help_imare.html

图像对齐（inage registration）工具箱

&lt;CP +> </font></b><font color="#000099">or </font><b><font color="#993300">&lt;CP
+ INV>.</font></b>
<br>&nbsp;
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="invc"></a><b><font color="#72BF64"><font size=+1>Invariant
computation</font></font></b>
<br><font color="#000099">Having CPs (either detected now&nbsp; </font><b><font color="#663300">&lt;CP
detection> </font></b><font color="#000099">or uploaded from the previous
session </font><b><font color="#993300">&lt;CP sets></font></b><font color="#000099">)
we can start computation of invariants. There are implemented three types
of invariants.</font>
<br>&nbsp;
<ul>
<li>
<font color="#000099"><b>Hu moment invariants</b> - allows user to compute
Hu's moment invariants, invariant to similarity transform</font></li>

<br>&nbsp;
<p><font color="#000099">Sequence of steps:</font></ul>

<ul><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Invariant computation
parameters>&nbsp;&nbsp; </font></b><font color="#000099">...&nbsp;&nbsp;

<b>Hu</b></font>
<p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the desired
radius of circular neighborhood of CP, which will be used for inv. computation</font>
<br><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Refinement radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the expected
error of CP localization (meaning : see <a href="#cperr">here</a>)</font>
<p><font color="#000099">&nbsp;&nbsp;&nbsp; </font><b><font color="#663300">&lt;INV
comp>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; activates automatic
process of invariant computation</font><p>&nbsp;</ul>

<ul>
<li>
<font color="#000099"><b>Rotational moment invariants</b> - allows user to compute
all independent rotational moment invariants up to given order. They can be used 
for translated and shifted images</font>&nbsp;
<p><font color="#000099">Sequence of steps:</font><ul>
  <p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Invariant computation
parameters>&nbsp;&nbsp; </font></b><font color="#000099">...&nbsp;&nbsp;

<b>Rotational</b></font> </p>
<p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the desired
radius of circular neighborhood of CP, which will be used for inv. computation</font>
<p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Order>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the desired maximum 
order (p+q) of invariants</font><p>
<br><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Refinement radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the expected
error of CP localization (meaning : see <a href="#cperr">here</a>)</font>
<p><font color="#000099">&nbsp;&nbsp;&nbsp; </font><b><font color="#663300">&lt;INV
comp>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; activates automatic
process of invariant computation</font></ul>

</ul>

&nbsp;<ul>
<li>
<font color="#000099"><b>Affine moment invariants</b> - allows user to
compute affine moment invariants, invariant to affine transform. They are
described in</font></li>

<br>&nbsp;
<p>&nbsp;
<p><i><font color="#000099">Flusser J., Suk T.: <a href="http://www.utia.cas.cz/cgi-bin/toASCII/library/prace/930019.pdf" TARGET="_blank">Pattern Recognition by
Affine Moment Invariants</a>. Pattern Recognition , 26 (1993), 1, 167-174</font></i>
<p><font color="#000099">Sequence of steps:</font></ul>

<ul><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Invariant computation
parameters>&nbsp;&nbsp; </font></b><font color="#000099">...&nbsp;&nbsp;

<b>Affine</b></font>
<p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the desired
radius of circular neighborhood of CP, which will be used for inv. computation</font>
<br><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Refinement radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the expected
error of CP localization (meaning : see <a href="#cperr">here</a>)</font>
<p><font color="#000099">&nbsp;&nbsp;&nbsp; </font><b><font color="#663300">&lt;INV
comp>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; activates automatic
process of invariant computation</font></ul>

<ul>
<li>
<font color="#000099"><b>Blur combined&nbsp; moment invariants</b> - allows
user to compute moment invariants, invariant to similarity transform
and blurring. Large neighborhood area is necessary due to the border error.&nbsp; They are described in</font></li>

<br>&nbsp;
<p>&nbsp;<i><font color="#000099">Flusser J., Zitova B.: <a href="http://www.utia.cas.cz/cgi-bin/toASCII/library/prace/990162.pdf" TARGET="_blank">Combined invariants
to linear filtering and rotation</a>. International Journal of Pattern Recognition
and Artificial Intelligence, 13 (1999), 8, 1123-1136.</font></i>
<p><font color="#000099">Sequence of steps:</font></ul>

<ul><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Invariant computation
parameters>&nbsp;&nbsp; </font></b><font color="#000099">...&nbsp;&nbsp;

