<html>
<head>
<style><!--
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Verdana
}
--></style>
</head>
<body class='hmmessage'>
<BR>
Hi Luis,<BR>
<BR>
Thank you for answering me.<BR>
<BR>
1) As you said, my images are from Microscopy and the spacing is about 2.85 micrometres/pixel. The images are in a raw format and I used the following .mhd file for reading it:<BR>
<BR>
<BR><SPAN lang=ES-MODERN>
NDims = 2<BR>
DimSize = 2000 2000<BR>
ElementType = MET_UCHAR<BR>
ElementSpacing = 1.0 1.0<BR>
Offset = 0.0 0.0<BR>
ElementByteOrderMSB = False<BR>
ElementDataFile = fixed.raw<BR>
<BR>
I also tried to as ElementSpacing 2.83 and 0.00283 (just in case...) but the results weren´t good either.<BR>
<BR>
2) I also try to put a relaxation factor of 0.9 as you said and I couldn´t see any improvement.<BR>
<BR>
3) I´m attaching three .txt files with some results obtained. <BR>
The results1.txt file shows the output of the program as you can find it on ITK/Examples/Registration/. <BR>
The results2.txt file shows the output of the program with a predifined center in the position 1000x1000. The results3.txt file shows the output of the program with a predifined center in the position 1000x1000, a predefined angle of 0.9 radians and a predefined traslation of -340 in the X axis and 800 in the Y axis (that are the values for an optimal registration).<BR>
<BR>
<BR>
I hope you can help me to fix my problem with that information... an be free to ask anything else :P.<BR>
<BR>
Thank you really much.<BR>
<BR>
Manuel.<BR>
</SPAN><BR> <BR>> Date: Wed, 21 Oct 2009 17:29:08 -0400<BR>> Subject: Re: [Insight-users] Problem with 2D Registration (local minima)<BR>> From: luis.ibanez@kitware.com<BR>> To: manueldavid.tabas@hotmail.es<BR>> CC: insight-users@itk.org<BR>> <BR>> Hi Manuel,<BR>> <BR>> Thanks for the detailed description of your problem.<BR>> <BR>> A couple of questions:<BR>> <BR>> <BR>> 1) What is the pixel spacing of your image ?<BR>> <BR>> The typical proportion of :<BR>> <BR>> > const double translationScale = 1.0 / 1000.0;<BR>> <BR>> is set up for images whose physical extent in<BR>> the units used by the spacing is in the range<BR>> of hundreds.<BR>> <BR>> For example, a typical CT scan will have 512x512<BR>> pixels with a spacing close to 1mm / pixel.<BR>> <BR>> Your images are probably from Microscopy or<BR>> from a Satellite, so I'm wondering about the units<BR>> that you use (microns or kilometers) and the<BR>> exact numeric values of the spacing.<BR>> <BR>> <BR>> 2) Could you post to the list the output of the<BR>> command observer that tracks the values of<BR>> the metric and the transform at every iteration ?<BR>> <BR>> <BR>> Something that happens often when using the<BR>> RegularStepGradientDescentOptimizer is that,<BR>> if the registration process have a couple of<BR>> bounces of direction in the parametric space,<BR>> the step length may be reduced too fast, and<BR>> not give the optimizer the oportunity to walk<BR>> far enough to get to the solution.<BR>> <BR>> <BR>> You may want to try using the method:<BR>> <BR>> optimizer->SetRelaxationFactor( 0.9 );<BR>> <BR>> <BR>> This method defines the factor by which the<BR>> step length is reduced at every time that the<BR>> parametric space step changes direction.<BR>> This may prevent the premature shrinkage<BR>> of the step length (assuming that this is the<BR>> real problem).<BR>> <BR>> <BR>> Anothe typical issue: It may simply be that you<BR>> need a better initialization for your images.<BR>> <BR>> <BR>> Note that the Transform initializer in the example:<BR>> <BR>> ImageRegistration6.cxx<BR>> <BR>> is using the methods MomentsOn().<BR>> <BR>> This may or may not be the right method to use,<BR>> depending on the content of your images.<BR>> <BR>> For example: If the two images have different<BR>> modalities, or if the main features of one image<BR>> may not overlap to the one in the other image.<BR>> <BR>> If you could post a screenshot of both images<BR>> we could avoid all this speculation... :-)<BR>> <BR>> Or...<BR>> If you can share your images in a public web<BR>> site, that will be even better.<BR>> <BR>> <BR>> <BR>> Thanks<BR>> <BR>> <BR>> Luis<BR>> <BR>> <BR>> ----------------------------------------------------------------<BR>> On Wed, Oct 21, 2009 at 11:56 AM, Manuel David Tabas Calle<BR>> <manueldavid.tabas@hotmail.es> wrote:<BR>> > Hello ITK,<BR>> ><BR>> > I´m working with ITK to do registration of 2D images and I think I reach a<BR>> > local minima that makes a poor registration.<BR>> ><BR>> > I´m using 2D images 2000x2000 pixels in what the moving image needs a<BR>> > rotation and a translation. So I reproduced the ImageRegistration6.cxx to do<BR>> > that. But the result I got is not good enough so I think that the<BR>> > MeanSquares metric reaches a local minima. The value of the metric is about<BR>> > 4000 (aprox.), and I run a test using as input parameters the optimal ones<BR>> > (that is, the result I think i should obtain) and the metric returns a value<BR>> > of 2500 (aprox.).<BR>> ><BR>> > The framework I use is:<BR>> ><BR>> > itkImageRegistrationMethod<BR>> > itkMeanSquaresImageToImageMetric<BR>> > itkLinearInterpolateImageFunction<BR>> > itkRegularStepGradientDescentOptimizer<BR>> > itkCenteredRigid2DTransform<BR>> ><BR>> > And the parameters of my registration are:<BR>> ><BR>> > const double translationScale = 1.0 / 1000.0;<BR>> > optimizerScales[0] = 1.0;<BR>> > optimizerScales[1] = translationScale;<BR>> > optimizerScales[2] = translationScale;<BR>> > optimizerScales[3] = translationScale;<BR>> > optimizerScales[4] = translationScale;<BR>> ><BR>> > optimizer->SetMaximumStepLength( 0.1 );<BR>> > optimizer->SetMinimumStepLength( 0.001 );<BR>> > optimizer->SetNumberOfIterations( 200 );<BR>> ><BR>> ><BR>> > The scales are well defined and I tried to vary the parameters of the<BR>> > optimizer but the results were similar, so I don´t know what to do next. Any<BR>> > idea??<BR>> ><BR>> > Thank you in advance, I hope you can have more ideas to try.<BR>> ><BR>> ><BR>> ><BR>> > Manuel.<BR>> ><BR>> ><BR>> ><BR>> ><BR>> > _____________________________________<BR>> > Powered by www.kitware.com<BR>> ><BR>> > Visit other Kitware open-source projects at<BR>> > http://www.kitware.com/opensource/opensource.html<BR>> ><BR>> > Please keep messages on-topic and check the ITK FAQ at:<BR>> > http://www.itk.org/Wiki/ITK_FAQ<BR>> ><BR>> > Follow this link to subscribe/unsubscribe:<BR>> > http://www.itk.org/mailman/listinfo/insight-users<BR>> ><BR>> ><BR> <br /><hr />Vive toda la experiencia de Messenger con Orange desde tu Blackberry <a href='http://serviciosmoviles.es.msn.com/messenger/blackberry.aspx ' target='_new'>¡Descúbrelo!</a></body>
</html>