Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

3 to 1 Scaling in the DCT Domain

IP.com Disclosure Number: IPCOM000103802D
Original Publication Date: 1993-Jan-01
Included in the Prior Art Database: 2005-Mar-18
Document File: 1 page(s) / 50K

Publishing Venue

IBM

Related People

Feig, E: AUTHOR

Abstract

An efficient method for scaling down 3 to 1 given that the data is presented as the output of a DCT on 8 points. The invention extends to multidimensions and is particularly suited when either efficient inverse DCT or scaled-DCT on 8 points (or 8x8, 8x8x8, etc.) is already available.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 79% of the total text.

3 to 1 Scaling in the DCT Domain

      An efficient method for scaling down 3 to 1 given that the data
is presented as the output of a DCT on 8 points.  The invention
extends to multidimensions and is particularly suited when either
efficient inverse DCT or scaled-DCT on 8 points (or 8x8, 8x8x8, etc.)
is already available.

      For the most important 2-dimensional case, where DCT data is
taken on 8 times 8 blocks (in both JPEG and MPEG standards), the
invention calls for transforming blocks comprised of the leading 3
times 3 DCT coefficients of each block of 8 times 8 DCT coefficients
with a 2-dimensional inverse DCT on 3 times 3 points.  Nine such
outputs are then used to form a 9 times 9 image Section, which is
then subjected to a forward 2-dimensional DCT on 9 times 9 points.
The output is a 9 times 9 set of DCT coefficients.  The last row and
column of this block are discarded, and the resulting matrix is
subjected to an inverse 2-dimensional DCT on 8 times 8 points.

      The first two parts can be combined to achieve significant
savings.  Namely, the inverse DCT on 3 times 3 points together with
the forward DCT on 9 times 9 can be written as a linear
transformation done row-column fashion, wherein the 1-dimensional
transform along each dimension is via the matrix product

                            (Image Omitted)

where F9 is the matrix of the DCT on 9-points and F3 is the matrix
of the DCT on 9-points.  When this product...