warped image blurs out certain channels. here 561 gets stretched along x.
In the K27/K4 dataset 750 is dramatically stretched along x-y in a correlated fashion.
Distortions get worse using higher order polynomial (e.g. 4th order). Distortions are no longer evident using lower order polynomial (2nd 0rder). Bead fit quality is uniformly better with lower order polynomial. 2nd order gives lower cdf90 and lower mean displacement than 3rd order, 3rd order is lower than second order, etc… (stupid matlab, this should be categorically false since all 2nd order polynomials are a subset of all 3rd and 4th order polynomials).
2nd order warp still introduces spreading of dots along a particular axis in some channels.
z-positions go from outrageously off (1000 nm+) to seriously off (200-500 nm) for Pc-GFP bead calibrated data. beads aligned beautifully, not sure why the fits are so bad.
Seems to be systematic problem from not having good sampling in z-data. the x-y polynomial is slightly different at each z-position and is warping out the z-cluster.