<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta http-equiv="Content-Type" content="text/html;charset=ISO-8859-1">
<title></title>
</head>
<body text="#000000" bgcolor="#ffffff">
Hi Pablo,<br>
<br>
In this email I talk about seams and control points.<br>
I have changed the subject line to reflect the most <br>
important idea.<br>
<br>
Pablo d'Angelo wrote:<br>
<blockquote type="cite" cite="mid20040503063922.GA2719@svalbart">
<pre wrap="">Hi Rik,
</pre>
<blockquote type="cite">
<pre wrap="">1. The PTStitcher masks often seem badly placed.
I would expect seams to be at the middle of overlap between
two images, so that every pixel in the output image comes
from the input image in which it is most central. PTStitcher
seems to do something different, so I was hoping to either
understand why I am thinking wrong, or fix the code.
</pre>
</blockquote>
<pre wrap=""><!---->
Hmm, the next version of nona will place the masks based on the nearest
feature transform, as implemented in enblend.
</pre>
</blockquote>
I don't know what "nearest feature transform" is, and I have<br>
not worked with enblend, so I can't follow what you wrote.<br>
<br>
However, I would like to share with you the strategy that<br>
I use to make panoramas. I think it is different from what<br>
others do, and I see a problem that I have not heard others<br>
talking about.<br>
<br>
Background is that I often shoot handheld panos, for example<br>
while backpacking in the wilderness. The most interesting<br>
ones always include some nearby objects as well as distant<br>
ones, and of course the nearby objects are vulnerable to<br>
parallax error as well as movement due to wind, etc.<br>
Even the clouds typically move quite a bit between frames.<br>
<br>
The strategy that I have evolved to deal with these<br>
difficulties goes like this:<br>
<br>
1. Roughly position source images into proper alignment.<br>
(I use the preview window of PTGui.)<br>
<br>
2. Estimate where the stitcher will put seams.<br>
<br>
3. Pick high quality control points near the seams.<br>
"High quality" means stable between frames<br>
and placed on visually important objects,<br>
places where a mismatch would be obvious.<br>
<br>
4. Pick additional control points needed to make<br>
step 6 work well.<br>
<br>
5. Optimize parameters.<br>
<br>
6. Render with morphing, so that all control points<br>
line up exactly. (It helps here if there are control<br>
points also away from the seam, to make the<br>
triangulation more uniform.)<br>
<br>
7. Adjust layer masks to minimize any remaining<br>
distracting seams.<br>
<br>
This strategy generally works quite well. However...<br>
<br>
There is one fundamental problem in the design<br>
of Panorama Tools and supporting gui's that makes<br>
it less effective than it could be.<br>
<br>
The problem is that control points come in pairs.<br>
<br>
Let me emphasize a bit here. <b><br>
<br>
</b>*** Control point PAIRS are the wrong model. ***<br>
<br>
The right model, from my standpoint, is that there<br>
are control points *in the world*. A world control<br>
point often appears in several images, and when<br>
it does, I want all of those images to line up, at<br>
that world control point, in the rendered image.<br>
<br>
But Panorama Tools and the gui's that I know about<br>
provide me no way to make that happen. I will not <br>
attempt to detail here what goes wrong. Suffice it to say<br>
that the morpher can generate nasty rendering<br>
artifacts if one image has two control points that<br>
are close together and are tied to separate other images.<br>
So I am forced to *avoid* using the same world<br>
feature as control in multiple image pairs, when using<br>
the morpher.<br>
<br>
What I would really like is:<br>
<br>
1. (Most important) Support for *world* control points<br>
in the data models, user interfaces, and underlying software.<br>
<br>
2. (In comparison, nothing else matters here.)<br>
<br>
I will not take space here to talk about ways in which<br>
this capability might be patched into Panorama Tools.<br>
<br>
The important thing is the idea that I want<br>
<b>control points that are to be matched across <br>
multiple images, not just pairs of images.</b><br>
<br>
Is this already included in the newer software,<br>
and I just don't know about it?<br>
<br>
--Rik<br>
<br>
</body>
</html>