<b>Blur combined</b></font>
<p><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the desired
radius of circular neighborhood of CP, which will be used for inv. computation</font>
<br><b><font color="#663300">&nbsp;&nbsp;&nbsp; &lt;Refinement radius>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; enter the expected
error of CP localization (meaning : see <a href="#cperr">here</a>)</font>
<p><font color="#000099">&nbsp;&nbsp;&nbsp; </font><b><font color="#663300">&lt;INV
comp>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;

</font></b><font color="#000099">...&nbsp;&nbsp;&nbsp; activates automatic
process of invariant computation</font></ul>
<font color="#000099">Having computed invariants, they can be saved </font><b><font color="#993300">&lt;Saving
CP, INV></font></b><font color="#000099">, their values vizualized 
</font><b><font color="#993300">&lt;INV
values> </font></b><font color="#000099">or the subset for the registration
can be chosen&nbsp;</font><b><font color="#993300"> &lt;Choice of invariants></font></b><font color="#000099"> 
(not in the case of <b>rotational invariants</b>).
Here, the window with checkboxes for all orders of computed invariants
is activated and desired subset can be selected for future application.
In the Reference and Sensed image areas the CPs which were used for invariant
computation are marked with green circles (points which are too close to
borders are excluded from the invariant computation). This
subset of points can be visualized by </font><b><font color="#993300">&lt;CP
+ INV>.</font></b>
<br>&nbsp;
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="match"></a><b><font color="#72BF64"><font size=+1>Matching</font></font></b>
<p><font color="#000099">Having CPs and computed invariants (either detected
</font><b><font color="#663300">&lt;CP
detection>&nbsp; </font></b><font color="#000099">and computed </font><b><font color="#663300">&lt;INV
comp></font></b><font color="#000099">&nbsp; or uploaded from the previous
session </font><b><font color="#993300">&lt;CP sets>,&nbsp; &lt;Invariants></font></b><font color="#000099">),
the two most probably corresponding pairs can be found by&nbsp;</font><b><font color="#993300">
&lt;Two best></font></b><font color="#000099">. The distance of their invariant
representation and maximum likelihood coefficients are used here (described
in <i>Flusser J.: Object matching by means of matching likelihood coefficients.
Pattern Recognition Letters, 16 (1995), -, 893-900.)</i>. First pair (one
points from Reference and one from Sensed images) is marked with red circles,&nbsp;
the second with blue circles. At the same time the window with paired CPs
is activated. There is no limit on maximum allowed distance between paired
points so possibly not all pairs are correct.</font>
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="reg"></a><b><font color="#72BF64"><font size=+1>Registration</font></font></b>
<br><font color="#000099">The registration algorithm implemented in IMARE
toolbox can be found in <i>Zitova B.: Registration of Radiometrically Deformed
Images by Invariant Descriptors. Ph.D. Thesis. Faculty of Mathematics and
Physics, Charles University, Prague 2000, 92 pp. </i>Before the registration
itself starts, three parameters should be set. Firstly, it is&nbsp; </font><b><font color="#663300">&lt;Matching
threshold></font></b><font color="#000099">, which defines the threshold
used for labeling point pairs as corresponding or non-corresponding (the
spatial distance of the CPs from the Reference image and CPs form transformed
Sensed image is checked, if the distance between points is less then threshold,
the points are set as corresponding ones). Then, </font><b><font color="#663300">&lt;Radius></font></b><font color="#000099">
and </font><b><font color="#663300">&lt;Refinement></font></b><font color="#000099">,
which define the radius of circular neighborhood, used for invariant computation
in the refinement step, and expected error in localization (meaning : see
<a href="#cperr">here</a>).
After setting the parameters, </font><b><font color="#663300">&lt;Registration>
</font></b><font color="#000099">invokes
the registration process. Firstly the window with coordinates of CPs from
Reference image and transformed Sensed image is shown. Transformation parameters
of the Sensed image is estimated from the two best corresponding point
pairs. Then the mutual distance between CPs from Reference image and transformed
Sensed image is used to establish more corresponding pairs. When the registration
process ends, the computed transformation parameters are listed at the
message line and transformed Sensed image is shown. Transformed image (in
TIFF file) and parameters (in MAT file) can be saved </font><b><font color="#993300">&lt;Saving
transform></font></b><font color="#000099">. The achieved registration
can be evaluated by </font><b><font color="#993300">&lt;Evaluation>.</font></b>
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="eval"></a><b><font color="#72BF64"><font size=+1>Evaluation</font></font></b>
<br><font color="#000099">Evaluation of the achieved registration is implemented
as follows. </font><b><font color="#993300">&lt;Evaluation></font></b><font color="#000099">
opens the window, which enables to see computed parameters&nbsp; </font><b><font color="#993300">&lt;Parameters></font></b><font color="#000099">.

