JPEG-2000: Background, Scope, And Technical Description Majid Rabbani Eastman Kodak Research Laboratories Rochester, NY 14650 rabbani@kodak.com 1 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Presentation Outline • Background – What is JPEG? – JPEG-2000 - A new compression paradigm – JPEG-2000 milestones • Technical description – Current DCT-based JPEG standard – Emerging wavelet-based JPEG-2000 – JPEG 2000 features • Future directions – JPEG 2000 timetable 2 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
What is JPEG? • The JPEG ( Joint Photographic Experts Group ) committee, formed in 1986, has been chartered with the – “ Digital compression and coding of continuous-tone still images ” • Joint between ISO and ITU-T • Has developed standards for the lossy, lossless, and nearly lossless of still images in the past decade • Website: www.jpeg.org 3 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG Summary • The JPEG committee has published the following standards: – ISO/IEC 10918-1 | ITU-T Rec. T.81 : Requirements and guidelines – ISO/IEC 10918-2 | ITU-T Rec. T.83 : Compliance testing – ISO/IEC 10918-3 | ITU-T Rec. T.84: Extensions – ISO/IEC 10918-4 | ITU-T Rec. T.86: Registration of JPEG Parameters, Profiles, Tags, Color Spaces, APPn Markers Compression Types, and Registration Authorities (REGAUT) – DIS 14495-1 | ITU-T Draft Rec. T.87 : Lossless and Near-Lossless Compression of Continuous-Tone Still Images - Baseline 4 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
ISO Organizations JTC 1 IEC ISO SC26 SC29 WG 1 WG 11 WG 12 JPEG/JBIG MPEG MHEG 5 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Old Compression Paradigm (JPEG Baseline) Encode Encoder choices No decoder choices color space only one image quantization post-processing entropy coder pre-processing 6 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
New Compression Paradigm Encode choices Contone or binary Tiling Decode choices Lossy/lossless Image resolution + Old paradigm choices Image fidelity Region-of-interest Fixed size Encode Components Lossless/lossy 7 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000 Objectives • Advanced standardized image coding system to serve applications into the next millenium • Address areas where current standards fail to produce the best quality or performance • Provide capabilities to markets that currently do not use compression • Provide an open system approach to imaging applications 8 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000: Requirements And Profiles • Internet applications (World Wide Web imagery) – Progressive in quality and resolution, fast decode • Mobile applications – Error resilience, low power, progressive decoding • Electronic commerce – Image security, digital watermarking • Digital photography – Low complexity, compression efficiency 9 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000: Requirements And Profiles • Hardcopy color facsimile, printing and scanning – Compression efficiency, strip or tile processing • Digital library/archive applications – Metadata, content management • Remote sensing – Multiple components, fast encoding, region of interest • Medical applications – Region of interest coding, lossy to lossless 10 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000 Features • Improved compression efficiency (estimated 30% depending on the image size and bit rate) • Lossy to lossless • Multiple resolution • Embedded bit stream (progressive decoding) • Region of interest coding (ROI) • Error resilience • Bit stream syntax (proposed by DIG 2000) 11 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000 Milestones • 7th WG1 Meeting - February 1996 – JPEG 2000 new work item proposed • 8th WG1 Meeting - June 1996 – JPEG 2000 call for proposal (N390) issued • 10th WG1 Meeting - March 1997 – JPEG 2000 call for contributions (N505) issued • 11th WG1 Meeting - July 1997 – JPEG 2000 Requirements and Objectives (N573) issued – Test plan for JPEG 2000 algorithm submissions (N557) 12 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000 Milestones • 12th WG1 Meeting - November 1997 – 