[ptx] Enblend 1.0

[Andrew C Mihal]

On Mon, 29 Mar 2004, alexandre jenny wrote:

> You made a great tool with enblend. It works great in many case ... But
> not all.  I uploaded here a result of a enblended panorama which shows a
> strange mis-interpolation line http://www.le-geo.com/temp/emblend1.jpg
> You can see that it breaks by a vertical line the luminosity of a single
> picture in that panorama.  I can provide you the inputs pictures if you
> need it for debug.

I noticed problems like that a couple times during development. It seems 
that occasionally the greedy assembly heuristic will make bad choices as 
to which images to blend next. If it picks an image with a long, skinny 
overlap region, there will not be enough space to use very many pyramid 
levels. Then there will almost always be a line in the result. Then in the 
next blending step, it will think that the line is really supposed to be 
there, and will blend future images against the line. Therefore the line 
will never totally go away. In cases like this the right thing to do is to 
blend the images in a different order. Save the image with small overlap 
for later, when there might be a bigger overlap region to use.

Another option would be to force it to use a specified number of blend 
levels no matter the size of the overlap region. Then enblend will use 
extrapolation to guess what happens off the edge of each image. But this 
can go wrong too. I've seen an example where parts of the image start to 
turn gray - I think this is because the blending zone exceeds the overlap 
zone too much.

So there should be a better way to automatically decide what image(s) to 
blend in next. For now you might try using the -s flag to specify an order 
manually and see if you can get that line to go away.

Andrew

---------------------------------
Andrew Mihal
www-cad.eecs.berkeley.edu/~mihal
mihal at eecs.berkeley.edu