[ptx] new reference for stitching algorithms

[Michael Norel]

Hi all! 

hi Andrew C Mihal !

I placed new version (1.1)of smartblend on 
http://www.minorlogic.com/projects/smartblend/index.htm

The main additions is Distance Error (the same
distance transform that enblend use ) added to
perpixel error function.

The perpixel error of seam area have 3 components
1. Direct error between images.
2. Error between high pass images (avoid the high pass
features that is noticeble by eye ).
3. Distance error that tend to place seam line in the
center of overlap zone.


Options
 -SeamVerbose  
and 
 -PyramidVerbose 
let you see all intermediate states of processing

 -SeamVerbose   - all with searching best seam line 

 -PyramidVerbose  - all multiresolution stuff. Here is
used approach like multiresolution blend , but used
MUCH faster transforms with respect of alpha channel. 

I know that you are using "boost" library. There is a
several solutions of MaxFlow/MinCut in that library.
Solution of Min Cut graph algorithm is the solution of
Max Flow with a little addition  steps. So you can
easy add the Minimal Seam error into Enblend. 

Second point is : smartblend use the FASTER pyramidal
operations. For example you can use Expand and Reduce
operations based on Good known wavelet bidirectional
transforms (for example such used in IW44 or JPEG2000
5/3 FDWT IDWT ) just ignoring the Hi frequensis
component.

//-------------------------------------------------------
Hi Aseem Agarwala ! 
Very impressive software. Good publication, and
exelent refrences.

Stitch in gradient domain is very promising but need
some more speed improvements.
Some time ago i experimented with GIST, and big
problem was speed. I hope you can solve it or
implement multiresolution approach.

//-------------------------------------------
Hi Pablo d'Angelo.

Yes smartblend based on similar fundamentals that
descrieabed in Aseem publication. 



		
__________________________________ 
Do you Yahoo!? 
Read only the mail you want - Yahoo! Mail SpamGuard. 
http://promotions.yahoo.com/new_mail