diff options
Diffstat (limited to 'ds9/doc/ref/how.html')
| -rw-r--r-- | ds9/doc/ref/how.html | 159 |
1 files changed, 159 insertions, 0 deletions
diff --git a/ds9/doc/ref/how.html b/ds9/doc/ref/how.html new file mode 100644 index 0000000..9a76d51 --- /dev/null +++ b/ds9/doc/ref/how.html @@ -0,0 +1,159 @@ +<!DOCTYPE doctype PUBLIC "-//w3c//dtd html 4.0 transitional//en"> +<html> + <head> + <meta http-equiv="Content-Type" content="text/html; + charset=windows-1252"> + <meta name="GENERATOR" content="Mozilla/4.78 [en] (X11; U; Linux + 2.4.7-10 i686) [Netscape]"> + <title>How</title> + </head> + <body link="#0000ee" alink="#ff0000" bgcolor="#ffffff" text="#000000" + vlink="#551a8b"> + <h3><img alt="" src="../sun.gif" height="98" align="middle" + width="100"> How it Works</h3> + <blockquote> + <p><b>Table of Contents</b></p> + <a href="#How">How DS9 Renders an Image</a><br> + <a href="#Scales">Scales</a><br> + <a href="#Smoothing">Smoothing</a><br> + <a href="#Contours">Contours</a><br> + <a href="#LargeFiles">Large Files</a><br> + <p><b><a name="How"></a>How DS9 renders an image</b></p> + <p>Here is a short description on how DS9 decides to paint a pixel + a color on the the screen, give an data value... you need a + color scale, a contrast/bias pair for the colorscale, clip + values for the data, a scale distribution, and finally, the + value of the pixel in question.</p> + <blockquote> + <p>Step 1. Select a color scale. A color scale is defined as a + number of colors (RGB triplets). The number of RGB triplets + can vary from just a few to over 200. DS9 contains a number of + predefined color scales (Gray, A, B, I8, ...) or the user may + load his own color scale.</p> + <p>Step 2. Apply a contrast/bias pair. This step takes the + result of step 1 and creates a new array with the + contrast/bias applied. The length of the new array will + between 200 (for pseudocolor) and 4096 (for truecolor).</p> + <p>Step 3. Calculate the data clip values (low/high data + values). The min/max data values may be used or an algorithm + may be used to determine the clip data values.</p> + <p>Step 4. Apply the scale distribution. This involves taking + the result of step 2, and creating yet another array, this + time of size 16384, redistributing the colors, based on the + scale algorithm selected (see <a href="Scales">Scales</a>).</p> + <p>Step 5. Based on your data clip values, and the value of the + pixel you have, index into the result of step 4, and you have + an index into lookup table (for pseudocolor) and an RGB pair + (for truecolor and postscript).</p> + </blockquote> + <p><b><a name="Scales"></a>Scales</b></p> + <p>The <tt>log</tt> function is defined as the following:</p> + <blockquote> + <p><b><img src="img/log.png" alt="log equation" height="32" + width="78"></b></p> + </blockquote> + <p>as <i>x</i> goes from 0 to 1. The user may specify an exponent + <i>a</i> to change the distribution of colors within the + colorbar. The default value of <i>a</i> is 1000. Typically, + optical images respond well at 1000, IR images as low as 100, + and high energy bin tables up to 10000. A value of 10000 closely + matches the <b><tt>log</tt></b> function of SAOImage as defined + as the following:</p> + <blockquote> + <p><b><img src="img/saolog.png" alt="SAOImage log equation" + height="34" width="65"></b></p> + </blockquote> + <p>The <tt>pow</tt> function is defined as the following:</p> + <blockquote> + <p><b><img src="img/pow.png" alt="pow equation" height="30" + width="51"></b></p> + </blockquote> + <p>as <i>x</i> goes from 0 to 1. The user may specify an exponent + <i>a</i> to change the distribution of colors within the + colorbar. The default value of <i>a</i> is 1000.