Changes between Version 1 and Version 2 of Durham_MDS_Testing2

Timestamp:

Sep 9, 2009, 5:13:22 AM (17 years ago)

Author:

shaun.cole@…

Comment:

--

Durham_MDS_Testing2

@@ -7 +7 @@
 
 [[Image(fullstack.png)]]
+
+This is a very small part of the 041 skycell stack from MD08 re-created at Durham from the 
+released warps. Looking at the intermediary convolved warps during processing you can see 
+that this small-scale high frequency ''mottling'' is already present, so this must be an 
+artifact of the convolution not the combination.
+
+The representative kernels used to convolve the warps are found in the third extension of 
+the ''.kernel'' files, and look like this:
+
+[[Image(fullkernels.png)]]
+
+Collapsing the data of the kernel along the X and Y directions and taking a single pixel slice, 
+shows the small-scale structure responsible for the ''mottling'':
+
+[[Image(fullmeankernel.png)]]
+
+[[Image(fullslicekernel.png)]]
+
+Note that convolving an image outside of IPP with this kernel confirmed it is the source of 
+this effect.
+
+
+== Improving the stack, aka removing the ''mottling'' ==
+
+Inspecting each of the skycells that contribute to the stack shows a range of imperfections, 
+some are quite serious, like double images and others less so, like trailing and low signal. 
+In general the image quality isn't good, with most images displaying some irregularities.
+
+To improve the stack and see what effect this would have on the matching kernels I removed 
+the images judged to be badly trailed, had double images or little signal, and re-stacked 
+(removing 6 of 23 images). This gave the following results. For comparison the before and is 
+shown again below the after.
+
+----
+
+[[Image(pickedstack.png)]]
+[[Image(fullstack.png)]]
+
+----
+
+[[Image(pickedkernels.png)]]
+[[Image(fullkernels.png)]]
+
+----
+
+[[Image(pickedmeankernel.png)]]
+[[Image(fullmeankernel.png)]]
+
+----
+
+[[Image(pickedslicekernel.png)]]
+[[Image(fullslicekernel.png)]]
+
+----
+
+It is immediately obvious that the mottling effect has been suppressed
+and that the matching kernels are somewhat less irregular, although still
+not symmetric or monotonic. This result must raise questions about the 
+selection of warps that are selected to create a stack.
+