Re: Parallel processing in Octave

[Olaf Till]

On Sun, May 22, 2011 at 11:30:28AM +0200, Thomas Weber wrote:
> On Wed, May 18, 2011 at 03:41:31PM +0530, Chethan S wrote:
> > Hi all,
> > 
> > The code I am using is fairly simple.I am using it to process satellite 
> > images of dimension - 3000X2000 pixels. The code goes as follows:
> > 
> > 
> > function texlayer(subfn)
> > clc;
> > [fn,pn] = uigetfile({"*.TIF;*.tiff;*.tif;*.TIFF;*.jpg;*.bmp;*.JPG;*.png", 
> > "Supported Image Formats"}, ...
> >                    'Select an Image', "/home/");
> > I = double(imread(fullfile(pn,fn)));
> > global ld
> > ld = input('Enter the lag distance = '); % prompt for lag distance
> > fh = eval(['@' subfn]); % Function handles
> > I2 = uint8(nlfilter(I, [7 7], fh))
> > imshow(I2);
> > imwrite(I2,'i1.tiff');
> > 
> > % Zero Degree Variogram
> > function [gamma] = ewvar(I)
> >     c = (size(I)+1)/2; % Finds the central pixel of moving window
> > global ld
> >     EW = I(c(1),c(2):end); % Determines the values from central pixel to 
> > margin of window
> >     h = length(EW) - ld; % Number of lags
> >     gamma = 1/(2 * h) * sum((EW(1:ld:end-1) - EW(2:ld:end)).^2);
> > 
> > 
> > According to the above code, I call the function as texlayer('ewvar')
> > and browse and select the input image and give the lag distance which
> > is usually 1. The processing exceeds 45 minutes and I get the output
> > image. Is there any scope for improvement in this code or will it be
> > possible to implement parallel processing for the code? I have no idea
> > about how to implement parallel processing. So any help would be a
> > great time-saver for me.
> 
> I've looked briefly at the nlfilter() function - it seems that it's
> running something along
> 
> for i=1:2000
>   for j=1:3000
>     here be dragons
>   endfor
> endfor
> 
> where 2000 and 3000 are the dimensions of your images. I don't think
> there's much you can do from your code - rewriting the time-critical
> part of nlfilter as .oct file seems your best way out.
> 
>       Thomas

I think this could be parallelized with 'parcellfun' of package
'general'. You could either modify 'nlfilter' to use parcellfun or
write your own specialized code instead of nlfilter. Additionally, you
could try to speed up the single calls to 'ewvar' by making an
oct-file of it.

Olaf