Browse Prior Art Database

Two-Dimensional Modified Run-Length Coding for Quantized Image/Video Subband Signals

IP.com Disclosure Number: IPCOM000120846D
Original Publication Date: 1991-Jun-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 3 page(s) / 169K

Publishing Venue

IBM

Related People

Lancon, E: AUTHOR [+3]

Abstract

Run-Length coding can be applied to the coding of image or video subband signals. The original signal is split into several two- dimensional temporal or spatial subband signals which are quantized prior to be Run-Length encoding. Taking into account the specific properties of those signals, i.e., their two-dimensional representation and correlation, and their statistical distribution (zero-mean Laplacian or Gamma distribution), a modified version of the typical Run-Length algorithm is proposed. A pre-processing, which consists of performing center clipping over the subband signals, further improves the performances of the proposed coder without significantly degrading the quality of the reconstructed picture (subband reconstruction).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 39% of the total text.

Two-Dimensional Modified Run-Length Coding for Quantized Image/Video
Subband Signals

      Run-Length coding can be applied to the coding of image
or video subband signals. The original signal is split into several
two- dimensional temporal or spatial subband signals which are
quantized prior to be Run-Length encoding.  Taking into account the
specific properties of those signals, i.e., their two-dimensional
representation and correlation, and their statistical distribution
(zero-mean Laplacian or Gamma distribution), a modified version of
the typical Run-Length algorithm is proposed.  A pre-processing,
which consists of performing center clipping over the subband
signals, further improves the performances of the proposed coder
without significantly degrading the quality of the reconstructed
picture (subband reconstruction).  The modified Run-Length coder
improves by about 30% the performances of the typical Run-Length
coder.

      Except for the lowest temporal or spatial subband to which this
coding cannot be applied, the subbands signals have a mean value of
zero and a Laplacian or Gamma statistical distribution.  The subband
signals comprise a large number of low amplitude samples that
fluctuate around zero and some pulses that represent strong
variations in space (contours - spatial subband decomposition) or in
time (motion - temporal subband decomposition).  A compression gain
can already be achieved by quantization, the number of quantization
levels having an impact on the quality of the decoded signal, and by
the use of lossless run- length-type coders.  A Block Companded
(Laplacian) Quantizer is computed with respect to the power of the
subband signal. The lower the number of levels, the higher the number
of zeros, the more efficient the run-length-type coder.

      The traditional run-length scheme, applied to the coding of bi-
level images, consists in a succession of counters that alternatively
count the '0' and '1' areas along the lines of the image.  In our
case, we use a succession of counters and values.  The counters count
the subsequent zero pixels areas that are separated by non-zero
values (pixels).  Two types of codewords follow each other
alternatively: the counter-words and the value-words.  The number of
bits assigned to a counter-word corresponds to the maximum length of
a zero-area that can be coded.  The number of bits assigned to the
value-words corresponds to the number of possible quantization levels
but zero. In addition, the following particular cases are taken into
account which contribute to a loss of efficiency: - when the
zero-value area is larger than the maximum allowed by the number of
bits of the counter value (counter overflow), - when several non-zero
values follow each other (null counter).

      The subband signals can be viewed as comprising a large number
of low amplitude samples that fluctuate around zero and some pulses
that represent strong variations in space (co...