Re: VIC2 images

From: Michael Huth (
Date: 2008-05-23 02:17:29

Wolfgang Moser wrote:
>> This probably has happened at some column and also ptgui had problems 
>> there finding control points.
>> Though it doesn't kills information.
> That's true, it does in no way harm later
> structure analysis.
> Nevertheless I may find it funny to preprocess
> the pieces a bit, if you're able to make them
> available too.
Ok I made them available under:

Skip this if you don't want a short introduction how to stitch rows 
together with ptgui.

Just a short note how I used PTGUI PRO 7.8. You won't need much memory 
just hdd space.
I put all images of one row in a directory, this is f.e. 0101.png , 
0201.png .... 1201.png.
So you got directories 1 to 14.
In ptgui click load images, choose tele lens (all images should be 
sorted in in reverse fitting order in source images - this is ok as long 
as they are not random, this is important for yourself interpreting the 
control point table later)
click Align Images, ignore complaints about problems of control point 
detection as you will need to adjust some parameters by yourself either way
click Advanced in the upper right corner
Tab Lens Setting set Field of View to 6 degree, set a,b,c lens parameter 
to 0. (as init for optimizer)
Tab Panorama Setting, choose Cylindrical
Tab Image Parameter, click on naming of column Yaw, press 0, click on 
Pitch, press 0, click on Roll, press 0, all Columns should be set to 0 
now (as init for optimizer)
Now choose Menu / Tools / Control Point Table, choose in the first drop 
down menu image 0, in the second any, click on second image columns 
naming to sort
Now check in the list how the points link the image as only subsequent 
images overlap, there should only be for image 0 control points to image 1.
For Image 1 control points with image 0 and image 2, image 2 with 1 and 
3 and so on. Delete all control points from the table that have more 
than one image step.
ok now it's time to check the control points visually, go to Tab Control 
Before start make sure all images were not rotated by the auto 
alignment. On the left side image 0 is chosen, go to drop down Rotate 
and press Pos1 to set it to 0.
Click Next, the tool switches to image 1, focus stays on rotate press 
Pos1 again, do it for all images.
Now choose image 1 on the left and image 0 on the right. There should be 
automated set control points in the overlapping area in the center. If 
they are there and in the right area you can be very sure the auto 
aligner has worked correctly.
Click next and both images will switch one further.
Case 1: no auto control points there, set a mark with mouse in the left 
and then click on the corresponding point in the right image. Repeat 
this about 8 times. Sometimes the tool recognized the pattern early and 
will choose the corresponding point by itself. Sometimes if it is unsure 
it just places the mouse. In most cases auto placing works with the 
third or forth set point.
Case 2: auto control points all over the image and definately not only 
in the overlapping area. Choose each point by hand or by control point 
table window and delete them. Go on as in Case 1.

Now you are ready to setup the optimizer:
Tab Optimizer, for a first try choose minimize lens distortion medium, 
click advanced, clear the checkboxes for roll as the original images are 
not rotated and it would make the final picture bend.
Now Yaw and Pitch should be checked except for the first anchor image.
Now click Optimize, result should be good, very good or too good to be true.

Press CTRL-E and check the generated panorama preview. Adjust the field 
of view so that everything is visible (sliders at the borders), the 
black border of the panorama will be saved transparent, don't worry if 
you size it a bit too big.
Preview should look ok. If not check control points and optimization 
init values again. Are there wrong crosslinked images?
Choose Tab Create Panorama, Set Optimum Size Maximum, should be around 
22000x1800 depending in your FOV.
Choose TIFF for lossless save and off you go.

Much text but I think it is after a bit practice alot faster for rows 
than stitchng manually in photoshop.
If you try to adjust the original images in the first place, ptgui 
automatic routines will probably even work better with some images.
On the other hand if ptgui can't find control points between two images, 
check if they are visually more different than the others.

> Hmmm, maybe....
> In the upper quarter to the left, this must
> be the line buffer with 12 rows of presumably
> 40 registers (D-flipflops or what?). Oliver
> Achten mentioned that in the Forum-64 thread
> where this story started for you.
Hmm I wonder myself where have all the electronic gurus gone?
> Hmmm, if I should be able to dig up some other
> 65xx chips in ceramic package, would you like
> to 'scope them also?
> Well, sure, this would be pretty useless until
> no one pops up who feels able to actually do
> some circuit reengineering with these, but _I_
> even like just watching to the circuits very much.

I can take shots of chips now and then. I also got the request to do 
this for a SID recently.
Since my time is limited to 24h a day, I can't push this too much.
So you might want to discuss how important it would be to map the die of 
a specific chip, not only for a 'nice to have' cause but also if there 
might be some real information gain in taking a look at the circuits.

Still for the VIC2 images I think it is in the 'nice to have' area, or 
did someone discovered something new or clarified some assumption?
Maybe put up some priority list.

Another thing is to make the data available. I'am still not sure after 
how many Gigs my provider will complain, but I would like to see 
everything on a more official server.
Could be connected to a Wiki to add information...


       Message was sent through the cbm-hackers mailing list

Archive generated by hypermail pre-2.1.8.