To appear in AJ. A Preprint typeset using L TEX style emulateapj v. 11/26/03 IMAGE-PROCESSING TECHNIQUES FOR THE CREATION OF PRESENTATION-QUALITY ASTRONOMICAL IMAGES Travis A. Rector University of Alaska Anchorage, 3211 Providence Drive, BMB 212, Anchorage, AK 99508 Zoltan G. Levay and Lisa M. Frattare Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218 Jayanne English Department of Physics and Astronomy, University of Manitoba, Winnipeg, MB R3T 2MB, Canada and Kirk Pu'uohau-Pummill Gemini Observatory, 670 N. A'ohoku Place, Hilo, HI 96720 To appear in AJ. ABSTRACT The quality of modern astronomical data, the power of modern computers and the agility of current image-processing software enable the creation of high-quality images in a purely digital form. The combination of these technological advancements has created a new ability to make color astronomical images. And in many ways it has led to a new philosophy towards how to create them. A practical guide is presented on how to generate astronomical images from research data with powerful image- processing programs. These programs use a layering metaphor that allows for an unlimited number of astronomical datasets to be combined in any desired color scheme, creating an immense parameter space to be explored using an iterative approach. Several examples of image creation are presented. A philosophy is also presented on how to use color and composition to create images that simultane- ously highlight scientific detail and are aesthetically appealing. This philosophy is necessary because most datasets do not correspond to the wavelength range of sensitivity of the human eye. The use of visual grammar, defined as the elements which affect the interpretation of an image, can maximize the richness and detail in an image while maintaining scientific accuracy. By properly using visual grammar, one can imply qualities that a two-dimensional image intrinsically cannot show, such as depth, motion and energy. In addition, composition can be used to engage viewers and keep them interested for a longer period of time. The use of these techniques can result in a striking image that will effectively convey the science within the image, to scientists and to the public. Subject headings: techniques: image-processing 1. INTRODUCTION But often images are now made from datasets that are For many decades astronomical color images have been either outside the optical window or do not match the generated using large-format photographic plates and characteristics of the color-detecting cones in the human traditional darkroom techniques, e.g., Malin (1992). In eye. the early 1980s, charge-coupled device (CCD) detectors The development of advanced astronomical instrumen- began to replace photographic plates as the instrument tation has been contemporaneous with the advance- of choice for astronomical research. However, until re- ment of computing power and, in particular, digital cently CCD arrays lacked the number of pixels neces- image-processing (IP) software for commercial applica- tions. These IP programs, e.g., Adobe r Photoshop r 1, sary to compete with the fine grain of photographic plate 2 emulsion. In recent years CCD detectors have grown in Photoshop r Elements and The GIMP , offer unprece- size and the physical size of pixels has decreased. And dented power, flexibility and agility in digital image gen- mosaics of CCD arrays have been implemented in many eration and manipulation. The combination of these instruments. The large number of pixels in these cameras technological advancements has led to a new ability to now allows for high-quality optical images to be gener- make color astronomical images. And in many ways, ated in a purely digital form. it has created a new philosophy towards how to create Furthermore, the continuous improvement of imaging them. No illustration of this is more apparent than the capabilities in non-optical windows of the electromag- famous Hester & Scowen (1995) \Pillars of Creation" im- netic spectrum have enabled the creation of high-quality age of the central region of the Eagle Nebula (M16), with images at other wavelengths as well. Historically, astro- 1 Adobe and Photoshop are either registered trademarks or nomical images have been made by combining grayscale trademarks of Adobe Systems Incorporated in the United States images taken through red, green and blue optical filters. and/or other countries. 2 The GIMP is written by Peter Mattis and Spencer Kimball, Electronic address: [email protected] and released under the GNU General Public License. 2 Rector, Levay, Frattare, English & Pummill the Hubble Space Telescope (HST). This image demon- and a color management workflow is established, steps strated the tremendous resolution of the HST Wide-Field that are described in x2.6.1 and x2.6.2. Planetary Camera 2 (WFPC2) camera. It also showed The techniques described herein assume that more how narrow-band imaging can change our view of an ob- than one dataset, preferably three or more, will be com- ject. And it demonstrated how color schemes can be used bined to produce a color image. A dataset is defined to imply depth, motion and texture in an astronomical as a two-dimensional image of a particular waveband, image. Just as importantly, it illustrated how effectively polarization or other distinct characteristic; e.g., an op- such images can inspire the public and generate enthusi- tical image through a single filter, or a radio image at asm for astronomy in general. or in a particular waveband and/or polarization. These The success of the HST image of M16, and images like techniques are designed to take advantage of the dis- it, inspired the creation of the Hubble Heritage Project tinct structural information in each dataset for which (Noll 2001). Since its inception in 1998, the project has they were obtained. For comparison, a popular technique released a new color image from HST every month. And among amateur astronomers is known as the LRGB the success of the Hubble Heritage project has inspired method, e.g., Gendler (2003), wherein an image is first many other observatories to invest resources into the cre- generated in the traditional \natural color" scheme (see ation of images from astronomical research data that are x3.2.1) from datasets obtained through red, green and primarily intended for the lay person. Many scientists blue filters. To improve the image quality, an unfiltered, have also become interested in ways to create such im- \luminosity" image is added that lacks color information ages from their data, not only for public consumption but but has a higher signal to noise ratio. This technique also as a visualization tool for colleagues. The goal of this is well suited for small-aperture telescopes because most paper is to demonstrate many of the techniques used by objects are limited by relatively poor signal to noise. The image creators of the Hubble Heritage team, the National LRGB method is effective for decreasing the noise in an Optical Astronomy Observatory, the Gemini Observa- image, but results in a loss of wavelength-specific struc- tory, the Canadian Galactic Plane Survey and others to tural information. This method is not well suited for use generate astronomical images from research-quality data. with scientific data, which is rarely obtained unfiltered. This paper is intended for a range of audiences, from Indeed, narrow-band observations are obtained specifi- professional astronomers to outreach specialists. The cally to increase the contrast between emission-line and procedural steps, from projecting the data into images continuum-emission regions of an object. Thus the qual- to the final electronic and print output, are discussed ity of an image can actually be improved by the exclusion in x2. This section serves primarily as an instructional of particular wavebands. guide. It will be of most interest to those with little The data must first be fully reduced, e.g., optical data image-processing experience who wish to learn how to are bias and flat-field corrected, with a standard data generate a color image from their data. As many of the reduction package, such as IRAF3, IDL4 or AIPS5. The steps described herein are subjective, different philoso- data must also be stored in a data file, e.g., FITS for- phies regarding image creation are discussed in x3. This mat, that can be projected into an image. The image includes details on how to select colors and use composi- will then be imported into an IP program which can tion to engage the viewer. This section will be of greater handle layers, e.g., Photoshop, Photoshop Elements or value to those who have prior experience in assembling The GIMP. The steps in this section will be discussed images but wish to improve their skills, although in many in general terms, without reference to a particular soft- ways the steps described in x2 will be more clear after ware program, where possible. A straightforward exam- reading x3. Step-by-step examples are provided in Ap- ple using IRAF, Karma (Gooch 1995) and The GIMP pendices A & B. These are of particular value as tutori- is given in Appendix A. And a more complex example als on how to implement the steps that are described using IDL and Photoshop is given in Appendix B. The in mostly general terms in x2. Novice image makers reader is encouraged to read both examples, as differ- are encouraged to work through the examples in Ap- ent techniques are illustrated in each. It is important pendices A & B after reading x2 and before reading x3. to note that the ESA/ESO/NASA Photoshop FITS Lib- Appendix C provides examples of the cosmetic cleaning erator (Christensen et al. 2004) is a very useful plug-in steps often necessary on images generated from optical that can scale and project FITS datafiles directly into and near-IR data.