Re: Fast gaussian convolution

From: Peter Heckert (ph_newsalias2_at_arcor.de)
Date: 02/22/05 

Date: Tue, 22 Feb 2005 22:00:25 +0100

Hello Hans-Bernhard,

Hans-Bernhard Broeker wrote:
> Peter Heckert <ph_newsalias2@arcor.de> wrote:
>
>
>>I am looking for a fast 1-dimensional gaussian filter algorithm.
>
>
> There ain't no such such thing as a "fast algorithm", without at least
> some specification of the parameters that define the problem. E.g.:
> how wide do you expect your filter to be?

 From 0.5 to 100 pixels where the pixel line is upto some 1000 pixels long.
Calculation will be in floating point.

> How big is the input? Are
> non-integer widths required?

Not necessarily. The sub-pixel case can be handled as special case,
because this already is very fast.

What input and output formats do you
> have?
8 bit integer, possibly 16 bit converted to floating point.
I want to make the "unsharp mask" faster for large radius and to do some
experiments just for fun.
>
>>It should work on an array and it must handle margin-condition.
>
> ... handle *which* margin-condition(s)?
When the kernel hits the start or end of the pixel row.
Of course this can be done by mirroring.
>
> You may want to ask the crowds over in comp.dsp or sci.image.analysis
> instead.
>
XPost & fup sci.image.processing

regards,

Peter

Relevant Pages

  • Re: Fixed vs Float ?
    ... computations can be just as fast as fixed-point. ... If the input to the digital filter is 8 bits wide and the coefficents ... Floating point calculations are always easier to code than fixed-point, if for no other reason than you don't have to scale your results to fit the format. ...
    (comp.arch.fpga)
  • Re: demonic problem descriptions
    ... It's not even USUALLY a disaster in common practice (which is to say, ... The most insidious thing about floating ... you have a better understanding of the representation than the ... calculation you make whose error estimate (which you won't compute, ...
    (comp.lang.lisp)
  • Re: Fixed vs Float ?
    ... I think that the added complexity of floating point in an FPGA will ... Fixed point implementations are often better than floating point ... filter structure that takes advantage of the long result. ... floating-point computations can be just as fast as fixed-point. ...
    (comp.arch.fpga)
  • Re: Testable Predictions by HdB
    ... > Jesse F. Hughes wrote: ... > floating point numbers often do not. ... This makes the Testable Predictions ... there are situations in which the calculation will give wrong ...
    (sci.math)
  • Re: Fixed vs Float ?
    ... computations can be just as fast as fixed-point. ... If the input to the digital filter is 8 bits wide and the coefficents ... Floating point calculations are always easier to code than fixed-point, if for no other reason than you don't have to scale your results to fit the format. ...
    (comp.arch.fpga)