Index: trunk/psphot/doc/notes.20090523.txt
===================================================================
--- trunk/psphot/doc/notes.20090523.txt	(revision 24583)
+++ trunk/psphot/doc/notes.20090523.txt	(revision 24583)
@@ -0,0 +1,41 @@
+
+20090523 : extended source analysis -- further development
+
+ issues to address:
+
+ * quality of moments used for initial psf selection
+   - what is the behavior of moments with Gaussian windowing as function of S/N & size?
+   - can we choose a sigma value that will give reliable moments for an expected range of seeing?
+
+ * moments for galaxy model guesses
+   - trade-off between Gaussian windowing, tophat windowing, and S/N windowing in footprint area
+
+ initial dev work:
+
+ * add Gaussian windowing, S/N windowing, (tophat windowing) and options to select
+
+ * check visualization of moments
+
+20090525 : some clarity of issues:
+
+ I've made some progress understanding why the moments analysis was
+ giving poor results.  I've added Gaussian (pseudo-gaussian)
+ windowing and the ability to set the per-pixel S/N limit.  The first
+ important conclusion was:
+
+ * the per-pixel S/N limit of 1.0 was driving the Mxx,Myy values low
+   for faint sources.  This should be completely disabled.
+
+ The next point is about the Gaussian vs tophat windowing.  One
+ interesting realization was that the Gaussian windowing (or tophat
+ for that matter) have a stronger impact on the measured moments for
+ sources with profiles that are non-Gaussian.  
+
+ * Gauss with seeing (FWHM) = 1.0 arcsec = 4pix -> sigma = 1.702 
+
+   Gauss window | measured Mxx
+   2.0	 	  1.75
+   4.0		  2.58
+   6.0		  2.85
+   
+ * QGAUSS with see 
Index: trunk/psphot/doc/stack.txt
===================================================================
--- trunk/psphot/doc/stack.txt	(revision 24583)
+++ trunk/psphot/doc/stack.txt	(revision 24583)
@@ -0,0 +1,38 @@
+
+20090606 : design notes on the multi-image photometry analysis
+
+  assumptions:
+  
+  * each input image represents the same sky pixels : they are warped to a common frame.
+
+  * each image has been previously processed, with the background subtracted and the psf model determined
+
+  basic outline:
+
+  * load images, masks, variance : for now, we should load all images.  in the future, we might be able to split this into segments.  
+
+  * option: model and subtract background for each image (should not be needed)
+
+  * option: smooth each image with a psf or fraction (not needed if stack smooths too much?)
+
+  * generate the significance (chisq) image: X_i = sum(f_i^2 / var_i)
+
+  * perform the peak detection on X_i
+
+  * for each of the input images:
+
+    * generate the significance image S_i = f_i^2 / var_i 
+    
+    * perform the peak detection on S_i
+
+  * merge the list of peaks
+
+  * generate the footprints
+
+  * linear fit to the peaks with the set of images (psf for each image)
+
+
+  design issues:
+  
+  * pmSource represents the analysis for an object on a single image
+  * extended pmSourceSet to group connected pmSources on multiple images?