IMAGE STACKING NOTES (brief version)
------------------------------------
Compiled MJI April 2001-2002
Issues relevant for stacking and mosaicing
[Observe=Image*PSF*Shah] * = convolution x = multiplication
NB. defect rejection simplest if already done by "local" stacking
PSF not varying
---------------
1. Lauer, T, 1999 PASP, 111, 227 - coaddition in Fourier domain (solve
matrix equation to recover Nyquist sampling), requires precise
fractional pixel offsets cf. WFCAM interleaving
2. Projection onto convex sets eg. Adorf, H,-M, ADASS IV 456 - iteration
technique (similar to drizzle in a way) promising for automatic
PSF generation
3. Drizzle technique - used for HST, ESO ESI - benchmark method, works,
well, review of this and others: Fruchter & Hook astroph 980887, see
also Hook, Pirzkel & Fruchter, also worth checking ADS publs.
4. Standard approach - resample to align and "average" - many different
interpolation schemes; precis:
nearest pixel
bilinear (similar to 1:1 drizzle)
sinc interpolation (textbooks)
hyperbolic tangent eg. Devillard, N. ESO Messenger ++
bicubic spline eg. Alard & Lupton 1998 APJ 503 325
interpolate with optimum weights Fischer & Kochanski 1994 AJ 107 802
PSF varying
-----------
1. Image coaddition and restoration using Lucy-type methods (iterative
deconvolution eg. IRAF acoadd). Bayes/ML MEM argument non-linear
iterative methods, trade off is S:N -v- resolution, issues local
flux conservation. Enormous literature on subject. CPU intensive.
2. Nic Kaiser approach based on optimal image detection, form:
Y = sum_i W_i x X_i * S_i ith weight image seeing
define Z * sum_i W_i x S_i = Y solve for Z
interesting if only want optimal image detection
3. Also relevant are adaptive kernel matching techniques (devised for
difference imaging viz transient phenomena detection SNe, NEOs....)
eg. Alard C., and also Alard & Lupton 1998 APJ 503 325
4. Magain ESO messenger, and Magain, Courbin & Sohy, 1998, ApJ 494 472
deconvolution with correct sampling plus application for stacking
images with different seeing