Is convolution in blob search done correctly?
Is the convolution, called from PeakFinder::findPrimaryBlobs, done correctly?
The convolution is done via Fourier forward transform, multiplication with transformed kernel, Fourier backtransform. This implies the assumption of periodic boundary conditions [1]. These make no sense in our application. However, the FT based method can be saved because the kernel is nonzero only for a small disk in pixel space. Let the disk have radius r. If only there is a fringe of width r around our rectangular detector where counts are zero, then there will be no artifacts from doing the convolution with periodic boundary conditions. Question thus: Is such a fringe added to the raw data? If not, I'd claim that the convolution is done incorrectly [1, fig 2c vs 2d].
As the entire peak finding is only an heuristic intermediate step in our entire data reduction, distortions at the boundary will normally not lead to incorrect final results. Nonetheless, an improved convolution algorithm might have favorable impact upon subsequent steps.
At this occasion, also consider whether other algorithms aren't better suited [1].