Python code to convert FITS files to images (including the OpenEXR format) - Min-Su Shin (msshin @ kasi.re.kr)

I present a simple Python code to scale FITS images by using linear, sqrt, pow, log10, asinh function, zscale, logistic distribution as well as histogram equalization, and then save them to image files such as PNG. The code also includes simple methods to measure sky levels and to find adequate ranges of pixel values for the scaling routine. I hope that the code is useful in your works. I always welcome comments. Please, note that these Python codes are presented as examples rather than a complete software package. The img_scale.py can be easily incorporated to your own code for various scaling methods. In the given examples, you can figure out how several Python graphics modules (SciPy, Matplotlib, and Python Image Library) can be used to convert FITS files to conventional image files.

The image scaling is given in the module file img_scale.py. There are multiple example Python codes that generate RGB color images and generate single-band images. Please, read Lupton et al. (2004) about how to produce RGB images and scaling methods, and Lupton et al. (1999) about the asinh scaling. If you are looking for general introduction to this problem, you may find the Youtube video explanation "How scientists colorize photos of space" useful. The code presented here is not for interactive production of images. I also suggest people to check STIFF, which is quite useful in producing high-quality picture files, APLpy, and dmimg2jpg in CIAO, which support multiple scaling relations to produce RGB images. f2n.py is also a simple useful tool written in Python to convert FITS images to PNG files by using Python Image Library.

Conversion from astronomical images to conventional images consists of two steps as commonly used in tone mapping with high dynamic range images. In the first step, you can change the pixel values of FITS images to values between 0.0 and 1.0 by using img_scale.py. In the second step, these values are divided into 256 intervals for 8bit images, and then are saved as graphic files.

Download : public_fits_image.tar.gz

The above file has example codes and FITS files as example data, as well as img_scale.py. The code needs NumPy. PyFITS is also required to read FITS files. The code build_rgb_w_asinh_SciPy.py uses SciPy to produce image files. Meanwhile, the code build_single_matplotlib.py and build_rgb_w_asinh_matplotlib.py use Matplotlib instead of SciPy. If you want to use Python Image Library (PIL) instead of Matplotlib and SciPy, check build_rgb_w_asinh_PIL1.py or build_rgb_w_asinh_PIL2.py. When you want to change color maps for a single-band image by using Matplotlib, look up the tables for the color maps (1D and 2D). The following are example pictures. When producing the following pictures, the pixel values of the FITS files were scaled down to some ranges which are divied into 256 intervals equally. Pixel values for images are assigned as nearest value for the 256 intervals. You might be interested in Matplotlib color maps to visualizae a single-band image by using Matplotlib. You also might be interested in interactive tools which can read high-dynamic range FITS images and can convert them to other formats. DS9 and Aladin are well-known tools in the astronomy community. ImageJ can be also used for this purpose with lots of image analysis methods.

* FAQ
- How can I change the size of output images?
If you use Matplotlib to produce images, use the statements such as the following: 

fig = pylab.gcf()
fig.set_size_inches(3.4,3.4)
rgb_array[:,:,0] = r 
rgb_array[:,:,1] = g 
rgb_array[:,:,2] = b 
print "Plotting a RGB image of (", width,",",height,")"
pylab.imshow(rgb_array, interpolation='nearest', origin='lower')
pylab.axis('off')
pylab.savefig(out_fn, dpi=(300))
where sizes are given in inches, and dpi is set to change the resolution. If you use Python Image Library to generate images, you can use resize function as the following does. 
use_image = Image.new('RGB', size=(height, width))
use_image.putdata(PIL_data)
use_image = use_image.resize((height/2,width/2))
use_image.save('rgb.png')
SciPy also supports the resize function as the following shows if you want to use SciPy instead of Matplotlib or PIL. 
rgba_array = numpy.empty((width,height,4), numpy.uint8) # assuming 8 bits per channnel
rgba_array[:,:,0] = scipy.misc.bytescale(r) # red
rgba_array[:,:,1] = scipy.misc.bytescale(g) # green
rgba_array[:,:,2] = scipy.misc.bytescale(b) # blue
rgba_array[:,:,3] = 255 # Alpha transparency
rgba_array = scipy.misc.imresize(rgba_array, (width/2, height/2), interp='bicubic')
scipy.misc.imsave('rgb.png', rgba_array)

- Some examples produced by others
- How can I produce a single-band image instead of a RGB image?
Check the example Python codes build_single_matplotlib.py, build_single_PIL_w_matplotlib.py, and build_single_Gnuplot.py which are included in the above tar.gz file. In principle, there are many ways to produce a single-band image after scaling by using img_scale.py. The example codes just show two possible ways.


linear


sqrt


log10


asinh(beta=1.0)


asinh(beta=5.0)


asinh(beta=20.0)


RGB type 1


RGB type 2


linear


sqrt


log10


asinh(beta=1.0)


asinh(beta=5.0)


asinh(beta=20.0)


histogram equalization


logistic distribution


RGB type 1


RGB type 2


RGB type 3

* Conversion to the OpenEXR format
It is also possible to convert FITS files to OpenEXR files, and then you can use tools for high dynamic range (HDR) images. The file fits2exr.py shows how you can do the conversion with PyFITS and Python binding of the OpenEXR library. The following images show how the converted EXR files can be used with exrtools. The original FITS files are SDSS DR8 images of NGC 5548 in g, r, and i-bands. 

exrnormalize galaxy.exr galaxy.n.exr
exrtopng galaxy.n.exr galaxy.n.png

exrpptm -c 1.0 galaxy.n.exr galaxy.pp.exr
exrnormalize galaxy.pp.exr galaxy.pp.n.exr
exrtopng galaxy.pp.n.exr galaxy.pp.png
exrblur galaxy.n.exr galaxy.b4.exr 0.25 exrchr galaxy.n.exr galaxy.b4.exr galaxy.chr.exr 0.1 exricamtm galaxy.chr.exr galaxy.b4.exr galaxy.icam.exr exrnormalize galaxy.icam.exr galaxy.icam.n.exr exrtopng galaxy.icam.n.exr galaxy.icam.png 
exrnlm galaxy.n.exr galaxy.b4.exr galaxy.nlm.exr
exrtopng galaxy.nlm.exr galaxy.nlm.png

If you want to see the converted EXR image of NGC 5548 by using HDR HTML viewer which is generated by pfsouthdrhtml of pfstools, click here!
ImageMagick and GEGL also support the OpenEXR format. Luminance HDR (a.k.a. Qtpfsgui) is a wondergul GUI program to process HDR images including OpenEXR and FITS formats. If you are looking for a tool to visualize FITS on Web interactively, you may find JS9 and jsFITS interesting.