suicao / aforge

Automatically exported from code.google.com/p/aforge
Other
0 stars 0 forks source link

Bilateral filter #283

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
Bilateral filter added to AForge.Imaging.Filters.Smooting.

Details on the filter can be found here: 
http://saplin.blogspot.com/2012/01/bilateral-image-filter-edge-preserving.html

Please find the .cs file attached.

Original issue reported on code.google.com by smaxm...@gmail.com on 29 Jan 2012 at 10:52

Attachments:

GoogleCodeExporter commented 9 years ago

Original comment by andrew.k...@gmail.com on 2 Feb 2012 at 9:50

GoogleCodeExporter commented 9 years ago

Original comment by andrew.k...@gmail.com on 2 Feb 2012 at 8:55

GoogleCodeExporter commented 9 years ago
1) Documentation, like the one shown below sounds quite funny and confusing. To 
me it sounds like: "Not sure what is the point of this parameter, but you can 
play with it." If you know what the property is for, why not to explain it in 
the code docs?
"/// You may play with it if you'd like to but in most cases default values 
should be fine"

2) Tweaking the algorithm for 24bpp (as you wrote in comments) does not have 
much sense. If you did not hard code pixel size to 3, but checked it from the 
provided image, the filter would easily work with pixel formats like 32 bit 
RGB. Also it can work for 32 bit ARGB, if we do smoothing of RGB channels only 
and ignore Alpha value. Also 8bpp grayscale version does not seem to be too 
difficult to implement.

Original comment by andrew.k...@gmail.com on 3 Feb 2012 at 9:59

GoogleCodeExporter commented 9 years ago
Good remarks.
1. I've got a mathcad file which shows a func for Power factors, it's not thet 
easy to describe... But I'll try to find some other words.
2. Hardcoding pixel size let me avoid extra loops inside the current loop 
(which is already 4 level nested). I'm quite concerned about the performance of 
the filter since it's quite bulky in it's core algorithm. Due to little tricks 
(except parallel processing) I've improved the original implementation (with 
single method, multiple pixel formats, functions evaluated in the loop body 
etc.) to smth about 2x-2.5x.
If from your experience you consider that having native support (without 
transformations) of 8bpp and 32bpp formats (any other) I may suggest not the 
best OOD but good from performance point of view aproach with presenting a 
number of Process methods which will work with specific pixel formats.

Original comment by smaxm...@gmail.com on 4 Feb 2012 at 7:04

GoogleCodeExporter commented 9 years ago
In the attachment you may find an updated .cs file. 
I've amended XML summaries for XxxPower properties and also added support for 
8bpp and 32 bpp formats.
The file increased in size significantly due to much repeating code which is 
necessary to get some performance benefits. I used the same approach and was 
targeting to have as few calls/operations in loop bodies as possible.

What do you think?

Original comment by smaxm...@gmail.com on 6 Feb 2012 at 2:36

Attachments:

GoogleCodeExporter commented 9 years ago
[deleted comment]
GoogleCodeExporter commented 9 years ago
Is it a bug?
spatialFunc[i, k] = M.Exp(-0.5 * (M.Pow(M.Sqrt((i - c) * (i - c) + (k - c) * (k 
- c) / spatialFactor), SpatialPower)));

>> (i - c) * (i - c) + (k - c) * (k - c) / spatialFactor

Should it be ???
( (i - c) * (i - c) + (k - c) * (k - c) ) / spatialFactor

Original comment by andrew.k...@gmail.com on 6 Feb 2012 at 8:42

GoogleCodeExporter commented 9 years ago
You're right, brackets were missing. The updated file is attached.

P.S.: Another code review pro.

Original comment by smaxm...@gmail.com on 7 Feb 2012 at 8:00

Attachments:

GoogleCodeExporter commented 9 years ago
I would say the documentation is still confusing ... Did not find any 
difference in documentation for SpatialPower and ColorFactor - it is identical. 
I can see these value are used to initialize different table, so I would expect 
they should have slightly different effect.

Original comment by andrew.k...@gmail.com on 8 Feb 2012 at 8:32

GoogleCodeExporter commented 9 years ago
The code is merged in development trunk after code review.

Updating Bilateral filter implementation after code review:
1) Did some polishing of the code and documentation so it looks more consistent 
to the rest of the framework;
2) Removed confusing documentation like “play with it and discover 
yourself”;
3) Fixed implementation of partial filtering, when a filter can be applied to 
rectangle of an image (original implementation did not care much about it);
4) Did some minor improvements, which result in slightly better performance 
(tests need to be done in Release build running a filter 100+ times to see any 
difference).

TODO: Documentation for filter properties still might be updated to clarify 
things.

Committed in revisions 1668-1669 (2 commits so result of code review could be 
seen as diff). Will be released in version 2.2.4.

P.S. If there are any further updates to the code, please, provide patch to the 
version in trunk.

Original comment by andrew.k...@gmail.com on 8 Feb 2012 at 9:44

GoogleCodeExporter commented 9 years ago

Original comment by andrew.k...@gmail.com on 23 Feb 2012 at 9:14