction, dark-current removal, and flat-fielding 1. Since the telescope will not point in exactly the same direction in each exposure, one should also carefully align individual frames. Finally, cosmetic blemishes (in particular, cosmic-ray hits and hot pixels) should be repaired. All this can be a demanding and very time-consuming process, especially if dealing with many images. Fortunately, the archive software includes a fully automated processing system (or pipeline ) that does much of this leg work for us. We ll use the pipeline processing in order to get to a quick-look result (which in practice is usually as good, or at least very nearly as good, as one could get through a much more interactive process). The pipeline works best if all the input images are free of obvious flaws, and are reasonably well aligned to start with. The alignment is usually pretty good, but the RoboScope target-acquisition process includes an initial exposure that s used to tweak up the telescope pointing. These initial exposures can be significantly displaced, and should be excluded. You should therefore click the check-box for each of your archive images that appears reasonable in terms of exposure and pointing. (In practice, it s often simpler to click the very first check-box, which will fill in all the others; then uncheck bad images.) It s important to get this right! 2.2 Downloading results Having ticked the relevant checkboxes to select a set of good, reasonably well-aligned images in the filters of your choice (these may be LRGB for a galaxy, or might include H-alpha for emission nebulae), click Download selected. You ll then have to make choices about the level of automated processing, and the file format. In the spirit of getting to a reasonable result in a reasonable time, choose Align and stack by filter ; the pipeline then does a lot of the hard work for you, to generate a single merged image for each filter used. (The stacking is done by applying a median filter to the intensities in each image, on a pixel by pixel basis; experience suggests this is the most reliable simple way to combine frames.) FITS (the Flexible Image Transport System) is the standard data format used in astronomy; it allows more information to be conveyed than do standard image formats, like jpeg. To retain that information at this stage, choose the FITS format, and then confirm download. You can save the result to the local disk of the PC you re working on, 2 /data_local, making an appropriately-named directory for your images (mkdir /data_local/<mydir>, substituting a suitable name for <mydir> ). They ll be saved in a zipped file with a name like ChRIS_yyyy-mmddUTChh-mm.zip where the date (yyy-mm-dd) and time (hh-mm) of your download should provide a unique identifier. Unzip this file ( unzip from the command line). You can have a quick look at the images using ds9 or gaia (each invoked these from the command line; ds9 or gaia); you may need to fiddle about to get a reasonable scaling (as a starting point in ds9, from the Scale drop-down menu, try setting Auto Cut at 98%, and Intensity Map to Histogram Equalization; in gaia the corresponding inputs are via the Auto Cut and Intensity Map buttons). 1 This processing is described in detail in the Spectrum extraction script. 2 The files will only be available on this particular computer, of course.

3 PHAS Making a colour image There are many ways to combine data taken through different filters. One standard approach is to combine R, G, and B images to set up the colours (which can have fairly low signal:noise), and use this to add colour to an L ( luminance ) image, taken through a broad-band filter that transmits more photons than the narrower colour filters and therefore normally of better signal:noise. 3.1 Making the RGB image The following instructions assume that you have images through three different filters e.g., RGB. If you have more, then you will need to combine then sequentially (as nearly all software presumes that images are made up of only red, green, and blue channels). Standard image-editing packages don t handle FITS images; we ll use ds9 for initial scaling and to convert the data, and the gimp for combining images. Both are available as free downloads for MacOSX and Windows platforms, 3 so, if you want to, you can take your images home on a USB stick (or download from the archive) and play with them there. 1. Read each (aligned and combined) FITS image in turn into the ds9 software and, from the drop-down menus, choose Scale -> Scale Parameters to adjust the histogram to bring out the detail you require. [To get started, try the Log and 98% display settings from the Scale drop-down menu before adjusting the limits in the Scale Parameters window.] 2. For each scaled image, File -> Save image as a 24-bit TIFF image; don t use any image compression when you save. 3. In the gimp, open all the TIFF images that are to be combined (typically, R, G and B; you may get warnings about 16-bit to 8-bit conversion, which you can ignore). 4. Select the red-channel image, and colour it by selecting Image -> Mode -> RGB, then Colours -> Colourify; choose a red colour. 5. Do the same with the images for the green and blue channels, picking the appropriate colour. 6. Select the red image and type Ctrl-L to bring up the Layers dialogue box. Double click on the Background label and change it to Red. 7. Select the green image, use Select All (Ctrl-A), and copy (Ctrl-C); then paste it (Ctrl-V) into the red image; rename the new layer by double-clicking on it, and calling it Green. 8. Select the blue image and do the same as for the green (calling it Blue). 9. In the Layers dialogue box, highlight the green layer and change the Layers Mode from Normal to Screen; then do the same for the blue layer. 10. Check that the images are well aligned. If not, you can turn different colour images off and on as required with the eye symbol, and use the move tool to align them relative to one another. 3 ESO has also produced FITS liberator, which is an arguably better alternative to ds9 for converting FITS images on Apple/Windows platforms.

