[prev in list] [next in list] [prev in thread] [next in thread] 

List:       koffice-devel
Subject:    Re: Review Request: Krita: Gaussian Blur Filter
From:       schumifer () hotmail ! com
Date:       2009-12-19 14:48:18
Message-ID: 20091219144818.21402.89964 () localhost
[Download RAW message or body]



> On 2009-12-19 09:33:13, Cyrille Berger wrote:
> > Out of curiosity, I wondered about the speed impact on using floats instead of \
> > integers for convolution. But a bigger concern is why the use of "float" instead \
> > of "double" (or "qreal") ? 
> > Also, the UI for the gaussian blur mention the use of radius, I think it would \
> > makes more sense to use sigma (as it was done in the krita-plugin, and in (future \
> > version of) gimp), and compute the radius from the sigma.

Regarding the choice of floats vs doubles, I chose in favour of floats to save \
memory. However, I've noticed that for example FFTW3, uses doubles by default, so I \
should probably switch to doubles to make things right.

Regarding the use of Radius, I've never seen a mainstream application which uses \
anything else. Probably most people do not understand the meaning of a Gaussian \
Distribution, so I think it is to much to expect them to understand the meaning of \
Sigma in the context. Radius on the other hand has a very graphical meaning (radius \
of effect) and is therefore understood by layman.


- antiquark


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://reviewboard.kde.org/r/2369/#review3454
-----------------------------------------------------------


On 2009-12-14 10:30:27, antiquark wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://reviewboard.kde.org/r/2369/
> -----------------------------------------------------------
> 
> (Updated 2009-12-14 10:30:27)
> 
> 
> Review request for KOffice.
> 
> 
> Summary
> -------
> 
> This patch adds a real Gaussian Blur Filter (floating point kernel).
> 
> The horizontal and vertical components are computed separately (significant speed \
> gains especially with larger radii) by performing convolution twice. 
> In the process a number of changes have been made:
> 
> - Convolution computation is performed using floating point numbers.
> - Kernels are defined using matrices provided by the Eigen Library.
> - Removed older 3x3 Gaussian Blur Filter.
> - Updated all other convolution filters to new system.
> 
> 
> Diffs
> -----
> 
> /trunk/koffice/krita/image/kis_convolution_kernel.h 1062217 
> /trunk/koffice/krita/image/kis_convolution_kernel.cc 1062217 
> /trunk/koffice/krita/image/kis_convolution_painter_impl.h 1062217 
> /trunk/koffice/krita/image/kis_generic_colorspace.h 1062217 
> /trunk/koffice/krita/image/tests/kis_convolution_painter_test.cpp 1062217 
> /trunk/koffice/krita/plugins/colorspaces/ctlcs/KoCtlConvolutionOp.h 1062217 
> /trunk/koffice/krita/plugins/colorspaces/ctlcs/KoCtlConvolutionOp.cpp 1062217 
> /trunk/koffice/krita/plugins/filters/blur/CMakeLists.txt 1062217 
> /trunk/koffice/krita/plugins/filters/blur/blur.cpp 1062217 
> /trunk/koffice/krita/plugins/filters/blur/kis_blur_filter.cpp 1062217 
> /trunk/koffice/krita/plugins/filters/blur/kis_gaussian_blur_filter.h PRE-CREATION 
> /trunk/koffice/krita/plugins/filters/blur/kis_gaussian_blur_filter.cpp PRE-CREATION \
>                 
> /trunk/koffice/krita/plugins/filters/blur/kis_wdg_gaussian_blur.h PRE-CREATION 
> /trunk/koffice/krita/plugins/filters/blur/kis_wdg_gaussian_blur.cpp PRE-CREATION 
> /trunk/koffice/krita/plugins/filters/blur/wdg_gaussian_blur.ui PRE-CREATION 
> /trunk/koffice/krita/plugins/filters/convolutionfilters/convolutionfilters.h \
>                 1062217 
> /trunk/koffice/krita/plugins/filters/convolutionfilters/convolutionfilters.cpp \
>                 1062217 
> /trunk/koffice/krita/plugins/filters/imageenhancement/kis_simple_noise_reducer.cpp \
>                 1062217 
> /trunk/koffice/krita/plugins/filters/unsharp/kis_unsharp_filter.cpp 1062217 
> /trunk/koffice/libs/pigment/KoColorSpace.h 1062217 
> /trunk/koffice/libs/pigment/KoColorSpaceAbstract.h 1062217 
> /trunk/koffice/libs/pigment/tests/TestConvolutionOpImpl.cpp 1062217 
> 
> Diff: http://reviewboard.kde.org/r/2369/diff
> 
> 
> Testing
> -------
> 
> No visible regressions.
> 
> Sometimes the filter does not work correctly (esp. with larger kernels).
> 
> 
> Thanks,
> 
> antiquark
> 
> 

_______________________________________________
koffice-devel mailing list
koffice-devel@kde.org
https://mail.kde.org/mailman/listinfo/koffice-devel


[prev in list] [next in list] [prev in thread] [next in thread]