johnpane
Well-known member
I am interested whether the statistics reported by SFS can be used to quantitatively evaluate integration results. This could help with evaluating rejection parameters, or how drizzle integration compares to non-drizzle.
One thing I notice is that the output from ImageIntegration and DrizzleIntegration are not on the same scale (drizzle is dimmer, on average). For an accurate comparison, is normalization necessary?
Here are some statistics from one dataset, with the output from drizzle and with that output normalized to the integration result:
Quantitatively, this would seem to demonstrate that the drizzle result is superior on most important metrics. Visually, none of these differences are apparent when comparing the integration and normalized drizzle integration with the same STF applied. The only visible difference I can see is that the drizzle result seems less sharp than the regular integration.
One thing I notice is that the output from ImageIntegration and DrizzleIntegration are not on the same scale (drizzle is dimmer, on average). For an accurate comparison, is normalization necessary?
Here are some statistics from one dataset, with the output from drizzle and with that output normalized to the integration result:
| File | PSF Signal Weight | PSF SNR | M* | N* | SNR | FWHM | Eccentricity |
| integration | 28.77 | 509.36 | 6.16E-05 | 9.05E-05 | 9.98 | 4.856 | 0.596 |
| drizzle_integration | 40.30 | 876.19 | 4.33E-05 | 6.36E-05 | 66.97 | 5.006 | 0.582 |
| drizzle_integration_normalized | 40.30 | 876.15 | 5.15E-05 | 7.55E-05 | 67.15 | 5.006 | 0.582 |
Quantitatively, this would seem to demonstrate that the drizzle result is superior on most important metrics. Visually, none of these differences are apparent when comparing the integration and normalized drizzle integration with the same STF applied. The only visible difference I can see is that the drizzle result seems less sharp than the regular integration.