ERICSSON RESEARCH Media Lab
1Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
JPEG2000JPEG2000The next generation still The next generation still
image coding systemimage coding system
Touradj Ebrahimi*, Charilaos Christopoulos**
*Ecole Polytechnique Federale de Lausanne, Switzerland
**MediaLab, Ericsson Research, Stockholm, Sweden
2Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Standards Organizations• International Organization for Standardization
(ISO)– 75 Member Nations– 150+ Technical Committees– 600+ Subcommittees– 1500+ Working Groups
• International Electrotechnical Commission (IEC)– 41 Member Nations– 80+ Technical Committees– 100+ Subcommittees– 700+ Working Groups
T
3Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ISO / IEC TerminologyISO / IEC Terminology• ISO: International Standardization Organization• IEC: International Electrotechnical Committee• ISO/IEC JTC1: Joint Technical Committee• SC29: Information Technologies
– WG1: still images, JPEG and JBIG• Joint Photographic Experts Group and Joint
Bi-level Image Group– WG11: video, MPEG
• Motion Picture Experts Group– WG12: multimedia, MHEG
• Multimedia Hypermedia Experts Group
4Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG: Summary
JPEG (Joint Photographic Experts Group) “Digital Compression and Coding of Continuous-tone Still Images”
• Joint ISO and ITU-T
• Published in 4 Parts:– ISO/IEC 10918-1 | ITU-T T.81 : Requirements and guidelines
– ISO/IEC 10918-2 | ITU-T T.83 : Compliance testing
– ISO/IEC 10918-3 | ITU-T T.84: Extensions
– ISO/IEC 10918-4 | ITU-T T.86: Registration of JPEG Parameters, Profiles, Tags, Color Spaces, APPn Markers Compression Types, and Registration Authorities (REGAUT)
5Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG: Summary (cont.)
JPEG derived industry standards
• JFIF (JPEG File Interchange Format, <xxxxxx.jpg>)
• JTIP (JPEG Tiled, Pyramid Format)
• TIFF (Tagged Image File Format)
• SPIFF (Still Picture Interchange File Format, JPEG Part 3)
• FlashPix– Developed by Hewlett-Packard, Kodak, Microsoft, Live Picture
(1996)
– Transferred to Digital Imaging Group (DIG), an industry consortium
6Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG 2000: Image Coding System
C
7Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Why another still image compression Why another still image compression standard?standard?
Low bit-rate compression: for example below
0.25 bpp
Lossless and lossy compression: No current
standard exists that can provide superior lossy and lossless compression in a single codestream.
Computer generated imagery: JPEG was
optimized for natural imagery and does not perform well on computer generated imagery.
In order to address areas that the current standards fail to produce the best quality or performance, as for example:
8Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Transmission in noisy environments: The
current JPEG standard has provision for restart intervals, but
image quality suffers dramatically when bit errors are
encountered.
Compound documents: Currently, JPEG is seldom
used in the compression of compound documents because
of its poor performance when applied to bi-level (text)
imagery.
Random codestream access and processing
Why another still image compression standard? Why another still image compression standard? (cont’d)(cont’d)
9Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Open Architecture: Desirable to allow open architecture to optimise the system for different image types and applications.
Progressive transmission by pixel accuracy and resolution
Why another still image compression standard?Why another still image compression standard? (cont’d)(cont’d)
10Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Internet Mobile Printing Scanning Digital
Photography Remote Sensing Facsimile Medical Digital Libraries E-Commerce
JPEG2000JPEG2000Markets and ApplicationsMarkets and Applications
11Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
The relation JPEG The relation JPEG JPEG2000 JPEG2000
• JPEG2000 is intended to complement and not to replace the current JPEG standards
12Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 DevelopmentJPEG2000 Development• Timeline
– Feb 96 (Geneva) started with original proposal– Nov 96 (Palo Alto) test method agreed– Mar 97 (Dijon) call for proposals – Jul 97 (Sapporo) requirements analysis started – Nov 97 (Sydney) algorithm competition & selection– VM 1 (Mar 98), VM 2 (Aug 98), split to VM 3A and 3B
Nov 98. Converged to VM4 and WD in Mar 99– Promotion to CD, FCD, FDIS as well as creation of
different parts
• Current status: VM 8, FDIS
13Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 contributorsJPEG2000 contributors• 21 countries / 80-100 meeting attendees
– EUROPE • Ericsson, Nokia, Philips, Canon, Motorola, IMEC, EPFL,
NTNU, Technical University of Denmark, VUB, Technical University of Berlin
– USA/Canada• Kodak, HP, Rockwell, Motorola, TI, Ricoh, Sharp, Adobe,
Sarnoff, SAIC, Teralogic, Univ. of Arizona, Univ. of Southern California, Univ. of Maryland, UBC, RPI
– ASIA/Australia• Samsung, Sony, Mitsubishi, CISRA, Univ. New South Wales,
Oki, Panasonic, ...
• 3-4 meetings per year
14Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
First steps of algorithm First steps of algorithm developmentdevelopment
• November 1997 (Sydney)– about 100 participants– 24 candidate algorithms– All of them intensively tested
• objective tests (quality metrics) ran on 22 test images at lossless and 6 different lossy bit rates (2, 1, 0.5, 0.25, 0.125, 0.0625 bpp)
• subjective tests by 40 evaluators at the 3 lowest bit rates
– selection WTCQ– VM established in March 98
JPEG2000
15Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 work planJPEG2000 work plan• Part I: A set of tools covering a good proportion
of application requirements (20-80 rules)
• Other parts are also defined and planned for a
further date
• Possible Amendment will be added to Part I
• Schedule for part I:Elevation to FDIS: 08/00Elevation to IS: 12/00
16Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 work planJPEG2000 work plan• Part II: Extension tools to cover specific
applications
• Part III: Motion JPEG2000
• Part IV: Conformance
• Part V: Reference software
• Part VI: Compound images file format
• Part VII: Technical Report
• Part VIII: ?
17Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
SStatus of tatus of existing existing implementationsimplementations
Software status – C implementation (SAIC / Univ. of Arizona / HP)
• JPEG2000 Verification Model used for the development of the standard
– JavaTM implementation (EPFL, Ericsson, Canon)• Reference implementation of JPEG2000 in part V and
publicly available
– C implementation (ImagePower / UBC) • Reference implementation of JPEG2000 in part V
18Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Features in Part JPEG2000 Features in Part II
• High compression efficiency • Lossless colour transformations • Lossy and lossless coding in one algorithm• Embedded lossy to lossless coding• Progressive by resolution, quality, position, …• Static and dynamic Region-of-Interest coding/decoding• Error resilience• Perceptual quality coding• Multiple component image coding• Tiling• Palletized image coding• Light file format (optional)• …
ERICSSON RESEARCH Media Lab
19Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
Some examplesSome examples
JPEG2000 JPEG2000 versus versus
JPEG baselineJPEG baseline
T
20Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEGJPEG at 0.125 bpp at 0.125 bpp
21Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000JPEG2000 at 0.125 bpp at 0.125 bpp
22Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEGJPEG at 0.25 bpp at 0.25 bpp
23Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000JPEG2000 at 0.25 bpp at 0.25 bpp
24Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEGJPEG at at 0.50.5 bpp bpp
25Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000JPEG2000 at at 0.50.5 bpp bpp
26Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG compound image 1.0 bpp
27Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 compound image 1.0 bpp
28Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Major Differences between Major Differences between JPEG and JPEG2000JPEG and JPEG2000
• New functionalities– ROI – Better error resiliency– More flexible progressive coding– ...
• Lossy to lossless in one system• Better compression at low bit-rates• Better at compound images and graphics
(palletized)
29Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 JPEG2000 and other standards and other standards
24
26
28
30
32
34
36
38
40
42
44
46
0 0.5 1 1.5 2
bpp
PSNR (dB)
J2K R J2K NR JPEG VTC
30Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Some lossless compression results
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
bike café cmpnd1 chart aerial2 target us average
com
pre
ssio
n r
ati
o
JPEG2000 JPEG-LS L-JPEG BZIP2
31Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Comparison of various algorithms Comparison of various algorithms from a functionality point of viewfrom a functionality point of view
JPEG 2000 JPEG-LS JPEG MPEG-4 VTC
lossless compression performance +++ ++++ + -lossy compression performance +++++ + +++ ++++progressive bitstreams ++++ - + ++Region of Interest (ROI) coding +++ - - +arbitrary shaped objects - - - ++random access ++ - - -low complexity ++ +++++ +++++ +error resilience +++ + + +++non-iterative rate control +++ - - +genericity +++ +++ ++ ++
32Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
More in depth comparisons More in depth comparisons between between JPEG2000JPEG2000 versus other versus other
standardsstandards• « JPEG 2000 still image coding versus other standards », D.
