improvement to source detection speed using size requirement

[anonymous]

Hi Matt,

just as a follow-up from our discussion last week, I'm making an official ticket for an improvement to duchamp. My idea is that the size requirement can be used during the search/detection phase to speed things up. Instead of checking every single voxel in a plane, you only need to check the voxels spaced Nx elements apart, where Nx is the size requirement in the x dimension. You then move Ny/2 rows ahead (where Ny is the size requirement in the y direction), offset by Nx/2 and again search through while advancing Nx elements at a time. This will result in a 'chess-board' pattern in a given plane, which has been spaced such that any detection of sufficient size will have shown up in at least one of the sample voxels. Any detections missed by this sampling method would have been rejected at the end by the size requirement. When moving to the next plane we can also use this method, by moving Nz - 1 planes ahead instead of just one plane ahead, and then using the alternate pattern. Again, the idea is to use the size requirements to figure out the minimal amount of voxels that need to be sampled to detect everything that satisfies the size requirements. This should result in a substantial speed increase in the detection phase.