[Insight-users] using Mutual Information

[Luis Ibanez]

Hi Jose,

Thanks for the additional details regarding your images.

What you probably want to do is to create a new Metric
that will take into account both the intensity values
and the Z values from your image.

This is like interpreting your image as a 2 components
dataset, where the first component is the intensity
and the second component is the Z depth.

NOTE that before you go further, you must check whether
the scanner was moved between the first reading and
the second. Because in that case, the Z values of one
image will not match the Z values from the other one
given that they are measured with respect to different
origins of coordinates.

You may want to take a look at the code implementing
Metrics such as MeanSquares in


          Insight/Code/Algorithms


That will give you a base for writing your own Metric.
You will also need a specialized interpolator, capable
of computing both intensity and Z values.


Once you have your new Metric and Interpolator you can
take advantage of the rest of ITK components such as
optimizers and transforms in order to complete a
registration method.


    Regards,


        Luis



-------------------------------
José Santamaría López wrote:

> Hi Luis,
> 
>   answering some questions...
> 
> Luis Ibanez
> 
>>Hi Jose,
>>
>>
>>Happy New Year !
>>
>>
>>The registration framework in ITK computes everything
>>in physical espace, not in pixel space.
>>
>>
>>
>>1) Are your images 2D ?
>>
> 
> 
> Yes, they are. But, I must to save those
> images in .png, .jpeg, or any 2D file format
> supported by ITK ... That is right?
> 
> 
>>2) Are the (x,y) coordinates regular ?
>>    that is, can they be simply computed as
>>
>>         (Ox + i * Sx ,  Oy + j * Sy )
>>
>>    where Ox,Oy is the origin of the image,
>>    Sx,Sy are the interpixel spacing, and i,j
>>    are the indices of a particular pixel.
>>
> 
> 
> I'm not sure but...I think it is as you say.
> 
> 
>>
>>3) Why do you encode in the Z coordinate the
>>    intensity of the image before normalization ?
>>    This sounds a bit like an elevation grid....
>>
> 
> 
> The images come from a range scanner. You are
> right thinking on Z coordinate encodes an elevation
> grid such as the depth from the scanner device to the
> surface of the scanned object.
> 
> 
>>
>>4) Yes, you can use ITK for registering images
>>    that have a reduced overlapping region.
>>
>>
>>
>>We *strongly encourage* you to read the chapter
>>on image resampling and the chapter on image
>>registration from the ITK Software Guide
>>
>>
>>    http://www.itk.org/ItkSoftwareGuide.pdf
>>
>>
>>- Section 6.7 "Geometric Transformations", pdf-page 199.
>>- Chapter 8   "Image Registration", pdf-page  241-340
>>
>>
>>
>>   Please let us know if you have further questions,
>>
>>
>>
>>     Thanks
>>
>>
>>
>>       Luis
>>
> 
> 
> Thanks Luis.
> 
> Jose.
> 
> 
>>
>>----------------------------
>>José Santamaría López wrote:
>>
>>>Hi Luis,
>>>
>>>  after of all...happy new year :-)
>>>I'm thinking on applying the mutual information
>>>approach as metric for imagen registration in
>>>a new problem.
>>>
>>>My doubt is if the registration framework needs
>>>to work with physical space instead on pixel
>>>space, that is, I have two gray scale images
>>>at given resolution where its intensity values
>>>indicates some feature. Each pixel value also
>>>has a (x,y,z) world coordenate value, where
>>>the z component references to the former normalized
>>>pixel intensity.
>>>
>>>The two images only share a partial common area with
>>>similar structure of gray scale pixels at their whole
>>>images. My intention should be to obtain a good
>>>pre-matching between those similar areas from
>>>both images by means a rigid body registration.
>>>
>>>Could I make use of ITK Software to develop a
>>>solution for such problem?
>>>
>>>
>>>Thanks.
>>>
>>>Jose
>>>
>>>
>>>
>>
>>
>>
>>
> 
> 
>