Closed dyang37 closed 7 months ago
I added a new script "test_backproject_numerical.py", which shows that $\langle y, Ax \rangle \approx \langle x, A^ty \rangle$ for random image x and sinogram y. Here's the output of the script: where $nrmse(a,b)=\frac{|a-b|}{|b|}$ for scalar values $a$ and $b$. Note that $\langle y, Ax \rangle$ and $\langle x, A^ty \rangle$ are not exactly the same. This is probably because there exists image pixels outside the reconstruction space that yields non-zero backprojection values.
Ok. I approve.
This PR contains the back projector for MBIRCONE.
Interface design
Implementation
I modified the C function "backProjectlike3DCone" to "backProject3DCone", and put a python interface on top of that. Other than that, none of the existing C or python functions are modified.
Testing
test script: "test/test_backproject.py" Test results: Phantom:
Back projection: