631F01 Lec9

ENEE631 Fall 2001 Lecture-9 Lecture-
Review of Last Time

"
Predictive coding Transform coding JPEG compression standard

Baseline block-DCT based algorithm
# #
Image Compression (3) Wavelet Coding & Vector Quantization

Min Wu
Electrical & Computer Engineering University of Maryland, College Park
" "
lossy part: quantization with different step size for each coeff. Band lossless part: differential coding, run-length coding, Huffman
" "
Binary image coding Today

Wavelet coding Brief introduction to Wavelet-based JPEG 2000 Brief introduction to Vector Quantization (VQ)
M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [2]
! !
http://www.ece.umd.edu/class/enee631/ minwu@eng.umd.edu
M. Wu: ENEE631 Digital Image Processing (Fall'01)
Recall Haar Transform

"
Wavelet Transform for Image Compression

"
Two major sub-operations

Scaling captures info. at different frequencies Translation captures info. at different locations
ENEE631 emphasis
Conceptual aspects related to image compression Wavelet is also useful for denoising, enhancement, and image analysis For more info. on wavelet: ENEE624, wavelet math course, & other ref.
" "
Can be represented by filtering and downsampling Poor energy compaction of Haar transform
x 1
"
K-level 1-D wavelet decomposition

Successive lowpass/highpass filtering and downsampling
# #
on different level: capture transitions of different frequency bands on the same level: capture transitions at different locations
From Matlab Wavelet Toolbox Documentation M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [3] M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [4]
Examples of 1-D Wavelet Transform 1-
2-D Wavelet Transform via Separable Filters
From Matlab Wavelet Toolbox Documentation
From Matlab Wavelet Toolbox Documentation M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [5] M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [6]
Orthonormal Filters
" "
Solutions to Coefficient Expansion

"
Circular convolution
Periodic extension of input signal Problem: artifacts by large discontinuity at borders
Equiv. to projecting input signal to orthonormal basis Energy preservation property

Convenient for quantizer design
#
"
Symmetric extension of input

Reduce border artifacts Problem: output at each stage may not be symmetric
From Usevitch (IEEE Sig.Proc. Mag. 9/01)
MSE by transform domain quantizer is same as reconstruction MSE
"
Shortcomings: coefficient expansion

N-element input & M-element filter $ (N+M-1)-element output $ (N+M)/2 after downsample Length of output per stage grows ~ undesirable for compression
Lec9 Wavelet Coding & VQ 10/2/01 [7]
Solutions to Coefficient Expansion (contd)

"
Biorthogonal Wavelets
"
Symmetric extension + symmetric filters

No coefficient expansion and little artifacts Symmetric filter => linear phase filters (no phase distortion except by delays) Problem:
#
Biorthogonal
Some orthogonality between filters but not strict orthogonality Not preserving energy But could be close to orthogonal filters performance
"
Advantage
Covers a much broader class of filters
#
only one set of linear phase filters for real, FIR orthogonal wavelets $ Haar filters: (1, 1) & (1,-1)
including symmetric filters that eliminate coeff. expansion
"
Commonly used filters for compression

9/7 biorthogonal symmetric filter
Subband Coding Techniques

"
Embedded Zero-Tree Wavelet Coding (EZW) Zero"
General coding approach

Allocate different bits for coeff. in different frequency bands Encode different bands separately Example: DCT-based JPEG and early wavelet coding
Modern lossy wavelet coding exploits multi-resolution and self-similar nature of wavelet decomposition
Energy is compacted into a small number of coeff. Significant coeff. tend to cluster at the same spatial location in each freq. subband
"
Minor difference between subband coding and early wavelet coding ~ Choices of filters
Subband filters aims at (approx.) non-overlapping freq. response Wavelet filters typically designed for certain smoothness constraints
"
Shortcomings of subband coding

Difficult to determine optimal bit allocation for low bit rate applications Not easy to accommodate different bit rates with a single code stream Difficult to encode at an exact target rate
M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [11] M. Wu: ENEE631 Digital Image Processing (Fall'01) From Usevitch (IEEE Sig.Proc. Mag. 9/01)
Two Key Concepts of EZW

"
EZW Algorithm and Example

"
Significance map coding via zero-Tree

Encode only high energy coefficients
#
Initial threshold ~ 2 ^ floor(log2 xmax)

Put all coeff. in dominant list
Need to send location info. $ large overhead
"
Dominant Pass (zig-zag across bands)

Assign symbol to each coeff. & entropy encode symbols
# # # #
Encode insignificance map using zero-trees

"
ps positive significance ns negative significance iz isolated zero ztr zero-tree root
Successive approximation quantization

Send most-significant-bits first and gradually refine coeff. value Embedded nature of coded bit-stream
#
Significant coeff.
# #
move. to subordinate list put zero in dominant list
"
Subordinate Pass
Output one bit for subordinate list
#
get higher fidelity image by adding extra refining bits

According to position in up/low of quantization interval
"
Repeat with half threshold

Until bit budget achieved
After 1st Pass
After 2nd Pass
Beyond EZW
"
JPEG 2000: A Wavelet-Based New Standard Wavelet"
Cons of EZW
Poor error resilience Difficult for selective spatial decoding
Targets and features

Excellent low bit rate performance without sacrifice at higher bit rate Progressive decoding to allow from lossy to lossless Region-of-interest (ROI) coding Error resilience
"
Set Partitioning in Hierarchal Trees (SPIHT)

Further improvement over EZW
"
EBCOT
Used in JPEG 2000 Address the shortcomings of EZW
"
For details
Tutorial @ IEEE Trans. on consumer electronics 11/00 Links and tutorials @ http://www.jpeg.org/JPEG2000.htm
Examples JPEG2K vs. JPEG

From Christopoulos (IEEE Trans. on CE 11/00)
From Christopoulos (IEEE Trans. on CE 11/00)
DCT vs. Wavelet: Which is Better?

