home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!cs.utexas.edu!sun-barr!ames!ncar!noao!amethyst!organpipe.uug.arizona.edu!delphinus.as.arizona.edu!newberry
- From: newberry@delphinus.as.arizona.edu (Mike Newberry)
- Newsgroups: comp.graphics
- Subject: Re: Blur Convolution kernel
- Keywords: Image Processing
- Message-ID: <1992Dec25.220644.27857@organpipe.uug.arizona.edu>
- Date: 25 Dec 92 22:06:44 GMT
- References: <47672@ogicse.ogi.edu>
- Sender: news@organpipe.uug.arizona.edu
- Organization: University of Arizona, Tucson, AZ
- Lines: 16
-
- In article <47672@ogicse.ogi.edu> bartb@ogicse.cse.ogi.edu (Bart Butell) writes:
- >I am looking for a convolution kernel (for both a 3x3 and a 5x5) that
- >will perform a BLUR on an image. Does anyone have any leads or have
- >them?
- >
- >BarT
-
- Just make an equal weighted circle or a Gaussian in a 5x5 box. That'll
- smooth the image, if that is what you want. 25 equal weights is the simplest
- and will smooth the thing using what is commonly called a "boxcar" filter
- but it leaves square artifacts at every point where there is a single "hot"
- or "dead" pixel in your image. Any convolution kernel will replicate
- itself at such pojnts in an image, but filters with circular symmetry
- leave an artifact that is usually similar to the point spread function unless
- there is mothing blur already in the image. Then you tailor your filter
- shape (contours) to the shape of the psf caused by the motion.
-