Santa-Cruz, T. Ebrahimi, J. Askelöf, M. Larsson and Ch. Christopoulos, in Proc. of SPIE, Vol. 4115
• « A study of JPEG 2000 still image coding versus other standards », D. Santa-Cruz, T. Ebrahimi, in Proc. of the X European Signal Processing Conference (EUSIPCO), Tampere, Finland, September 5-8, 2000
• « An analytical study of JPEG 2000 functionalities », D. Santa-Cruz, T. Ebrahimi, in Proc. of the IEEE International Conference on Image Processing (ICIP), Vancouver, Canada, September 10-13, 2000
33Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000JPEG2000
Algorithm descriptionAlgorithm description
C
34Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000JPEG2000: : Basic encoding Basic encoding schemescheme
Wavelettransform
Codeblockpartition
Quantization Entropycoding
Rateallocation
36Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Why block coding?Why block coding?• exploit local variations in the statistics of the
image from block to block
• provide support for applications requiring
random access to the image
• reduce memory consumption in hardware
implementations of the compression or
decompression engine
• Allow for parallel implementation
38Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
EBCOT: layered bitstream EBCOT: layered bitstream formationformation
• Each bitstream is organized as a succession of
layers
• Each layer contains additional contributions from
each block (some contributions might be empty)
• Block truncation points associated with each layer
are optimal in the rate distortion sense
• Rate distortion optimization can be performed but it
does not need to be standardized
39Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
EBCOT layered formationEBCOT layered formation
empty
emptyempty
empty
empty
empty
empty
empty
layer 5
layer 4
layer 3
layer 2
layer 1
block 1bit-stream
block 2bit-stream
block 3bit-stream
block 4bit-stream
block 5bit-stream
block 6bit-stream
block 7bit-stream
40Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Wavelet TransformWavelet Transform
• Two filters supported – W9x7 (Floating point)
for lossy coding – W5x3 (Integer) for
lossless coding• Only dyadic
decomposition supported
Dyadic decomposition
T
41Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
QuantizationQuantization• Explicit
– Define a specific quantization step for each subband– Smaller quantization steps for lower resolution subbands
• Implicit– Quantization steps derived from LL subband quantization steps– Smaller quantization steps for lower resolution subbands
• Reversible– No quantization but pure bit plane coding of transform coefficients
Possibility of visual weighting Fixed visual weighting Visual progressive coding (VIP)
42Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
LAZY CODING MODE
• Not all bitplanes need to be encoded by arithmetic coding
• Some bits are saved as raw bits
• This increases speed without sacrificing performance
43Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Lazy mode: Image “Gold” Gold: No lazy mode vs. lazy mode
0
5
10
15
20
25
30
35
40
45
50
0.0625 0.125 0.25 0.5 1.0 2.0
Bits per pixel [bpp]
PS
NR
[d
B]
No Lazy mode
Lazy mode
44Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
No lazy mode: 0.0625 bppNo lazy mode: 0.0625 bpp
45Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Lazy mode: 0.0625 bppLazy mode: 0.0625 bpp
46Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
No lazy mode: 0.25 bppNo lazy mode: 0.25 bpp
47Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
lazy mode: 0.25 bpplazy mode: 0.25 bpp
48Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Multi-component imageryMulti-component imagery
– up to 256 components– arbitrary dimensions/bit depths for
each component– reversible & non-reversible
component color transforms
49Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Reversible color Reversible color transformation:transformation:
making lossless color coding making lossless color coding possiblepossible
GBVr
GRUr
BGRYr
4
*2
GVrB
GUrR
VrUrYrG
)
4(
All components must have identical subsampling parameters and same depth before transformation
50Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Multiresolution decomposition
OriginalImage
51Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
LH1
HL1
HH1
LL1
Multiresolution decomposition
LL1
52Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Multiresolution decomposition
LL2
LH1
HL1
HH1
LH2 HH2
HL2LL2
53Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Multiresolution decomposition
LH1 HH1
LH2 HH2
HL2
HL3
HH3LH3
LL3
HL1
55Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
– Different modes are realized depending on the way information is written into the codestream
codestream
JPEG2000: JPEG2000: ScalabilityScalability
56Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Resolution
57Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Resolution
58Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Resolution
59Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Resolution
60Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Accuracy
61Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Accuracy
62Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Scalability - Progressive By Accuracy
63Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Example:Example:Progressive by resolutionProgressive by resolution
• Image: Woman• Resolution levels: 5• Decoded sizes: 1/16
1/81/41/21
64Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
65Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
66Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
67Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
68Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
69Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Example:Example:Progressive by qualityProgressive by quality
• Image: Woman• Bitrates: 0.125 bpp
0.25 bpp0.5 bpp1.0 bpp2.0 bpp
70Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.125 bpp
71Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.25 bpp
72Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.5 bpp
73Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
1.0 bpp
74Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
2.0 bpp
75Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Region Of Interest codingRegion Of Interest coding• Allows certain parts of an image to
be coded or decoded in better quality
• Static: The ROI is decided and coded once for all
en the encoder side
• Dynamic: The ROI can be decided and decoded on
the fly from a same bitstream
C
76Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ROI: Some visual resultsROI: Some visual results
No ROI
69:1 overall compression ratio
Rectangular ROI
77Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Regions Of Interest
78Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Regions Of Interest
79Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Regions Of Interest
80Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ROI coding: mask ROI coding: mask computationcomputation
81Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Region Of Interest codingRegion Of Interest coding• BASIC IDEA:
Calculate wavelet transform of whole image/time
calculate ROI mask == set of coefficients that are needed for up to lossless ROI coding
Encoding is progressive by accuracy and resolution
• NOTE: ROI mask need NOT be transmitted to decoder (location and shape of ROI needs however)
82Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Creation of ROI maskCreation of ROI mask
• The ROI masks are acquired by looking at the inverse transform
• For each pixel (X) that is in the ROI, the low and high frequency coefficients (L:s and H:s) that are needed to reconstruct the pixel, are included in the ROI mask
n-1 n n+1
Low High
n-1 n n+1
X:s
2n 2n+1
Inverse transform of the 5-3 filter
83Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ROI Scaling based methodROI Scaling based method
Coefficient values
ROI Coefficients
Highest BG coeff value is found
84Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ROI MaxShift methodROI MaxShift method
ROI Coefficients
Coefficient values
After shifting, all the ROI coefficients are larger than the largest BG coefficient
85Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Example:Example: ROI coding ROI coding
• Image: Woman• ROI: rectangular• Scaling value: 6• Progressive type: SNR• Bitrate: 4bpp
86Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.125 bpp
87Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.25 bpp
88Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
0.5 bpp
89Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
1.0 bpp
90Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
2.0 bpp
91Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
4.0 bpp
92Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Example:Example: ROI coding ROI coding
• Image: Woman• ROI: rectangular• Scaling value: MAXSHIFT• Progressive type: SNR• Bitrate: 4bpp
93Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
1.0 bpp
94Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
3.0 bpp
95Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
4.0 bpp
96Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ROI Maxshift mode: what ROI Maxshift mode: what is the gain?is the gain?
– Support for arbitrary shaped ROI’s with minimal complexity
– No need to send shape information– No need for shape encoder and decoder– No need for ROI mask at decoder side– Decoder as simple as non-ROI capable decoder– Can decide in which subband the ROI will begin
– therefore it can give similar results to the general scaling method
97Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab ROI coding: what do we ROI coding: what do we
pay?pay?Lossless image coding with Lossless image coding with
ROIROIGold: Rectangular ROI
0,99
0,995
1
1,005
1,01
1,015
1,02
No ROI
S=2
S=4
Maxshift
98Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab ROI coding: what do we ROI coding: what do we
pay?pay?Lossless image coding with Lossless image coding with ROIROI Target - approx. 25% circular ROI - Relative sizes
0,94
0,96
0,98
1
1,02
1,04
1,06
1,08
1,1
No ROI
S=2
S=4
Max shift
99Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Block transform codingBlock transform coding
• Tiling– Allow random access to portions of an
image
• Single-Sample Overlap Discrete Wavelet Transform (SSODWT) Exploit overlapping in order to reduce
blockiness In part II
T
100Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
TilingTiling (128x128, 0.25 bpp) (128x128, 0.25 bpp)
101Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
SSODWTSSODWT (128x128, 0.25 bpp) (128x128, 0.25 bpp)
102Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Error resilience Error resilience capabilitiescapabilities• Most still image coders use Entropy Coding
• Variable Length Coding is known to be prone to channel or transmission errors– Loss of synchronization
CHeader
Residual
DCHANNEL
Bit errors (Noise)Burst errors (Fading)
103Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
• Error resilience is achieved at two levels:– Entropy coding level
• Code-blocks
• Termination of arithmetic coding
• Reset of context
• Selective arithmetic coding bypass
– Packet level• Short packet format
• Resynchronization markers
Error resilienceError resilience
104Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Visual Frequency Visual Frequency WeightingWeighting• Allows system designers to take advantage of
visual perception• Utilize knowledge of the visual system’s
varying sensitivity to spatial frequencies as measured in the contrast sensitivity function (CSF)
• CSF is determined by the visual frequency of the transform coefficients; One CSF weight per subband
• Design of CSF weights is an encoder issue; depends on viewing condition of decoded image
105Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Visual Frequency Weighting Visual Frequency Weighting (cont.)(cont.)
