1D and 2D Least-squares Gaussian fit

[zeddie888]

Test how good these are before trying Gaussian Grid

[zeddie888]

According to the Delabie paper:

• The 2D least-squares Gaussian fit (with 9x9 window) has the best accuracy
• Although Gaussian Grid is better than COG, it isn't better than 2D fit in terms of accuracy. In fact, it's accuracy is comparable to that of 1D fit

The hybrid Gaussian Grid (for determining better initial parameters) is probably the only thing worth implementing to speed up the least-squares fit methods

[markasoftware]

To test in ideal conditions:

• Turn down the brightness so there aren't any "saturated" (full brightness) pixels.
• Disable all generation noise (--generate-read-noise 0 and --generate-shot-noise fase)

Then the pixel brightnesses will actually be generated by a Gaussian function (

lost/src/io.cpp

Line 435 in ddccdfc

return generatedStar.peakBrightness * exp(-d0.MagnitudeSq() / (2*stddev*stddev));

), so there's no excuse for worse performance in that case. (It is possible that there's a bug in the generator meaning that the brightness distribution isn't actually Gaussian, but so far I've seen nothing to indicate that)