A Quick Start to Using MR/1 Version 2.0

The Very Start

MR/1 Version 2.0 extends Version 1.0 substantially with more wavelet and multiscale transforms, various extra image processing tool commands, lossless compression, support tools for large images, commands for edge detection, and multifractal analysis functionality. A new package, MR/2, is available separately, with functionality for filtering, deconvolution, source detection and other operations based on multiscale entropy.

On the MR/1 Version 2.0 CD which also contains the codes for MR/2 Version 1.0, you will find the executables (for varying operating systems - Dec, HP, Linux, IBM, and Sun; and Windows/DOS). Here are the top directories which you will find on the CD:

COPYRIGHT           doc/                javadec.tgz         windows/
COPYRIGHT.windows   idl/                javamr1/
README              idlv5.tgz           javamr1.tgz
README.windows      javadec/            unix/

The IDL directory contains various subdirectories, one of which (pro) is the suite of programs you will use to run MR/1 directly in IDL. Another subdirectory contains the public suite of programs including inter alia the readfits and writefits IDL programs. A further IDL directory contains help files for the MR/1 IDL programs.

The javadec directory contains a demonstration Java applet for selective decompression of a previously compressed image. The javamr1 directory contains the Java user interface.

The windows and unix directories each contain the license manager, and the executables, for (respectively) W95/W98/WNT and a range of Unix platforms. To activate your license, you will first need to tell us (preferably by email to fmurtagh@multiresolution.com, or to multires@hotmail.com or fmurtagh@acm.org) your hostname (xxx.yyy.edu or whatever) and hostid. The latter can be obtained on either Windows or Unix platforms by running the appropriate mrhost program, available in either the windows or unix directory.

Please also let us know your operating system, whether your license is to be valid indefinitely or is a limited time demo, whether your license is for all machines of the same operating system, whether multi-user, and finally whether MR/1 alone, or both MR/1 and MR/2, are wanted.

Two remarks on the license manager. Firstly, if your machine is not networked, then you will have to set the value of the environment variable CEA_LICENSE_HOST to localhost. Secondly, if there is any problem in running the license manager, cea_lm, let us know immediately (since optimization compilation options used caused a problem in the past on some systems) and we will explain what needs to be done.

When we return the license key to you, you can install your license manager.

First Steps

a = readfits('ngc2997.fits')
write_gif, 'ngc2997.gif', bytscl(a)

im_convert -b ngc2997.fits ngc2997.gif

With IDL one can display images of course. Otherwise you can use xv or ImageMagick's display program under Unix, or Photoshop under Windows, to display GIF files. To directly display FITS files, one possibility is to use SAOimage, which is publicly available for a range of Unix platforms. Another FITS viewer program for MS Windows, Apple Mac and Unix platforms is available at http://www.cv.nrao.edu/~bcotton/fitsview.html.

Some FITS Images for Downloading

• NGC 2997 galaxy. FITS image (267840 bytes).
• Point pattern clusters. FITS image (267840 bytes).
• Mammogram. FITS image (1054080 bytes).
• R Aqr jet. FITS image (265024 bytes).
• NGC 2997 again, rotated. FITS image (267840 bytes).
• Lena test image. FITS image (265024 bytes).
• Comet core. FITS image (1088640 bytes).
• Galaxies and stars - small objects. FITS image (135360 bytes).
• M51. FITS image (36840 bytes).
• Sky simulation. FITS image (267840 bytes).
• A pre-refurbishment HST PSF. FITS image (265024 bytes).
• X-ray image (cluster A2390). FITS image (1054080 bytes).
• Finally a 384-valued spectrum (5760 bytes.)

A Range of Examples

• For visualization of the cometary core, using 6 resolution levels (-n 6), and writing the different level images out (-x) for independent display: mr_transform -n 6 -x comet.fits comet. The file comet.mr contains the wavelet transform (by default à trous method).

• For the point pattern data, the significance limits for detection were set up with mr_abaque -d. This was followed by the determining of the support image at each of 6 resolution scales, mr_psupport -I gauss2.fits -s ptpatt -p ptout. The support images were extracted using mr_extract -x ptout.mr ptout.

  Another important command is mr_detect which not only detects the clusters, with support for lots of noise models, but also calculates and outputs parameters for all objects or `blobs' found. The objects or blobs can be at different resolution scales, or a combination of them.

• In the NGC2997 image, we looked for structures on different scales which were based on the wavelet transform.

  To detect structures at levels 1 and 2, i.e. essentially stars, we use mr_detect -n 7 -L 2 ngc2997.fits im_stars.fits Options -n 7 -L 2 indicate, respectively, 7 scales for the analysis, and scales 1 and 2 for object detection. To detect larger-scale structures, we use mr_detect -n 7 -F 3 -L 6 -K ngc2997.fits im_gal.fits The option -K keeps objects close to the image boundaries which would otherwise (for safety) be ignored, and the -F 3 option implies scale 3 is the starting scale used for object detection. Extended structures were obtained by ngc2997 - im_stars (for example, in IDL, followed by writing of the images as GIFs).

• Compression of the mammogram image was carried out using mr_comp -R 20 mamo.fits mamo.fits which produces the compressed file, mamo.fits.MRC. Decompression is carried out by the following command: mr_decomp -i mamo.fits.MRC -o mamo.fits

• The X-ray image was filtered based on mr_filter -m 9 which, for a 512x512 images takes about 2-3 minutes elapsed time.

• The Earth observation example, using the urban/rural scene around Derry, used mr_transform -t 5. The images were then histogram-equalized and dimmed using xv.

• Using mr_pfilter on sparse image data (i.e. many pixels were zero-valued - this data is being working on currently and will be made available here at a later point) and finding that the filtered images looked devoid of almost any structure, we checked the input data, and found that it had been preprocessed such that pixel values were of the order of 1.0 or 1.5. This is not in keeping with the assumption of Poisson noise. We instead used mr_filter with a model of undefined noise in this case.

• For various industrial vision inspection problems, where the images contain lots of structure, and where the faults to be discovered consist of small irregular blemishes, we have found that the multiscale morphological (min-max) transform performs very well. Some resolution level, or levels, allow us to easily find and subsequently measure the blemishes.

• For the example of microstructure of concrete using SEM (scanning electron microscope) images, we used mr_rfilter -n 3 -t 2. The images were histogram equalized for display in IDL. A general remark - we have found speckle removal to be far more effective is a relatively low value of paramter -n is used, as we do here.

• More descriptions of commands used will follow, based on the "Palette of Multiresolution Applications" (http://hawk.infm.ulst.ac.uk:1998/multires), or the book "Image and Data Analysis: the Multiscale Approach" (Starck, Murtagh and Bijaoui, Cambridge University Press, 1998), or the document "Multiresolution and Its Applications: an Overview" (Starck, 1998).

Bugs - Corrections

• The mr_visu option

  -v Type_Visu
  

  should read

  -V Type_Visu
  

  (not least because option -v is already used for Verbose).

• See anonymous ftp area

  strule.cs.qub.ac.uk

  directory area pub/updates

  for other updates, categorized by operating system.