4 4 Image Processing You can change the Screen slider bar to adjust the colour balance here, if it helps (you don t always need to): you might set green to be 65% for example, and blue to 35%. 12. Select Image -> Merge Visible Layers (Ctrl-M) to collapse the layers into one image. 13. Choose Colours -> Curves to make further adjustments to the colour balance. Needless to say, producing the final image can be a fiddly process, and is more an art than a science. You should be prepared to spend some time getting it right (or as nearly right as you can). You final product should be an image which has a colour rendition that looks good to you. Don t worry if it looks a bit noisy, as long as the colour rendition is suitable - the detail will come from the L image discussed below. (In fact, if you re worried that your RGB image looks too grainy or noisy, then you can apply some smoothing. From the Filters drop-down menu, choose Blur, and select Gaussian blur. Experiment with the blur radius (use the same value for the horizontal and vertical directions); a number like 10 is more likely to be useful than 1 or 100, but you can always undo anything you don t like with Ctrl-Z.) 3.2 Making an LRGB image You should now open the L image for your target (following steps 1 and 2 given above). If necessary, convert the image from Greyscale to RGB mode (from the Image drop-down menu, choose Mode, and select RGB). Note that this does not actually change the greyscale image to a colour image it merely increases the bit depth of the image so that it can accept colour information. You should make the L image look as good as you can quite a lot can be done using the Curves option in the Colours drop-down menu. This is again very subjective; be prepared to spend some time experimenting. You might (or might not) want to use the clone tool to try to fix up any obvious flaws, but don t get carried away to the point of inventing data! (The clone tool is part of the toolbox, where its icon looks like a little rubber stamp; ctrl-b should bring up the toolbox if it isn t already visible.) When you re happy with your L image, take a copy of the RGB image, and paste it as a new layer above your L image. (This can all be done from the Edit drop-down menu.) Check that the two images are well aligned (e.g., by setting the blending mode in the Layers dialogue box to Normal, and the Opacity to around 50%), and adjust if necessary although chances are that this won t be required. With the RGB image as the top layer, set the Opacity to 100%, and the blending mode to Colour. This will impose the colour from the RGB image onto the L image (which should be of much better signal:noise). From the drop-down Layer menu (not the Layers dialogue box), Merge down the two layers into a single standard image. You may (or may not...) want to make some final tweaks to the colour rendition, contrast, etc.

5 PHAS Limiting magnitude You should make a rough estimate of the limiting magnitude in your image. This section describes a quick-and-dirty, but still reasonably good, method; a more elaborate alternative is discussed in the Appendix. Fire up the Aladin Sky Atlas by typing aladin at the command line. 4 Enter your target name or co-ordinates in the Location bar to bring up an image. Next, click on the small open folder icon at top left of the main sub-window. Alternatively, use the Aladin File drop-down menu, choosing Load catalog, Vizier. This should open the Server selector dialogue window note that this may pop up behind another window, and so may not be immediately obvious! In the Server selector window, click the Surveys tab (on the right), highlight NOMAD1 from the list, and submit. You ll get an overlay on the image, showing all the stars from the (US) Naval Observatory Merged Astrometric Dataset, which includes approximate photometry to rather deep limits. Try to find the faintest star you can see in your image that s identified in Simbad. If you click on it in the Aladin window, you get some basic data, including magnitudes. Your colour image doesn t correspond well to any particular photometric passband, of course; but it s reasonable to suppose that, overall, it s roughly comparable to Johnson V or R. (You might consider trying to estimate limiting magnitudes in each of the RGB channels, correspondingly roughly to Johnson R, V, B passbands.) 5 Report preparation The following are some suggested points which you could discuss, if relevant for your data; lengthy reports are not required, but you should summarize what you did (including any problems encountered). A typical word-processed report might be one or two pages in length. You should hand in a hard copy of your report, and a copy of your final (jpeg) image to idh@star.ucl.ac.uk. The hard-copy should include a small version of your image (perhaps 5cm on a side; a bigger one just wastes printer ink), with the orientation and image size clearly indicated. [The actual orientation of your image doesn t matter, 5 as long as you indicate the directions of north and east, and the approximate width of the image in arc minutes you should be able to establish this information having compared your image with Sky Survey results from Aladin.] 1. You should give a table of basic observational parameters dates, exposure times, numbers of images in each filter. If you use your own data then you might include details for each image; if you use large numbers of exposures from the archive, you might just give some summary statistics (including the total integration time in each filter). 2. What are the typical counts in the object of interest in a typical image, and in the background? (You should get values using ds9 or gaia.) Does it depend strongly on filter? 4 If you re working on this away from UCLO, then go to for an on-line version. 5 Astronomical images are often displayed with north at top and east to left, but that s just a convention.

