Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!samsung!umich!sharkey!msuinfo!chroma.cps.msu.edu!raja
From: raja@bombay.cps.msu.edu (Narayan S. Raja)
Newsgroups: comp.graphics
Subject: Re: Stripe tease
Message-ID: <1990Dec13.114125.29972@msuinfo.cl.msu.edu>
Date: 13 Dec 90 11:41:25 GMT
References: <1990Dec11.031331.22486@msuinfo.cl.msu.edu>
Sender: news@msuinfo.cl.msu.edu
Reply-To: raja@cpswh.cps.msu.edu
Organization: PRIP Lab, Comp. Sci. Dept., MSU
Lines: 45


In article <1990Dec11.031331.22486@msuinfo>, I wrote:

< Some of my images have noise stripes (i.e.
< periodic noise at a particular orientation,
< that visually looks like stripes).
 

Well, there weren't any good tips about
this one, but I did get some requests to
summarize the info.

I managed to solve the problem anyway, so
here goes:  

1.  by separating the red, blue, and green 
      planes I found that the red image was 
      very stripy, while blue and green were 
      absolutely clean -- must have been 
      something goofy about the red sensors.  
2.  Moreover, the stripes were fairly narrow 
      and in any given image, were either 
      vertical or horizontal.

I was able to remove the stripes with very
little loss of sharpness by convolving the
stripy (red) image plane with a small mask 
tailored to average the intensities in either 
horizontal or vertical directions (depending 
on whether the stripes were vertical or 
horizontal.  In case of horizontal stripes, 
average the intensities vertically, and vice 
versa).

E.g., in case of vertical stripes, the mask
    0.11 0.00 0.11
    0.11 0.34 0.11
    0.11 0.00 0.11
worked well for me.  After de-striping the
red image satisfactorily, a better-quality
colour image is obtained by recombining the
r,g,b planes.


Narayan Sriranga Raja.