</p> + <p>The <tt>sqrt</tt> scale function is defined as the following:</p> + <blockquote><img src="img/sqrt.png" alt="sqrt equation" + height="21" width="42"><br> + </blockquote> + <p>as <i>x</i> goes from 0 to 1. </p> + <p>The <tt>square</tt> scale function is defined as the + following:</p> + <blockquote><img src="img/square.png" alt="square equation" + height="21" width="35"><br> + </blockquote> + <p>as <i>x</i> goes from 0 to 1.<br> + </p> + <p>The <tt>asinh</tt> scale function is defined as the following:</p> + <blockquote><img alt="asinh" src="img/asinh.png" height="29" + width="80"><br> + </blockquote> + <p>as <i>x</i> goes from 0 to 1. </p> + <p>The <tt>sinh</tt> scale function is defined as the following:</p> + <blockquote><img alt="sinh" src="img/sinh.png" height="29" + width="69"><br> + </blockquote> + <p>as <i>x</i> goes from 0 to 1. </p> + <p>The <tt>histogram equalization</tt> scale function distributes + colors based on the frequency of each data value.</p> + <p><b><a name="Smoothing"></a>Smoothing</b></p> + <p>The user may select one of three types of smoothing kernels. + The parameter, <i>r</i> or <tt>kernel radius</tt>, is defined + as the following:</p> + <blockquote> Boxcar function, where the width = 2<i>r</i>+1<br> + Tophat function, where the radius = <i>r</i> and the diameter + of kernel is 2<i>r</i>+1<br> + Gaussian function, defined as: + <blockquote><img src="img/gauss.png" alt="Gaussian Equation" + height="38" width="173"><br> + </blockquote> + where the mean = 0 and sigma =<i> r</i>/2, and the diameter of + kernel is 2<i>r</i>+1 </blockquote> + <p><b><a name="Contours"></a>Contours</b></p> + <p>The contour algorithm is from an unknown author and originally + came from FV. The difference between the two modes are:<tt><br> + </tt></p> + <blockquote><tt>block</tt> : the image is blocked down before the + contour is generated <br> + <tt>smooth</tt> : the image is smoothed via a Gaussian kernel + before the contour is generated. </blockquote> + <p><tt>block</tt> mode is faster as the smoothing parameter + increases. Inversely, <tt>smooth</tt> mode is much slower as + the smoothing parameter increases.</p> + <p><b><a name="LargeFiles"></a>Large Files</b></p> + There are several factors that determine if DS9 will be able to + load a large file.<br> + <p>32 bit OS vs 64 bit OS : to address very large files, you may + need to use an 64 bit OS with a 64bit version of DS9. 32bit apps + can address up to 4Gb of address space. However, depending on + the OS, this limit may be less. Linux for example, the limit + appears to be ~3GB (the OS and shared libs eat up a lot of + address space). Under 64bit Solaris, 32bit ds9 has a full 4Gb of + space. MacOSX appears to have a limit ~3Gb. Under windows, ~2Gb.</p> + <p>Large File Support: is the ability to sequence thru files + larger than 4Gb. DS9 is compiled with LFS.</p> + <p>File system: the OS file system must be able to support files + larger than 4Gb. Most recent file systems fully support 4GB>.</p> + <p>Memory Management: There are a number of memory management + techniques supported in DS9 that will greatly affect the ability + and speed of loading large files:</p> + <blockquote> <tt>$ ds9 foo.fits # uses mmap</tt><br> + <tt>$ cat foo.fits | ds9 - # allocates memory</tt> <br> + <tt>$ xpaset -p ds9 file foo.fits # uses mmap</tt> <br> + <tt>$ xpaset -p ds9 fits foo.fits # allocates memory</tt><br> + </blockquote> + <p>Memory Map (<tt>mmap</tt>) is very fast, limit is memory + address space (see above). Allocate is very slow, limit is + amount of physical memory + swap partition.</p> + <p>Scanning Data: DS9 needs to determine the min and max data + values to correctly display your image. For large files, such as + Mosaics and Data Cubes, this can take time. You have the + option of using using the FITS keywords DATAMIN/MAX or + IRAFMIN/MAX - great if present, bad because they are always + wrong. Another option is to specify the low and high clip values + via the command line or Scale dialogbox.<br> + </p> + <blockquote> <br> + </blockquote> + </blockquote> + </body> +</html> |