[ptx] Re: [PanoTools-devel] automatic gain adjustment

[douglas wilkins]

Another approach:
>From a paper I found at http://grail.cs.washington.edu/projects/multipano called
Photographing long scenes with multi-viewpoint panoramas. ACM Transactions on Graphics (Proceedings of SIGGRAPH 2006), 2006 

The relevant text:
The next pre-processing step is to compensate for exposure variation
between the source photographs, which is a common problem
for panoramic stitching [Uyttendaele et al. 2001]. 
.....
We associate a brightness scale factor Ki to each image, and for two 
photographs Li, Lj we assert that Ki x Li = Kj x Lj for pixels that depict the same geometry.
Each SIFT point match between two images gives us three linear
constraints of this form (one for each color channel). We solve
for the values of Ki that best meet these constraints in a least-squares
sense by solving a linear system. (The linear system is defined only
up to a global scale, so we include a weak prior that each Ki = 1.)
 
This is the kind of compensation that perhaps could be implemented within hugin.

Doug
 
----- Original Message ---- 
From: Bruno Postle 
 
To: Panorama Tools software developers  
Sent: Monday, 5 June, 2006 3:08:00 PM 
Subject: Re: [PanoTools-devel] automatic gain adjustment 
 
On Mon 05-Jun-2006 at 13:13 -0700, Daniel M. German wrote: 
> 
> Bruno> I have used pfscalibration to calculate and correct the curve: 
> 
> Bruno> http://www.mpii.mpg.de/resources/hdr/calibration/pfs.html 
> 
> Bruno> pfscalibration also includes a tool to read the JPEG EXIF header and  
> Bruno> determine the exposure from that. 
 
> I guess this means that there is no need to include this functionality 
> in PTmender. Users can run their images through pfscalibration before 
> stitching them. 
 
pfscalibration just does merging to a single HDR file, it would need  
some work to do this job.  I would also say it's a bit flakey at the  
moment. 
 
However it is a good tool for calculating the curve itself, this  
might be useful to feed to PTmender rather than the current  
histogram matching stuff. 
 
--  
Bruno 
 
 
_______________________________________________ 
PanoTools-devel mailing list 
PanoTools-devel at lists.sourceforge.net 
https://lists.sourceforge.net/lists/listinfo/panotools-devel