"
Summary of Compression Tools

"
3dB improvement?
Wavelet compression was claimed to have 3dB improvement over DCTbased compression Comparison is done on JPEG Baseline
Lossless compression tools

Entropy coding
# #
Huffman, Arithmetic, Lempel-Ziv run-length
Predictive coding
"
Improvement not all due to transforms

Main contribution from better rate allocation, advanced entropy coding, & smarter redundancy reduction via zero-tree DCT coder can be improved to decrease the gap
[Ref] "A comparative study of DCT- and wavelet-based image coding", Z. Xiong, K. Ramchandran, M. Orchard, Y-Q. Zhang, IEEE Transc. on Circuits and Systems for Video Tech., v.9, no.5, 8/99, pp692-695.
reduce the dynamic range to code
Transform
#
enhance energy compaction
"
Lossy compression tools

Discarding and thresholding Quantization
# #
( previous lectures ) Scalar quantizer Quantize more than one component => vector quantization
Vector Quantization
"
Outline of Core Parts in VQ

"
Encode a set of values together

Find the representative combinations Encode the indices of combinations
Design codebook
Optimization formulation is similar to MSE scalar quantizer Given a set of representative points
#
"
Stages
vector quantization of 2 elements
Nearest neighbor rule to determine partition boundaries
Codebook design Encoder Decoder

" "
Given a set of partition boundaries

#
Probability centroid rule to determine representative points that minimizes mean distortion in each cell
Scalar vs. Vector quantization

VQ allows flexible partition of coding cells VQ could naturally explore the correlation between elements SQ is simpler in implementation
From Boviks Handbook Sec.5.3 M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [23]
Search for codeword at encoder

Tedious exhaustive search Design codebook with special structures to speed up encoding
scalar quantization of 2 elements
E.g., tree-structured VQ
vector quantization of 2 elements
Summary
"
Assignment
"
Required readings
[Wavelet coding]
# # #
Wavelet coding
Filters for wavelet transf. EZW
Usevitchs tutorial in IEEE Sig. Proc. Magazine 9/01 Xiongs paper on DCT vs. Wavelet on IEEE Trans. CSVT MATLAB 2-D Wavelet Demos
(Univ. Librarys E-journal)
[JPEG2000] Christopoluos tutorial in IEEE Trans. CE 11/00

#
"
Brief introduction
JPEG 2000 Vector quantization
"
http://etro.vub.ac.be/~chchrist/paper_ieee_ce_jpeg2000_Nov2000.pdf
[VQ] Boviks Handbook Sec.5.3
Recommended readings
www.jpeg.org Grays tutorial paper on VQ Boviks Handbook Sec.5.4
"
Next time
Review of image processing basics and image compression
"
Reminder
Project BB1 Due Thursday before class 1st in-class exam ~ Next Thursday 10/11/2001
M. Wu: ENEE631 Digital Image Processing (Fall'01) Lec9 Wavelet Coding & VQ 10/2/01 [26]

631F01 Lec9

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

631F01 Lec9

Uploaded by

Copyright:

Available Formats

ENEE631 Fall 2001 Lecture-9 Lecture-

Review of Last Time

Predictive coding Transform coding JPEG compression standard

Image Compression (3) Wavelet Coding & Vector Quantization

Binary image coding Today

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Recall Haar Transform

Wavelet Transform for Image Compression

Two major sub-operations

K-level 1-D wavelet decomposition

Examples of 1-D Wavelet Transform 1-

2-D Wavelet Transform via Separable Filters

From Matlab Wavelet Toolbox Documentation

Solutions to Coefficient Expansion

Equiv. to projecting input signal to orthonormal basis Energy preservation property

Symmetric extension of input

MSE by transform domain quantizer is same as reconstruction MSE

Shortcomings: coefficient expansion

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [7]

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [8]

Solutions to Coefficient Expansion (contd)

Symmetric extension + symmetric filters

including symmetric filters that eliminate coeff. expansion

Commonly used filters for compression

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [9]

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [10]

Subband Coding Techniques

Embedded Zero-Tree Wavelet Coding (EZW) Zero"

General coding approach

Shortcomings of subband coding

Lec9 Wavelet Coding & VQ 10/2/01 [12]

Two Key Concepts of EZW

EZW Algorithm and Example

Significance map coding via zero-Tree

Initial threshold ~ 2 ^ floor(log2 xmax)

Need to send location info. $ large overhead

Dominant Pass (zig-zag across bands)

Encode insignificance map using zero-trees

ps positive significance ns negative significance iz isolated zero ztr zero-tree root

Successive approximation quantization

move. to subordinate list put zero in dominant list

get higher fidelity image by adding extra refining bits

According to position in up/low of quantization interval

From Usevitch (IEEE Sig.Proc. Mag. 9/01)

Repeat with half threshold

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [13]

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [14]

After 1st Pass

From Usevitch (IEEE Sig.Proc. Mag. 9/01)

After 2nd Pass

From Usevitch (IEEE Sig.Proc. Mag. 9/01)

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [15]

M. Wu: ENEE631 Digital Image Processing (Fall'01)

Lec9 Wavelet Coding & VQ 10/2/01 [16]

JPEG 2000: A Wavelet-Based New Standard Wavelet"