47
Distortion estimators for bitplane image coding Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain Francesc Aulí-Llinàs

Distortion estimators for bitplane image coding

  • Upload
    hana

  • View
    32

  • Download
    0

Embed Size (px)

DESCRIPTION

Distortion estimators for bitplane image coding. Francesc Aulí -Llinàs. Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain. Distortion estimation Code-stream transcoding Interactive image and video transmission. Applications. - PowerPoint PPT Presentation

Citation preview

Diapositiva 1

Distortion estimatorsfor bitplane image codingDepartment of Information and Communications EngineeringUniversitat Autnoma de Barcelona, SpainFrancesc Aul-Llins

1TABLE OF CONTENTS3. EXPERIMENTS4. CONCLUSIONS2. PDF-BASED DISTORTION ESTIMATORS1. INTRODUCTION

Distortion estimation

Code-stream transcoding

Interactive image and video transmission

Applications

2Distortion estimation

Code-stream transcoding

Interactive image and video transmission

Applications

3. EXPERIMENTS4. CONCLUSIONS2. PDF-BASED DISTORTION ESTIMATORS1. INTRODUCTIONTABLE OF CONTENTS3INTRODUCTIONBITPLANE CODING STRATEGYoriginal coefficients0 0 01recovered coefficientssignificance coding0 0 0 01 1MOST SIGNIFICANT BITPLANE (MSB)mid-point reconstruction4INTRODUCTIONBITPLANE CODING STRATEGYsignificance coding0 0 0 011 MSB - 1original coefficientsrecovered coefficients5INTRODUCTIONBITPLANE CODING STRATEGY 0 0 01refinement codingMSB - 1original coefficientsrecovered coefficients6INTRODUCTIONBITPLANE CODING STRATEGYProgressive refinement of image distortionLossy-to-lossless codingWell-suited digital representation for computersMid-point reconstruction does NOT correspond with the nature of the signaloriginal coefficientsrecovered coefficients7INTRODUCTIONDISTRIBUTION OF WAVELET COEFFICIENTSwavelet coefficientsin a subbanddistribution8INTRODUCTIONDISTRIBUTION OF WAVELET COEFFICIENTS9INTRODUCTIONDISTRIBUTION OF WAVELET COEFFICIENTS0+-Common model:generalized Laplaciandistributionnumber of coefficients10INTRODUCTIONDISTRIBUTION OF WAVELET COEFFICIENTS0+-Common model:generalized Laplaciandistributionnumber of coefficients11INTRODUCTIONMID-POINT RECONSTRUCTION1) RECONSTRUCTION PROCEDURE CARRIED OUT IN THE DECODER2) DETERMINATION OF DISTORTION DECREASES IN THE ENCODER 3) DISTORTION ESTIMATORSFAST COMPUTATIONOF DISTORTION DECREASESTRANSCODING OPERATIONSDISTORTION ESTIMATION0 0 0 0 1 1 1D12INTRODUCTIONMID-POINT RECONSTRUCTION1) RECONSTRUCTION PROCEDURE CARRIED OUT IN THE DECODER2) DETERMINATION OF DISTORTION DECREASES IN THE ENCODER 3) DISTORTION ESTIMATORSFAST COMPUTATIONOF DISTORTION DECREASESTRANSCODING OPERATIONSDISTORTION ESTIMATIONRESEARCH PURPOSE:DETERMINATION OF DISTORTION ESTIMATORSTHAT BETTER CAPTURE THE SIGNALS NATUREApplication to the PCRD process of JPEG200013Distortion estimation

