mainpixmaptest.html

Delaunay三角形的网格剖分程序

<TITLE>COG 2.0: Pixmaps</TITLE>

<H1>Pixmaps</H1>

<PRE>
#include "<A HREF="cogenpixmap.hxx">cogenpixmap.hxx</A>"
</PRE>

The class <B>CogenPixmap</B> is defined in
 <A HREF="cogenpixmap.cxx">cogenpixmap.cxx</A>.  In our example, we
read a 2D pixmap (PPM format, 512 x 259 pixels)

 <P><IMG SRC="pixmapinput.gif">

 <P>and create the following grid with 21879 nodes:

 <P><IMG SRC="pixmaptest.gif">

 <P>The refinement near the boundary we have made slightly more
accurate than pixel size.  This gives the correct geometry even for
the highly non-convex parts of this picture:
 <P><IMG SRC="pixmap1test.gif">
 <IMG SRC="pixmap2test.gif">

<H2>Reading the pixmap</H2>

 <P>The idea is defined by the principle <B>one color - one
region</B>.  If you have more than necessary or a continuum of colors,
the algorithm works too: close enough colors will be identified.
A coarse upper bound is <B>maxcolors</B>:

<PRE>
#include "<A HREF="coglib.hxx">cog/coglib.hxx</A>"
int main()
{
  cogIndex maxcolors=20; // maximal number of colors allowed
  cogenPixmap gen = new CogenPixmap("cog/pixmapinput.ppm",maxcolors);
</PRE>

 <P>Currently we support only <B>PPM</B> format, but it is easy to
transform pictures into this format (I use <B>xv</B> for this
purpose).

<H2>Specification of the box containing the picture</H2>

 <P>Now, let's define the box.  You possibly want to have a
non-distorted result but don't know the pixel numbers in above
directions.  For this purpose, you can define the range in one
direction (setHorizontal or setVertical) and define a single value for
the other direction with specification of the alignment (alignMinimum,
alignCenter, alignMaximum):

<PRE>
  gen-&gt;setHorizontal(-1,1);
  //  gen-&gt;setVertical(-1,1);
  gen-&gt;alignCenter(0);
</PRE>

 <P>If you want to specify all values (possibly with distortion of the
picture) simply use setHorizontal and setVertical.

<H2>Identifying colors and regions</H2>

 <P>To identify the colors with region numbers, define a list of
colors and their related region numbers:

<PRE>
  int listlength=2, region[3];
  wzColor color[3], black(""), blue("#11f"), yellow("#a70");
  color[0] = black;
  color[1] = blue;	region[1] = wzRegion(3);
  color[2] = yellow;	region[2] = wzRegion(2);
  gen-&gt;fitColorList(listlength,color,region);
</PRE>

 <P>To find out which colors have been found you can print out the
colors with <B>printColors</B>:

<PRE>
  gen-&gt;printColors();
</PRE>

<H2>Refinement</H2>

 <P>Now, you have to be careful with the refinement. In our case, the
refinement near the faces defined by <B>faceRefinementGlobal</B> is
approximately pixel size:

<PRE>
  gen-&gt;refinementGlobal(0.2,0.2);
  gen-&gt;faceRefinementGlobal(0.005,1.0);
</PRE>

 <P>Thus, once a boundary has been detected, it will be found accurate
enough.

<H2>Special Points</H2>

 <P> But there are some regions (islands and lakes) too small to be
found with certainty with the global refinement used here.
Especially, the lake near the center will not be found.  To catch this
lake, we add the coordinates of one point inside the lake:

<PRE>
  gen-&gt;addPlanesOfPoint(-0.0901,-0.208);
</PRE>

 <P>Now, we generate the grid and write it out as usual:

<PRE>
  wzgrid grid = (*gen)();
  grid->write("test.sg");
}
</PRE>

<H2>Warnings</H2>

<A NAME="badParameter"><HR></A>
<H3>bad input parameter</H3>

 <P>Check your input paramters.  If you define alignment, you cannot
define as horizontal, as vertical boundaries.