MathGroup Archive 1999

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Image-Sequence analyse with M4

  • To: mathgroup at smc.vnet.net
  • Subject: [mg19782] Re: Image-Sequence analyse with M4
  • From: Jens-Peer Kuska <kuska at informatik.uni-leipzig.de>
  • Date: Fri, 17 Sep 1999 01:36:40 -0400
  • Organization: Universitaet Leipzig
  • References: <7rnkhb$em7@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

Hi,

I can't help you with your image correlation, but I can give you some
of impressions. We have an image and signal processing laboratory
here and one of the tasks is to mask the brain from  MR 3d scan of 
a human head. The scan has a resolution of 250 x 250 x 190
voxels with gray scale data 0.. 255. The task is to create a 3d image
mask
from the data set. This needs binarization, masking, a distance
transform and 
a labeling of the connected voxels.
The data set is stored in the vista format and a MathLink C-function
load
the planes of the image. The original file is 12 MByte huge, Mathematica
need
32 MByte to store the data (Integer instead of unsigned char). Only the
binarization
of the data set is done in Mathematica, all othe image operations like
masking, bitand, distance transform .. are inside the MathLink program. 
Other versions to implement the bitand and masking operations in
Mathematica end up with a memory crash after several minutes or hours
...

Doing allmost every in the MathLink binary take a half hour on a SGI O2 
(324 MByte RAM, MIPS R10000) and need 200 MByte of memory. 
The pure C-implementation need 6 minutes.

Every time I tryed to make a Mathematica implementation of the image
operations
I found out that 
  a) it is to slow
  b) needs to much memory
thats why all the tasks are done in a MathLink program and the speed of
the C-code
weigths out the need of the transfer of up to 64 MByte data (image &
mask).

If you have a workstation with 2GByte of RAM and a 2 GHz CPU 
you may be able to handle your data inside Mathematica but a 4 GHz CPU
and more
RAM may be help full.

You will need more RAM than I, because FFT and wavelet transforms need
and generate
Complex or real data and it will be incredible slow.

As long as you only have to compare two succsesive images and keep not
more than
4-5 images in the memory it may work but it will still need more than
256 MByte	
RAM.

Image processing is well known to need a huge memory, Mathematica is
well known
to need a huge memory and both together are a deadly couple.

Hope that helps
  Jens

"R. Maurus" wrote:
> 
> Hello,
> i am working at a reserche projekt (bubble-observation)
> and i have problems in evaluating  my image sequences.
> 
> My image sequences exist of 500 Pictures ( bmp) and Size of 128x256
> Pixel with 8 bit grey.
> (Matrix of 128x256 with values from 0 to 255).
> My goal is to get statistical coefficents of cross- and autocorrelation
> (FFT or wavelet) of the
> changing value of my pictures.
> 
> My questions:
> There are tools in Mathematica 4 to load images and correlate them ?
> Which Application Package would solve my problem: (Experimental Data
> Analyst , Time Series, Wavelet Explorer)  ?
> 
> Are sequences of 500 Pictures still to handle ?
> Have someone experiences with image analysing or (FFT wavelet)
> applications with M4?
> 
> Cause it is a new field of reserche for me, i would be glade to get your
> help
> 
> Thanks in advance
> Reinhold


  • Prev by Date: Re: typesetting piecewise function
  • Next by Date: Re: Mathematica can't win against Tiger Woods
  • Previous by thread: Image-Sequence analyse with M4
  • Next by thread: Where's the Speed?