![]() |
Blitz Devel : |
From: Peter Nordlund (petern_at_[hidden])
Date: 1998-08-14 10:08:32
I have tried the stencil-stuff for some image-processing
and I am very pleased with the performance so far.
For all small blitz-examples that I have compared with corresponding
lowlevel code written by people here in our lab
(not too thoroughly optimized but anyway),
the blitz-code has better performace, and is very much more readable.
I do not have any (very good) ideas how the following best should
be implemented, but on my wish-list are the following:
1. Image smoothing operations
2. Boundary-conditions, for 2-d arrays see examples below:
All new pixel values in the extension region
are set to zero. (ZeroExtension)
The image is extended using the last value of
each column/row. Hence, the extension will be
constant along straight lines perpendicular to
the previous image boundary. (LastValueExtension)
The image is mirrored at each boundary. Thus,
continuity in the normal direction will be
preserved at the previous image boundaries.
(MirrorBoundExtension)
The pixel values in the extended region are set
to the periodic extension of the input image.
The size of the input image is used as period.
(PeriodicExtension)
The input image is mirrored at each boundary.
Then the pixel values of each column/row are
reversed around the pixel value at the boun-
dary. Thus, continuity in the first normal
derivative will be preserved at the previous
image boundary. (MirrorAndReverseExtension)
//Peter Nordlund, CVAP, NADA, KTH
--------------------- blitz-dev list --------------------------------
* To subscribe/unsubscribe: mail to majordomo_at_[hidden], with
"subscribe blitz-dev" or "unsubscribe blitz-dev" in the body of the message
* Blitz++ web page: http://oonumerics.org/blitz/