set offOrigin and offDetector

[zeminLi66]

20250328_142024.mp4

Hello.
My projection method is shown in Figure 1, where the object and detector are rotated around the x-axis at the same angular velocity, while the source remains stationary and the object rotates 90 ° along the y-axis. The code is shown in Figure 2. The obtained projection image is shown in video recording 3.

[zeminLi66]

reconstruct.mp4

SIRT is iterated twice, and the reconstructed image obtained is shown in the figure. I am curious why white noise is initially all on the left and then all on the right? Is this normal? For my projection method, which iterative reconstruction method would work better? I am now planning to write my own refactoring method based on TIGRE for my projection method. What aspects do you think I can modify and improve from? Thank you very much for your valuable feedback

[AnderBiguri]

SIRT is quite a slow algorithm, try OS-SART instead!

But indeed, in laminography, your depth resolution is really bad, and out of the middle plane you will get artifacts, as you have a special version of "limited angle/view CT"

[zeminLi66]

Lol, you’re absolutely right, but our machine operates with this specific imaging method, and there’s no way to change it. Has TIGRE been specifically optimized for CT 3D reconstruction with limited angles and less-than-ideal depth information? For example, is there a particular iterative algorithm in TIGRE that performs well in such cases? If not, based on your experience, could you kindly suggest some directions I could explore for improvement?

[AnderBiguri]

@zeminLi66 Yeah, thats OK that the machine operates that way, its what laminography is. But on that operation mode you don't get a lot of depth, so when testing with TIGRE you should just try to see if the central slice is well reconstructed. In your real case scenario you will almost likely be reconstructing something very thin, so bad depth reconstruction will be marginally an issue.

If you are not 100% conviced the geometry is right, use OS-SART (or SIRT with hundreds of iterations). Once you know 100% the geometry is right, algorithms like OS-ASD-POCS, OS-SART-TV will produce a better recon for this cases, however they are very hard to tune and the wrong parameters will cause very bad artifacts.

[zeminLi66]

@AnderBiguri Great， I'll give it a try right away

[zeminLi66]

abnormal.mp4

And I discovered a strange phenomenon, if I set the rotation angles of the object to all 0, that is, angles=np. hstack (zeros_array, zeros_array, zeros_array). The obtained projection image will be abnormal, as shown in the video recording.

[zeminLi66]

testProjection.mp4

But if I move 600 * np. sin (angle 1) from the z position of offOrigin to the X position, the projected image can be restored to normal., as shown in the video recording. Is this normal and why is it like this？

[AnderBiguri]

Thinking of arbitrary geometries and what they do is not very useful, there are infinite values for angles and geometry that would make infinite different behaviours that are not useful at all.

[zeminLi66]

Hello，Biguri, TIGRE has delivered excellent reconstruction results for my CT dataset. However, I’ve observed some residual noise in the reconstructed images that I’m currently unable to pinpoint the origin of. Would you be able to share your expertise on potential causes of such artifacts based on your experience with TIGRE? Any guidance would be greatly.

20250415_174409.mp4

The projected image is as follows

20250415_175310.mp4

[zeminLi66]

[zeminLi66]

By the way, I made some changes to the Python code. Previously, the blockSize updated by ossart was 20 each time. I changed it to 1 and only fed one image for each projection and backprojection. Will this have a significant impact.

[AnderBiguri]

@zeminLi66 You changed OS-SART to become SART! You can just SART instead ;) But in any case, its an optional input parameter, there is no change in the code needed to make it different.

About the results: no idea! clearly laminography has very little out of plane resolution, but its hard to know if what you are getting is errors in the geometry or laminography artefacts. You know more than I do here.

[zeminLi66]

I feel that the probability of this being an artifact is higher because I cannot be certain that the calibrated size is absolutely accurate, and I have tried debugging different sizes, but the artifact still exists. And my shooting method may not have enough depth information. Assuming this is an artifact, is there any way to remove it at this stage

[AnderBiguri]

@zeminLi66 you keep asking "is there a way to remove the artifacts if I done something wrong" and the answer is yes, by not doing it wrong. But you need to understand that I don't have your data, your CT machine or your code, so I have absolutely no way of fixing it for you! I can help if you have specific, technical, well posed, questions. But I cant help anyone that says "this is wrong, help". For what I know you may be multiplying everything by 100, I just can't really know!

Sorry for the tone, I just don't really know how to help and you keep asking the same questions. I don't do magic, there is nothing that I know that you can't also know, so I don't know how to help you. Please consider refining the questions before asking.

[zeminLi66]

What algorithms in TIGRE can achieve TV noise reduction similar to the one shown in the picture above？I tried ossart_tv, but the results were almost identical to ossart.

[AnderBiguri]

Yes of course, you just need to find the right parameters for the algorithm

[zeminLi66]

Hello Biguri,
I hope this message finds you well. First and foremost, I wanted to extend my sincere gratitude for developing the TIGRE toolbox—it has been an invaluable resource for my research. I also want to thank you again for the helpful advice you provided previously.
Building on that, I've recently encountered a new challenge. While I have been achieving promising results with TIGRE, I am facing persistent difficulties with certain cases involving significant depth information. As shown in the attached image, adhesion artifacts remain quite problematic, and I feel the layer separation is still somewhat insufficient.
My current reconstruction strategy relies on TIGRE's siddonProjection. To enhance the reconstruction quality, I am now considering switching to the Separable Footprint (SF) forward projection algorithm.
Could you please advise if adopting the SF algorithm is likely to yield significant improvements for these specific artifacts? Alternatively, I would be very grateful for any other suggestions or insights you might have.
Thank you once again for your incredible work and your time.

35-2-improve.75.2025-08-11.13-36-08.mp4

[AnderBiguri]

@zeminLi66 Thanks for the nice words! what you are seing are the artifacts you get with laminography, they are quite standard on this modality. I seriously doubt that changing the projectors would make any difference. Perhaps using a TV algorithm will improve them, particularly an anisotropic one (not available in TIGRE), but in general those artifacts are the cost of doing laminography.

[zeminLi66]

Thank you for your suggestion. I tried using TV noise reduction in TIGRE, but it didn't weaken the adhesion artifact effect. Instead, some defects were blurred. I'm not sure if it's because of my parameter settings error. Would anisotropic be much better than it is now？

[AnderBiguri]

Yes, this is also expected sadly. Anisotropic TV may do better, but not significantly better. I am afraid you are playing in the boundaries of what people know about CT/laminography, so there is no clear asnwer on what to do to improve image quality.

[zeminLi66]

That's helpful, thank you. Speaking of reconstruction quality, I've been looking into the Separable Footprints (SF) method. Based on what I've read, it seems to provide a substantial improvement in accuracy, and I'm planning to experiment with it.
I was curious to get your perspective—are there other key methods, besides SF, that you would recommend for significantly boosting image quality?"

[AnderBiguri]

Sure, give it a shot. I don't think it will meaningfully change anything though, as SF is not a reconstruction algorithm, just a different way to do forward/backprojection. But if you implement it I would be interested to know how big/small is the difference.