Fixed Visual Weighting (FVW) &
Progressive Visual Coding (PVC)
• FVW: CSF are chosen according to the
final viewing condition
• PVC: Visual weights are changed during
the embedded process
106Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Line based transformsLine based transforms
• Most acquisition devices are serial in nature
• Most common scanning patterns work on a line-by-line basis
• Traditional wavelet transforms require whole image to be buffered and filtered
• Filtering along a line, requires one line
• Filtering along a column requires whole image
That is too complex!
107Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Line based transformsLine based transforms
• A way for low memory implementation
of the wavelet transform
Same wavelet coefficients as full frame
wavelet transform
• Same encoding results as in full frame
wavelet transform
108Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
File FormatFile Format• File Format extension .jp2• Possible to include XML data• Possible to include vendor specific
information• Possible to include IPR information• Possible to add URL to file format
– Can be used by an application to acquire more information about the associated vendor specific extensions
109Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part IJPEG2000 Part I
Core Coding System
• Schedule– March 2000, FCD – September 1, FDIS– December 2000, IS
• Only editorial changes allowed
• File extension, .jp2
C
110Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part IIJPEG2000 Part IIExtensions• Schedule
– March 2000, WD– September 2000, CD – December 2000, FCD– April 2001, FDIS– July 2001, IS
• File extension .jpx
111Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part IIIJPEG2000 Part IIIMotion-JPEG2000• Schedule
– March 2000, WD– December 2000, CD – March 2001, FCD– July 2001, FDIS– November 2001, IS
• Based on JPEG2000 Part I• No inter-frame dependencies
112Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part VJPEG2000 Part VReference Software• Schedule
– March 2000, ED– July 2000, CD – December 2000, FCD– April 2001, FDIS– July 2001, IS
• Software – JavaTM implementation (EPFL, Canon, Ericsson)– C implementation (UBC / ImagePower)
113Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part IVJPEG2000 Part IV
Compliance Tests
• Schedule– July 2000, WD– December 2000, CD – March 2001, FCD– July 2001, FDIS– November 2001, IS
114Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part VJPEG2000 Part V
Reference software
• Schedule– July 2000, CD– December 2000, FCD – March 2001, FDIS– July 2001, IS
115Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part VIJPEG2000 Part VI
Compound Image File FormatCompound Image File Format
• Schedule– August 2000, CD– December 2000, FCD – March 2001, FDIS– July 2001, IS
116Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
JPEG2000 Part VIIJPEG2000 Part VII
Technical reportTechnical report
• Schedule– December 2000, PDTR – March 2001, DTR– July 2001, TR
117Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
ConclusionsConclusions
• Advanced Still Image Coding System
• More complex than JPEG but it offers many
interesting functionalities
• No IPR associated to Part I of the standard (free
licensing)
• Intended to become the key standard for still
image coding in the next millennium
118Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
More informationMore information• JJ2000
– http://jj2000.epfl.ch
• JPEG Web site:– http://www.jpeg.org
• EUROSTILL– http://ltswww.epfl.ch/~eurostill
• SPEAR– http://spear.jpeg.org/
119Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
Contact usContact us for any further for any further informationinformation
• Touradj Ebrahimi– [email protected]
• Charilaos Christopoulos– [email protected]
120Signal Processing Laboratory Swiss Federal Institute of Technology, Lausanne
ERICSSON RESEARCH Media Lab
AcknowledgementsAcknowledgements• Mr. Joel Askelöf, Ericsson• Mr. Nicolas Aspert, EPFL• Dr. Eiji Atsumi, Mitsubishi, Japan• Mr. Martin Boliek, Ricoh• Dr. A. Chien, Eastman Kodak Company, USA• Dr. Troy Chinen, Fuji• Mr. Raphael Grosbois, EPFL• Prof. Faouzi Kossentini, UBC• Mr. Mathias Larsson, Ericsson• Dr. Daniel Lee, HP Labs• Dr. Eric Majani, CRF• Prof. Michael Marcellin, Univ. of Arizona• Prof. Andrew Perkis, NTNU• Dr. Majid Rabbani, Eastman Kodak Company• Mr. Diego Santa Cruz, EPFL• Prof. Athanasios Skodras, Univ. Of Padras• Dr. David Taubman, HP Labs & Univ. New South Wales• and many others ...
**
* In alphabetical order* In alphabetical order