Browse Prior Art Database

3 X 3 Transform Coding of Images with Super Fast Decoding

IP.com Disclosure Number: IPCOM000109376D
Original Publication Date: 1992-Aug-01
Included in the Prior Art Database: 2005-Mar-24
Document File: 2 page(s) / 48K

Publishing Venue

IBM

Related People

Feig, E: AUTHOR [+3]

Abstract

Presented is an efficient method for compression coding of images, with a corresponding fast decoder. The image is sectioned into 3 x 3 blocks of pixel values. Each such block M is subjected to a transform of the form N = CMCt where the 3 x 3 matrix C is 1 1 1 1 0 -2 1 -1 1 The resulting output N is then scaled by pointwise division by (or equivalently, pointwise multiplication by the reciprocals of the entries in) a quantization matrix (Image Omitted) The resulting values are then either rounded to nearest integer or truncated to least integer in absolute value, and these are then coded losslessly using either Lempel-Ziv, Huffman, or any other entropy coder combined with run-length coding.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 64% of the total text.

3 X 3 Transform Coding of Images with Super Fast Decoding

       Presented is an efficient method for compression coding
of images, with a corresponding fast decoder.  The image is sectioned
into 3 x 3 blocks of pixel values.  Each such block M is subjected to
a transform of the form
                          N = CMCt
where the 3 x 3 matrix C is
                         1    1    1
                         1    0   -2
                         1   -1    1
The resulting output N is then scaled by pointwise division by (or
equivalently, pointwise multiplication by the reciprocals of the
entries in) a quantization matrix

                            (Image Omitted)

The resulting values are then either rounded to nearest integer or
truncated to least integer in absolute value, and these are then
coded losslessly using either Lempel-Ziv, Huffman, or any other
entropy coder combined with run-length coding.

      At the decoder, the coded data is first decoded, and quantized
transform data from each 3 x 3 block are descaled by pointwise
multiplication by a descaling matrix
to form the 3 x 3 matrix N.  This resulting matrix is then subjected
to a transform
                           M = CtNC

      Disclosed too are particular choices of quantization and
corresponding descaling matrices.  For the quanti...