104 delegates from 15 national bodies attended – 24 complete algorithm, 5 partial algorithm, and 7 architecture proposals presented – Ad hoc groups on “Requirements and Profiles”, “Core Experiments” and “Features and Functionality” were formed – First round of 37 core experiments were defined • 13 WG1 Meeting - March 1998 – 100 delegates from 15 national bodies attended – New ad hoc group on Verification Model (VM) was formed – Second round of 27 core experiments were defined 13 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG 2000 Milestones • 14th WG1 Meeting - July 1998 – 95 delegates from 14 national bodies attended – JPEG 2000 VM version 0.1 published – Third round of 42 core experiments were defined • 15th WG1 Meeting - November 1998 – 100 delegates from 13 national bodies attended – VM 2.1 was released – Fourth round of 28 core experiments were defined – DIG 2000 File Format proposal submitted 14 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Overview Of The Current DCT-Based JPEG 15 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Building Blocks Of A Compression Algorithm • In general, image compression schemes consist of: – Transformation or decomposition – Quantization – Symbol modeling and encoding Original Compressed Image Data Image Data Lossless Transformation Modeling & or Encoding Quantization Decomposition Lossy 16 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG DCT Encoder Block Diagram Header Compressed 8x8 Blocks Data FDCT Quantizer Huffman Encoder Quantization Huffman Tables Tables 17 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Example Block From Lena Image Following is an 8x8 block of the Lena image where each pixel value has been level-shifted by a value of 128 to place in the range (-128,127). 8 14 23 37 52 68 73 82 6 14 24 37 46 67 74 81 3 11 28 35 48 62 72 82 4 13 22 28 44 61 69 86 5 11 18 30 40 59 72 86 5 9 16 29 39 58 74 83 -1 8 16 31 38 59 75 80 2 11 18 30 37 57 69 82 18 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
DCT Of 8x8 Image Block The 8x8 discrete cosine transform (DCT) of the block packs its energy into a small number of coefficients. -215.8 16.1 -10.7 -3.7 -1.5 4.2 -6.7 327.5 18.1 3.4 -9.9 3.7 0.5 -3.2 3.5 2.2 2.5 1.3 -5.4 2.8 -1.0 2.3 -1.6 -2.6 0.6 -2.5 3.0 5.0 1.8 2.2 -2.6 -1.4 0.3 1.6 3.4 0.0 2.5 -5.1 1.6 -0.7 -0.6 -1.8 -2.4 0.5 -0.4 -1.6 -0.1 2.1 0.9 1.6 -0.6 -0.7 2.1 -0.5 0.9 2.8 0.6 -1.0 -2.9 -1.4 0.2 1.9 -0.6 0.7 19 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
DCT Basis Functions DCT coefficients can be viewed as weighting functions that, when applied to the 64 cosine basis functions of various spatial frequencies (8 x 8 templates), will reconstruct the original block. 20 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Quantization/Dequantization Procedure • Encoder quantization (example: input value = 18.1) • Scale (normalize) by the step size: 18.1/12=1.51 • Round to the nearest integer to get quantizer index = 2 • Decoder dequantization (example: index=2, step size=12) • Multiply quantizer index by the step size: 2 x 12 = 24 12 -24 -12 +12 +24 0 -30 -18 -6 +6 +18 +30 -2 -1 0 +1 +2 21 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Quantized DCT Coefficients For typical blocks in an image, the process of normalization followed by quantization results in many zero-valued coefficients that can be coded efficiently. 0 0 0 0 20 -20 2 -1 0 0 0 0 0 0 2 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 22 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Variable-Length Codes Symbol Probability Code I Code II A 0.60 00 0 B 0.30 01 10 C 0.05 10 110 D 0.05 11 111 • Average length of Code I = 2.0 bits/symbol • Average length of Code II = 1.5 bits/symbol • Code I is a fixed-length code, while code II is a variable- length code (VLC). An example of VLC is Huffman coding. 23 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
JPEG DCT Decoder Block Diagram Header Reconstructed Image Data Compressed Data Huffman Decoder Dequantizer IDCT Quantization Huffman Tables Tables 24 JPEG 2000 - Majid Rabbani, Eastman Kodak Company - December 1998
Recommend
More recommend
