Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: $Revision: 1.6.2.16 $; site datacube.UUCP
Path: utzoo!linus!philabs!cmcl2!seismo!harvard!think!datacube!shep
From: shep@datacube.UUCP
Newsgroups: net.graphics
Subject: Re: FFT of image in sections..
Message-ID: <6700030@datacube.UUCP>
Date: Mon, 26-Aug-85 11:35:00 EDT
Article-I.D.: datacube.6700030
Posted: Mon Aug 26 11:35:00 1985
Date-Received: Fri, 30-Aug-85 00:08:07 EDT
References: <298@ur-laser.UUCP>
Lines: 21
Nf-ID: #R:ur-laser:-29800:datacube:6700030:000:1052
Nf-From: datacube!shep    Aug 26 11:35:00 1985


>/**** datacube:net.graphics / ur-laser!nitin / 11:18 am  Aug 19, 1985 ****/
>Given this information, my question is, can we now process the megabyte image
>in such sections and get any increase in speed. Also, does linear system
>theory allow such a process.. because an FFT of a part is NOT the FFT of the
>whole image. 
>/* ---------- */

You -can- use linear systems theory... The DFT is separable! So you
break the n*m image into m-rows and n-columns. You may then process
the more manageable 1-dimensional arrays in core. I grant you that
separability speed-ups may be easier to take advantage of with
pipelined, not-in-place FFT butterflies; but it does reduce the amount
of memory required in core -and- reduce the computron requirement.
Look at "Digital Signal Processing", Oppenheim/Schafer page 320.

I know of no way to split the image up into 2-d sections and recombine
that will yield a computron reduction.

Shep Siegel                           UUCP: ihnp4!datacube!shep
Datacube Inc.; 4 Dearborn Rd.; Peabody, Ma. 01960; 617-535-6644