software update: multipoint alignment and Theophilus
So I've been whiling away the last few days on a major overhaul to ppmcentre, including name change for the package and a few new features.
The most important new feature will be "multipoint alignment", otherwise known as image morphing. The basic idea is to do a normal align as best you can and then start distorting parts of the image to make it match up with your reference frame even better.
I'm still a ways off having it working properly, but I thought you might like to see the output of the software as it currently stands - there are many artifacts (the horizontal/vertical lines) which will go away when I implement more of the algorithm that I have so far. What I have so far is a very crude test implementation.
The thing to notice about this image is that it's equally sharp all the way across - the algorithm tries to get all of each frame to line up with the reference frame.
This is also just a stack of 220 frames, so it's a bit noisy. However when you compare it to the earlier Theophilus (in the other thread) remember that the other image is a stack of about 2000 frames using registax, and it gets blurry when you look closer to the edges.
Wow - this looks very interesting work indeed. Thanks for the report on your development work; it looks like it will be a major breakthrough in image alignment and stacking. When using Registax3, I try to place the alignment square over the area of interest in my image so I can optimize how it will turn out. This leaves the rest of the image to "look after itself" with sub-optimal results.
Thanks for posting the results and a very Merry Xmas and Prosperous New Year to you too!
Merry CHristmas Bird.. this looks great, as Dennis mentions if it can help to optimise sharpess and alignment over a wider part of the image frame it will be fantastically useful and sought after piece of software
How do you select the multiple areas? Or does it do it itself?
Will it work with avi's that are less-than-perfectly tracked?
Looking forward to trying it out!
The software is automatic at present, no user input required. It creates tiles across and down the image and then aligns each tile as best it can. Then I make a mesh and distort the image according to the offsets in those tiles.
It has extremely good tracking, better than registax :-) I'll make a demo avi later and show you something....
I think I'm on the right track, this version shows up a hairline thin fissure or rille that runs from the lower right of the crater rim across the floor of the crater and through a small craterlet. This isn't visible in either of the earlier versions, which leads me to believe that this image is showing more of the detail.
This is a stack of 2000/4000 frames, all going well tomorrow I'll try and stack up all 10,000 frames that I captured :-)
