[Rtk-users] Lateral blur in a FDK reconstructed volume

[Vincent Libertiaux]

On 25.04.22 09:07, Simon Rit wrote:
> Hi Vincent (and Jasper),
> I agree with Jasper that this is most likely due to a difference in 
> geometric calibration between the two datasets. Do you have more 
> information on what does B use (projection matrices, detailed 
> parametrization, etc.)? I think that such a blur can be caused by a 
> difference in source to detector distance (or, equivalently, detector 
> pixel size).
> Simon
>
> On Tue, Apr 19, 2022 at 11:08 AM Vincent Libertiaux <vl at xris.eu> wrote:
>
>     On 16.04.22 13:52, Jasper Albertus Nijkamp wrote:
>     > Hi Vincent,
>     >
>     >  From just these two images, it is a bit hard to help. However,
>     I have seen similar challenges when the detector vertical offset
>     is not properly set. If you could share a bit more data (fx
>     projection data and the geometry), more people might be able to help.
>     >
>     > Jasper
>     >
>     > -----Original Message-----
>     > From: Rtk-users <rtk-users-bounces at public.kitware.com> On Behalf
>     Of Vincent Libertiaux
>     > Sent: Friday, 15 April 2022 17:15
>     > To: rtk-users <rtk-users at public.kitware.com>
>     > Cc: Damien Koch <dk at xris.eu>
>     > Subject: [Rtk-users] Lateral blur in a FDK reconstructed volume
>     >
>     > Hello rtk users !
>     >
>     > I am facing a problem for which I have exhausted all the
>     possibilities except asking you.
>     > I have performed a standard FDK reconstruction of a lego bricks
>     assembly.  I used a custom-made code to compute the detector
>     horizontal offset and tilt angle, found to be 1.15 mm and 0.02°
>     respectively.  The result of the reconstruction is shown in the
>     picture
>     > https://ibb.co/LdMzJF2 . The volume looks mostly sharp, except
>     on the lateral edges, let's say on the last half brick.
>     >
>     > We had the opportunity to have the same volume reconstructed
>     with two commercial solutions.  The first one, "A", produced the
>     same results than rtk.  The second, "B", produced the result shown
>     in the picture https://ibb.co/VwXMmRH
>     >
>     > In this case, the edges are sharp too.  The offset values found
>     with this software were very close (1.13mm and 0.025°
>     respectively) and feeding them to rtksimulatedgeometry didn't
>     change my result.  No other correction was allegedly applied.
>     >
>     > I thought that the edge blurring was due to a wobbling artefact
>     but it can't be the case according to the result with the "B"
>     software.
>     >
>     > Do you have any idea on what could cause this blurring on the
>     edges ?
>     >
>     > I thank you very much for any clue.
>     >
>     > Best regards,
>     >
>     > Vincent
>     >
>     > _______________________________________________
>     > Rtk-users mailing list
>     > Rtk-users at public.kitware.com
>     > https://public.kitware.com/cgi-bin/mailman/listinfo/rtk-users
>
>     Hi Jasper,
>
>     thank you for your reply. I had already tried to play with the
>     vertical
>     offset but setting it to a value different than 0 progressively
>     decrease
>     the overall quality of the reconstruction.
>     Following your advice, here are the projections:
>
>     http://share.xris.eu/d91b09673bba
>
>     word of warning, the set is very large (900 projections on a
>     3072x3072
>     pixels detector, approx. size = 16Go)  I could try and make it
>     smaller
>     by downsampling it but I am afraid it would mask the problem. I
>     can try
>     and do it on request.
>     The geometric parameters I used were: SDD = 810 mm, SID = 410 mm,
>     proj_iso_x = 1.15mm and in_angle = 0.02°.
>
>
>     Best regards,
>
>     Vincent
>
>     _______________________________________________
>     Rtk-users mailing list
>     Rtk-users at public.kitware.com
>     https://public.kitware.com/cgi-bin/mailman/listinfo/rtk-users
>
Hi Simon, hi Jasper.

I have investigated a bit more on that issue.  I told you in my first 
message that both software we used besides rtk found the same set of 
parameters.  After playing a bit more with the software "B" (the one 
that gives the sharp edges), I found out that the tilt angle is found to 
be 0 and it is the out-of-plane angle around the vertical axis of the 
detector which is non zero (and very coincidentally is very close to the 
tilt angle we computed).  The value is quite low (0.025°).  I am a bit 
puzzled as, if I am not mistaken, the literature always assumes that 
both out of plane angles are negligible as long as they don't come close 
2°.  We are two order of magnitude below this value, hence my surprise.

I am also wondering if this angle can be introduced in the rtk 
geometry.  From the doc (http://www.openrtk.org/Doxygen/DocGeo3D.html), 
it seems that only the out-of-plane angle around the detector horizontal 
axis is considered.  Is that correct ?  If yes, is there a way to model 
the other one with the available parameters ?

Once again, thank you for any help you can provide.

Kindest regards,

Vincent




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/rtk-users/attachments/20220510/08cef593/attachment.htm>