It appears Cliff's algorithm uses J's sort /:~ and I suppose that's where it 
spends most of its time. 
If you sort 169 small numbers 3072x4608 times in SAC you will have a lower 
limit of the time Cliff's algorithm requires, as compared to 47s in J (on my 
computer).

Ben
________________________________________
From: [email protected] [[email protected]] on 
behalf of Robert Bernecky [[email protected]]
Sent: Wednesday, April 24, 2013 20:10
To: [email protected]
Subject: Re: [Jchat] J for Image Processing

Hi, Cliff,

I'm looking to turn your medianf function into APL, but
will have to relearn J in order to do so. Can you give me a
quick translation into APLish of that?  If that's not trivial,
or if you're up to your butt in alligators, it's no big deal:
I'll dig out the J books and figure it out.

What I want to do is to compile the APL code into
SAC, then compile that and see how well it works in
both serial and parallel environments.

Thanks,
Bob

On 13-04-24 11:01 AM, Cliff Reiter wrote:
> In addition to my visualization book (OK fractals) you might look at
> preprints of APL quote quad articles
> http://webbox.lafayette.edu/~reiterc/j/withj/withj12_im_proc01_pp.pdf
> http://webbox.lafayette.edu/~reiterc/j/withj/withj14_im_proc02_pp.pdf
> and the vector paper
> http://archive.vector.org.uk/art10500860
> and ask the forum (or me) if you have questions -- I tend to always be
> a couple days out of date on the forum, but I do browse it.
> Those books and articles are best used with J6 but I hope that in a
> semester or two I will transition to J8.
> Best, Cliff
>
> On 4/24/2013 10:32 AM, Sarah Arnold wrote:
>> Hi Ben
>>
>> Yes, in mathematica it takes approximately 2 minutes to do the 13x13
>> medianfilter for a 2004 x 2006 picture... I usually use much bigger
>> boxsizes (about 120x120), so I can pick up a coffee while the program
>> is running.
>>
>> What you write about your use of J sounds promising to me. I ordered
>> the book from Reiter in the library and will have a look at it. The
>> tutorials you mentioned: are they somehow available?
>>
>> Thanks a lot for your answers!
>>
>> On 04/24/2013 04:14 PM, Ben Gorte - CITG wrote:
>>> By the way, Cliff Reiter's median in J isn't slow at all, for
>>> example compared to famous software like ImageMagick:
>>>
>>> $ time convert -median 3x3 p19.pgm m3.pgm
>>>
>>> real    0m11.961s
>>> user    0m22.333s
>>> sys    0m0.292s
>>> $
>>> $ time convert -median 13x13 p19.pgm m13.pgm
>>>
>>> real    1m13.204s
>>> user    2m22.517s
>>> sys    0m0.352s
>>>
>>> Good old xv uses 10s resp. 5 minutes.
>>>
>>> Ben
>>>
>>> ________________________________________
>>> From: [email protected]
>>> [[email protected]] on behalf of Ben Gorte - CITG
>>> [[email protected]]
>>> Sent: Wednesday, April 24, 2013 15:47
>>> To: [email protected]
>>> Subject: Re: [Jchat] J for Image Processing
>>>
>>>     6!:2 'med3 =: 3 medianf p19'
>>> 17.9526
>>>     6!:2 'med13 =: 13 medianf p19'
>>> 47.2474
>>>     $p19
>>> 3072 4608
>>>
>>> (A median 3x3 median filter on a 3072x4608 image takes 18s, and a
>>> 13x13 median takes 47s).
>>>
>>> This is on an Intel Core 2 Duo CPU at 2.00GHz , using the median
>>> function from the already-mentioned book of Cliff Reiter "Fractals,
>>> Visualization and J":
>>>
>>>     medianf =: 1 : '(2#m) "_ (<.-:*:m)&{@:(/:~)@,;._3 ]'
>>>
>>> I have a C program doing the same in 0.7s resp. 1.2s, but when I
>>> consider how much time it took me to write that one, I must compute
>>> an awful lot of medians with it before earning back that time.
>>>
>>> Ben
>>> ________________________________________
>>> From: [email protected]
>>> [[email protected]] on behalf of Sarah Arnold
>>> [[email protected]]
>>> Sent: Wednesday, April 24, 2013 15:02
>>> To: [email protected]
>>> Subject: Re: [Jchat] J for Image Processing
>>>
>>> Hi
>>> I should have been a bit more precise: I'm into astronomy. I'm mostly
>>> interested in noise reduction / signal enhancement and therefore apply
>>> different filters to the image. Some of this operations, like for
>>> example the MedianFilter[] function, take a lot of computation time.
>>>
>>> ----------------------------------------------------------------------
>>> For information about J forums see http://www.jsoftware.com/forums.htm
>>> ----------------------------------------------------------------------
>>> For information about J forums see http://www.jsoftware.com/forums.htm
>>
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm
>>
>

----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to