Full HTML for

Basic foilset Wavelet Video Compression

Given by Miguel del Rosario at Rome Lab Quarterly Review for CIV on October 1 96. Foils prepared 22 February 97
Outside Index Summary of Material

Objective: real-time wavelet video codec
Problems to Solve
Current Approach
All-pass Filters
Codec Complexity
Complexity Issues
All-pass Filter Data Flow

Table of Contents for full HTML of Wavelet Video Compression

 Denote Foils where Image Critical
Denote Foils where HTML is sufficient
1 Wavelet Video Compression
2 Project Objectives
3 State of the Research
4 Problems to Solve
5 Current Approach
6 All-pass Filters
7 All-pass Filters
8 Codec Complexity
9 Complexity Issues
10 All-pass Filter Data Flow
11 All-pass Filter Data Flow

Outside Index Summary of Material

HTML version of Basic Foils prepared 22 February 97

Foil 1 Wavelet Video Compression

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Miguel del Rosario, Tom Major
HTML version of Basic Foils prepared 22 February 97

Foil 2 Project Objectives

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Objective: real-time wavelet video codec
Current state of research:
  • Wavelet compression developed primarily in image processing context.
  • Zerotree algorithm has arisen as preferred quanitzation method.
  • Zerotree variants (e.g., vector zerotree) currently being proposed.
HTML version of Basic Foils prepared 22 February 97

Foil 3 State of the Research

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Current state of research: (con't) ...
  • Several wavelet filters have been established as suitable for image processing.
    • Filter is critical in determining transform quality and complexity.
    • Short filters (e.g., 5/3) reduces complexity significantly but does not always result in acceptable rate/distortion characteristics.
HTML version of Basic Foils prepared 22 February 97

Foil 4 Problems to Solve

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Problems with wavelet video compression:
  • Complexity of current codecs exceed real-time constraints.
  • Motion compensation not amenable for use with wavelet transform.
    • codec still requires full transform after estimation process.
  • Hybrid wavelet / H.263 motion is inferior to pure H.263 (with DCT).
HTML version of Basic Foils prepared 22 February 97

Foil 5 Current Approach

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Our present strategy:
  • Improve image codec to within real-time requirement.
  • Use image codec to produce intraframe video stream.
Our approach:
  • Develop wavelet transform code with all-pass filters in a polyphase network.
  • Incorporate transform into Said/Perlman algorithm.
HTML version of Basic Foils prepared 22 February 97

Foil 6 All-pass Filters

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
All-pass polyphase filters:
  • All-pass filters are IIR a opposed to FIR for QMF filters.
  • Non-causal requirement for perfect reconstruction nullified due to finite extent of images.
  • All-pass polyphase design reduces high order QMF filter bank to 0th and 1st order filters (much lower complexity).
HTML version of Basic Foils prepared 22 February 97

Foil 7 All-pass Filters

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
All-pass polyphase filters: (con't) ...
  • 0th and 1st order all-pass polyphase filter equivalent to order 16 QMF filter with factor 5 complexity reduction for 2x2 bands.
    • (J.H.Husoy and T.A.Ramstad, March 1990).
HTML version of Basic Foils prepared 22 February 97

Foil 8 Codec Complexity

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Complexity issues:
  • For tree structured FIR NxN band filter bank:
    • NM2log2(N) multiplications
  • Filtering for all-pass polyphase filter:
    • 2(N-1)(M/N)M multiplications
  • DFT requires (e.g., for 8x8 DFT):
    • 2M2 non-trivial multiplications
    • where NxN = no. of subbands, MxM = picture dimension,
    • and L = length of the FIR filter
HTML version of Basic Foils prepared 22 February 97

Foil 9 Complexity Issues

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
Complexity issues: (con't) ...
  • Complexity reduction from FIR filter bank to all-pass polyphase filter bank:
    • R = 1 L log2(N)
    • 2 (2-1 / N)
    • Previous table resulted from computation of typical values of N and L.
HTML version of Basic Foils prepared 22 February 97

Foil 10 All-pass Filter Data Flow

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
All-pass polyphase filter structure:
    • analysis/synthesis 2x2 filter bank with coder
z-1
2
2
H0(z)
H1(z)
IDFT
x(n)
coder
z-1
2
H0(z)
H1(z)
x '(n)
2
DFT
HTML version of Basic Foils prepared 22 February 97

Foil 11 All-pass Filter Data Flow

From Wavelet Video Compression Rome Lab Quarterly Review for CIV -- October 1 96. *
Full HTML Index
All-pass polyphase filter structure:
wavelet implementation of all-pass polyphase filters
z-1
2
2
H0(z)
H1(z)
IDFT
z-1
2
2
H0(z)
H1(z)
IDFT
© Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu

If you have any comments about this server, send e-mail to webmaster@npac.syr.edu.

Page produced by wwwfoil on Thu Aug 14 1997