Browse Prior Art Database

Method and Apparatus for Implementing Scaled or Non-scaled DCTs

IP.com Disclosure Number: IPCOM000108359D
Original Publication Date: 1992-May-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 2 page(s) / 60K

Publishing Venue

IBM

Related People

Feig, E: AUTHOR

Abstract

Disclosed are software and corresponding hardware implementations for the scaled and the non-scaled inverse DCT on 8 x 8 points, which can be used for fast decompression of image data. Extensions of this invention to arbitrary multi-dimensional scaled and non-scaled DCTs and inverse DCTs on input sizes which are powers of 2 are straightforward. Each computation path will compute only one multiplication, which means that the bit accuracy requirement for the implementation is considerably relaxed.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 65% of the total text.

Method and Apparatus for Implementing Scaled or Non-scaled DCTs

       Disclosed are software and corresponding hardware
implementations for the scaled and the non-scaled inverse DCT on 8 x
8 points, which can be used for fast decompression of image data.
Extensions of this invention to arbitrary multi-dimensional scaled
and non-scaled DCTs and inverse DCTs on input sizes which are powers
of 2 are straightforward.  Each computation path will compute only
one multiplication, which means that the bit accuracy requirement for
the implementation is considerably relaxed.

      The device is sectioned into two parts.  The first part
performs only permutations, shifts, sign changes and vector sums (on
8-point) vectors.  The second part performs eight identical
one-dimensional scaled or non-scaled inverse DCTs, and then outputs
the data in some predetermined order, and may be implemented with
off-the-shelf devices.

      The first part is best described as follows:  perform a
modified one-dimensional inverse DCT (MIDCT) on the input, where each
column in the 8 x 8 input array is considered an input point (vector)
in the MIDCT, each addition of the IDCT becomes a vector addition in
the MIDCT, and each multiplication in the IDCT becomes a modified
vector multiplication in the MIDCT.  Choosing any of the various
known efficient one-dimensional DCT (the best use 29 additions and 13
multiplications) each multiplication is by a factor

                 ...