The visualization of the difference
image (abs(Reference-transformed Sensed)) </font><b><font color="#993300">&lt;Difference
image></font></b><font color="#000099">, average image ((Reference + transformed
Sensed)/2 ) </font><b><font color="#993300">&lt;Average image></font></b><font color="#000099">,
chess-board image (alternating parts from the Reference and transformed
Sensed images) </font><b><font color="#993300">&lt;Chess-board image></font></b><font color="#000099">,
and transformed Sensed image </font><b><font color="#993300">&lt;Transformed
image></font></b><font color="#000099"> is implememted. Default state is,
that images are shown in the same window, if the check box </font><b><font color="#993300">&lt;New
window></font></b><font color="#000099">is activated, every image has its
separate window.</font>
<br>&nbsp;
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="ncc"></a><b><font color="#72BF64"><font size=+1>Normalized
cross-correlation</font></font></b>
<br><font color="#000099">Alternatively, for shifted images, the <a href="#notes">normalized
cross-correlation</a> (NCC) can be applied. Here, firstly the window from
the Sensed image has to be chosen. You can either specify the desired size
of squared window in input box next to </font><b><font color="#993300">&lt;Window
selection> </font></b><font color="#000099">or you can let it empty and
then the window size is specified interactively (even rectangular shape
in this case). </font><b><font color="#993300">&lt;Window selection> </font></b><font color="#000099">evokes
in the Sensed image area selection of the size/position of the window.
If the window size is specified, you left mouse click at the desired top
left corner and the window boundary of this size appears. The right mouse
click confirms the selection otherwise continue with the left mouse clicking
till the desired localization is achieved. If no size is specified, interactively
the position of the top left corner is chosen with the left mouse click
and the shape is set with the mouse moving while at the same time holding
down the left mouse button. At the end (after confirmation/after stopping
the left mouse click and mouse moving) the selected area is shown in the
separate window, which is closed after hitting Space Bar. Now the NCC can
be start. </font><b><font color="#993300">&lt;Normalized CC> </font></b><font color="#000099">begins
the process of computation. After finishing, the computed NCC matrix is
shown in separate window as 3D mesh, position of global maxima is found
and shift parameters are computed. The transformed Sensed image is shown
as well. When the registration process ends, the computed transformation
parameters are listed at the message line and transformed Sensed image
is shown. Transformed image (in TIFF file) and parameters (in MAT file)
can be saved </font><b><font color="#993300">&lt;Saving transform></font></b><font color="#000099">.&nbsp;
The achieved registration can be evaluated by </font><b><font color="#993300">&lt;Evaluation>.</font></b>
<p>
<hr SIZE=3 WIDTH="100%">
<br><a NAME="auto"></a><b><font size="+1" color="#72BF64">Phase correlation</font></b>
<p>F<font color="#000099">or shifted images, even form different modalities, 
which preserves borders, the <a href="#notes">phase correlation</a>&nbsp; can be applied. 
Firstly the window from
the Sensed image has to be chosen in the same manner as in the previous case by </font><b><font color="#993300">&lt;Window
selection>.</font></b><font color="#000099"> </font><b><font color="#993300">&lt;Phase 
correl&gt; </font></b><font color="#000099">begins
the process of computation. After finishing, the computed PHC matrix is
shown in separate window as 3D mesh, position of global maxima is found
and shift parameters are computed. The transformed Sensed image is shown
as well. When the registration process ends, the computed transformation
parameters are listed at the message line and transformed Sensed image
is shown. Transformed image (in TIFF file) and parameters (in MAT file)
can be saved </font><b><font color="#993300">&lt;Saving transform></font></b><font color="#000099">.&nbsp;
The achieved registration can be evaluated by </font><b><font color="#993300">&lt;Evaluation>.</font></b>
</p>
<p>&nbsp;<p>
<hr SIZE=3 WIDTH="100%">
<p><a NAME="info"></a><b><font color="#72BF64"><font size=+1>Info</font></font></b>
<p><font color="#000099">In case of any questions / problems / suggestions,
please contact <a href="mailto:[e-mail]">Barbara Zitova</a>.</font>
</body>
</html>