| | 28 | |
| | 29 | * Noise model fix: |
| | 30 | * Comparison with magictest.%.20100210% reveals fewer streaks, with factors ranging from 0.2 to 1.0 (except at very small counts) |
| | 31 | * Sent before/after fix comparison data to MOPS |
| | 32 | * Majority of streaks in exposure with most streaks appears to be due to bad convolution kernel; might be identifiable using low avg(diffSkyfile.norm)/std(diffSkyfile.norm), but this is unlikely to be generally useful since normalisation is a function of number of input images |
| | 33 | * One effect causing bad convolution kernels is from using (convolved) stacks as templates, where the noise in the stack is sufficiently low as to clip the tops of stars of moderate brightness, and then the diff is forced to use faint stars for stamps |
| | 34 | * Some badly subtracted stars masked CONV.BAD or CONV.POOR contribute to magic streaks; adding these mask bits can reduce streaks, 23 --> 7 in this data. |
| | 35 | * Hough threshold calculation does not include duty cycle. Reducing the Hough threshold by the minimum duty cycle helps find long, faint streaks with low duty cycle. It does introduce more false positives than simply reducing the image threshold to 2.0 (17 streaks from tweak, 9 streaks from threshold reduction), so perhaps a combination of the two would work well? Concerned about impact on shorter exposures. |
| | 36 | * Shorter exposure has 14 streaks from tweaking the code, 19 streaks from reducing image threshold, and 10 streaks without either. |
| | 37 | * Artifacts contributing to streaks in all modes: bathtub rings, diffraction spikes (for bright stars, outer regions aren't masked because angle is slightly off), burn cross-talk. |
| | 38 | * Running all exposures through magic with threshold = 2.0, and separately with code tweak |
| | 39 | * Code tweak is preferred to lowering threshold since the median streak count is 40 vs 50. |
| | 40 | * Sent report to Paul Sydney |
| | 41 | * Extracted and delivered microtest data |
| | 42 | * Jacobian when warping: |
| | 43 | * Average plate scale correction necessary for flux conservation; has been applied for several months |
| | 44 | * Differential plate scale correction shouldn't be necessary, since flat-field correction does that |
| | 45 | * To double-check, processing STS image through warp to check photometry |
| | 46 | * Differential plate scale correction is probably necessary for covariance scale change |
| | 47 | * Chip-warp magnitudes show unexplained offset of ~0.05 mag, and positional variation of ~ 0.015 mag (peak-peak). |
