1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
|
<!DOCTYPE doctype PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<html>
<head>
<meta name="generator" content=
"HTML Tidy for Mac OS X (vers 31 October 2006 - Apple Inc. build 15.18.1), see www.w3.org">
<meta http-equiv="Content-Type" content=
"text/html; charset=us-ascii">
<meta name="GENERATOR" content=
"Mozilla/4.78 [en] (X11; U; Linux 2.4.7-10 i686) [Netscape]">
<title>How</title>
</head>
<body vlink="#551A8B" text="#000000" link="#0000EE" bgcolor=
"#FFFFFF" alink="#FF0000">
<h3><img alt="" src="../sun.gif" width="100" height="98" align=
"middle"> 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" id="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" id="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" width="78" height=
"32"></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" width=
"65" height="34"></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" width="51" height=
"30"></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" width="42"
height="21"><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" width=
"35" height="21"><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" width="80" height=
"29"><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" width="69" height=
"29"><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" id="Smoothing"></a>Smoothing</b></p>
<p>The user may select one of three types of smoothing kernels. Two
parameters, <i>r</i><i> (radius)</i> and <i><tt>k</tt></i>
<i>(kernel size</i><i>)</i>, are defined as the following:</p>
<blockquote>Kernel width/height = <i>2k+1</i> with <i>r</i>
<i><=</i> <i><tt>k</tt></i><br>
Boxcar function, width = <i>2</i><i>r</i><i>+1</i><br>
Tophat function, where the diameter = <i>2r</i><i>+1</i><br>
Gaussian function, defined as:
<blockquote><img src="img/gauss.png" alt="Gaussian Equation" width=
"173" height="38"><br></blockquote>
where the mean = 0 and sigma = <i>r/2</i><br></blockquote>
<p><b><a name="Contours" id="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" id="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>
|
