Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site brand.UUCP
Path: utzoo!linus!philabs!cmcl2!seismo!lll-crg!dual!qantel!hplabs!sdcrdcf!oberon!brand!shankar
From: shankar@brand.UUCP (Shankar Chatterjee)
Newsgroups: net.graphics
Subject: Re: FFT of image in sections ?
Message-ID: <159@brand.UUCP>
Date: Fri, 9-Aug-85 13:42:10 EDT
Article-I.D.: brand.159
Posted: Fri Aug  9 13:42:10 1985
Date-Received: Wed, 14-Aug-85 22:11:23 EDT
References: <360@ur-laser.uucp>
Reply-To: shankar@brand.UUCP (Shankar Chatterjee)
Distribution: net.graphics
Organization: U. of So. Calif., Los Angeles
Lines: 14
Keywords: Images, matrix transpositions, FFT

In most cases, when a 2-D FFT of an image is required, 1-D FFT is performed on
one row at a time if the image is too large to fit into the core. After this
the image in the disk is 'transposed' using some block allocation strategy. The
block size can be fixed or can be determined (alongwith the no. of passes over
the image) from the available core. [Note here that the transpostion has to be
done block by block to speed up processing. One could refer to J.O. Eklundh
(IEEE Tr. on Computers, vol. C-21, 1972) or a more generalized version of it in
H.K. Ramapriyan (IEEE Tr. on Comp., Dec. 1972). I had implemented a modified
version of the latter in software on a PE-3220 machine.] After this another
1-D FFT is performed on the rows. However some smart addressing scheme is 
needed (butterfly ?) to center the d.c. term.

Shankar Chatterjee         USC Signal & Image Processing Institute
shankar!brand%oberon%usc-cse@csnet-relay