W3C home > Mailing lists > Public > www-svg@w3.org > July 2006

Re: Gaussian Blur with very small stdDeviation

From: Dr. Olaf Hoffmann <Dr.O.Hoffmann@gmx.de>
Date: Sat, 8 Jul 2006 19:05:17 +0200
To: www-svg@w3.org
Cc: amenzie@us.ibm.com
Message-Id: <200607081905.17410.Dr.O.Hoffmann@gmx.de>

Hello,

Just my observations about this:
I think in your example there are several reasons, why 
the blurred element is still visible or the result is not as
expected.

1. second scale is 1100, first scale only 1200
2. stdDeviation="0.001" + transform="scale(1200)"
    is not small compared to 'one pixel'.
3. I checked the behaviour with other examples and
    discovered some minor errors in Opera9 (maybe
    related to optimisation of processing time and 
    used working memory, maybe in parts avoidable,
    I'm not sure).
4. The adobe plugin seems to stop calculating 
    those filters completely if the dimension of
    the filter gets too big (width and height 
    attributes of the filter element).
    Even if it calculates, sometimes remain some
    smaller errors for really neglectable standard
    deviations.  

If you want to look into details, it might be
useful to animate width and height of filter.

I think, in a real world with limited capabilities of
computers one cannot expect a complete perfect
behaviour for those filters in an acceptable time.
Therefore one has to take a very close look, why
something is not completely perfect - sometimes
it is only a compromise with current technical
possibilities. Of course, sometimes it is a mixture
with minor errors, too ... 
Received on Saturday, 8 July 2006 17:03:57 GMT

This archive was generated by hypermail 2.3.1 : Friday, 8 March 2013 15:54:35 GMT