6 6 Image Processing How is the noise reduced by combining images (just from inspection)? Do certain filtered images appear particularly noisy relative to others obtained through a different filter? Does this correlate with counts? 4. Explain how contrast stretch was used to adjust colour balance during RGB colour combine. What governed your final choice of colour balance? 5. Briefly describe what your image actually shows. Are the results you obtain consistent with what you know about the objects? (E.g., is a nebula the colour expected from what you know about the line-emission processes which are going on?) 6. Evaluate the final images: how clean are they? Are the colours right? Are there residual flat-field features? Hot pixels? Cosmic rays? Trailing of stars? 7. State your estimate of the limiting magnitude of your image (with some rough indication of the associated uncertainty). An ambitious student might attempt to estimate the limiting magnitudes in the individual merged R, G, and B frames; these filters correspond reasonably well to Johnson R, V, and B passbands. Initial author: SJF, IDH, 2/2013 Other contributors: Latest revision: IDH, 3/2016

7 PHAS Appendix: precise magnitude measurements in GAIA For those with the time and inclination, quantitative measurements of star brightnesses can be obtained reasonably easily using gaia. 1. Open a FITS image in gaia (this would normally be the median image in L, R, G, or B), and from the Image Analysis pull-down menu, select Aperture photometry and Results in data counts. In the new window which appears, pull down the Options menu, and select Keep apertures same size (this is very important). 2. To draw an aperture for measuring, click the Define object aperture button and then click on a star; a small aperture will be displayed. You can set the size of the aperture in the semimajor axis entry box in the photometry window, and scale the sky annulus also. A suitable aperture size would be about 2 3 times the FWHM of a star image (use the slice tool [from the View drop-down menu] to estimate this). The sky annulus should include sky in the region immediately around the source, but should exclude starlight from the source itself. 3. Click the Parameters tab in the photometry window and set the Sky estimator to mode. This will avoid problems due to stars falling within the sky annulus. 4. When measuring many star images at the same time, all one has to do is to define a new aperture for each star and drop it in place over the star to measured. For each star, click Define aperture and then click on the star. If you set an aperture on the wrong star, you can drag it to select the correct one at any time. Do not worry if the apertures are not precisely centred; the program centroids the apertures on the star images when the results are calculated. 5. When all apertures are in place, click on Calculate results. All the results can be viewed by selecting the Results tab in the photometry window and clicking the button to show them all. You can save your measurements in a text file. 6. If you wish to measure more than one image of the same field, the same apertures can be reloaded (so you only have to define them once). The stars for which you might want to make measurements of counts would be at least one for which you have a magnitude measurement from Simbad, and perhaps two or three that are among the faintest you can securely detect. We can relate the gaia measured flux in counts, N, to magnitude, m: m = 2.5 log N + k. (1) Use a known reference star to calculate the constant k for the frame. (You should match the magnitude system as closely as reasonably possible to the image colour; e.g., V for G.) Thence estimate the limiting magnitude in the frame (by calculating the apparent magnitude of your faintest targets). The photometry output from gaia also gives you a mean sky flux per pixel, which can be used to determine the sky brightness. First determine the mean sky flux per square arcsecond (the image scale for the RoboScope is 0.86 arcseconds per pixel), then use equation 1 to calculate the sky magnitude per square arcsecond.