[Insight-users] # of spatial samples

[pinpress]

Hi Luis,

Thanks for your expertise.

I wonder if there is a similar default value of # of spatial samples used in
MI evaluation in other types of mutual information metric, e.g., normalized
mutual information. Of course, I could do an experiment to find it out.

In addition, I wonder what happens when the requested # of spatial samples
exceeds the total # of image voxels? For example, my moving image has 1000
voxels, but I requested 2000 spatial samples to use. Then the actual # of
spatial samples would be 1000 (i.e., the maximum # of voxels) or 2000? 

Thanks a lot.

Pinpress


Luis Ibanez wrote:
> 
> 
> Hi Pinpress,
> 
> 
> Your observation is correct.
> 
> 
> 1) There is a default value for the number of samples
>     to be used by the MattesMutualInformation. This is
>     set in the constructor.
> 
> 
> 2) The initial value of 500 was based on regression tests
>     performed on 2D images of dimensions 180 x 217.
> 
> 
> 3) It is *your responsibility* to replace the default value
>     with a number of samples that matches your image size.
> 
>     A typical rule of thumb is to use between 1% and 5% of
>     the total number of pixels in the image. You will find
>     examples on how to compute this number in the ITK
>     Software Guide, and the associated .cxx files in
> 
>               Insight/Examples/Registration
> 
>     Note however, that depending on the characteristics
>     of your registration problem you may need to use more
>     samples.
> 
> 
> 4) The samples are drawn "randomly" from the FixedImageRegion
>     that you pass to the ImageRegistrationMethod.
> 
> 
> 
>   Please let us know if you have any further questions,
> 
> 
>       Thanks
> 
> 
>           Luis
> 
> 
> -----------
> pinpress wrote:
>> Hi,
>> 
>> I just found out that when I was using
>> MattesMutualInformationImageToImageMetric in registration, there is a
>> default number of spatial samples used in the Mutual Information
>> evaluation,
>> which is 500. If I do not explicitly specify by calling:
>> SetNumberOfSpatialSamples, then the registration will not generally yield
>> good result.  Is this observation correct? I have only tested on one
>> example
>> program so far.
>> 
>> The other question is: Exactly how these spatial samples were determined?
>> Are they all from the moving image? Are they randomly selected voxels
>> from
>> the moving image? The reason I am asking is because I accidentally found
>> out
>> that the number of spatial samples could be even larger than the total
>> number of voxels. How can this be possible? 
>> 
>> Thanks.
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
> 

-- 
View this message in context: http://www.nabble.com/--of-spatial-samples-tp20220158p20227272.html
Sent from the ITK - Users mailing list archive at Nabble.com.