Code-stream transcoding

Interactive image and video transmission

Applications

3. EXPERIMENTS4. CONCLUSIONS2. PDF-BASED DISTORTION ESTIMATORS1. INTRODUCTIONTABLE OF CONTENTS14DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORS2P*2P*+1DsigP* = ( 2P* + P* 2P* )DrefP* = x -)2(x2 -[]p(x)2P*2P*+1dxP*= 0.5probability density function (pdf)coefficientsdistribution within a subband15DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORSto minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervaltransmitted bitplane P*P*mean of non-transmitted bitsP*= P* / 2P*DsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dx10100011001010110010101111011010101111101010100011001016DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORSto minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalDsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dx2P*2P*+1p(x)17DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORSaccumulated probability = 0.5to minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalDsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dx2P*2P*+1p(x)P*18DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORS2P*2P*+1to minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalp(x)DsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxP*19DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORS2P*2P*+1to minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalp(x)DsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxP*20DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORS2P*2P*+1to minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalp(x)DsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxP*21DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORS2P*2P*+1to minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalp(x)DsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxP*22P*2P*2P*+1p(x)DETERMINATION OF THE ESTIMATORSPDF-BASED DISTORTION ESTIMATORSto minimize the average distortion coefficients have to be reconstructed as the centroid of the quantization intervalDsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxCENTROIDS ARE NOT COMMONLY AVAILABLE IN PRACTICE23CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

dyadic wavelet decomposition24CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

HH1HL1LH1dyadic wavelet decomposition25CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

HH1HL1LH1HL2LH2HH2LL2

Portrait image - 5 wavelet levels (9/7 DWT)dyadic wavelet decomposition26CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

HH1HL1LH1HL2LH2HH2LL2

Portrait image - 5 wavelet levels (9/7 DWT)dyadic wavelet decomposition27CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

HH1HL1LH1HL2LH2HH2LL2

Portrait image - 5 wavelet levels (9/7 DWT)dyadic wavelet decomposition28CENTROIDS ESTIMATIONPDF-BASED DISTORTION ESTIMATORS

HH1HL1LH1HL2LH2HH2LL2

Portrait image - 5 wavelet levels (9/7 DWT)dyadic wavelet decompositionP,b =5+ 0.475log10 (Kb PKb )Practical advantage: implementations can use pre-computed lookup tables for and DrefP DsigP 29APPLICATION TO JPEG2000PDF-BASED DISTORTION ESTIMATORS

MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamcore coding system30

APPLICATION TO JPEG2000PDF-BASED DISTORTION ESTIMATORSMULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestream

core coding system31APPLICATION TO JPEG2000PDF-BASED DISTORTION ESTIMATORSMULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamcore coding system

32APPLICATION TO JPEG2000PDF-BASED DISTORTION ESTIMATORSMULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestream

core coding system33APPLICATION TO JPEG2000PDF-BASED DISTORTION ESTIMATORSMULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamcore coding system

34

MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamPDF-BASED DISTORTION ESTIMATORSAPPLICATION TO JPEG2000core coding system35

MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamPDF-BASED DISTORTION ESTIMATORSAPPLICATION TO JPEG2000core coding system36MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamPCRDPDF-BASED DISTORTION ESTIMATORSAPPLICATION TO JPEG2000core coding systemfinal codestream37MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamR1,D1R3,D3R4,D4R5,D5R2,D2RDconvex hullgeneralized Lagrange multiplier12345PCRDPDF-BASED DISTORTION ESTIMATORSAPPLICATION TO JPEG2000core coding system38MULTI-COMPONENTTRANSFORMWAVELETTRANSFORMQUANTIZATIONTIER-1 CODINGORIGINAL IMAGETIER-2 CODINGJP2 codestreamR1,D1R3,D3R4,D4R5,D5R2,D2RDconvex hullgeneralized Lagrange multiplier12345PCRDPDF-BASED DISTORTION ESTIMATORSAPPLICATION TO JPEG2000core coding systemD = DsigP #S + DrefP #R39Distortion estimation

Code-stream transcoding

Interactive image and video transmission

Applications

3. EXPERIMENTS4. CONCLUSIONS2. PDF-BASED DISTORTION ESTIMATORS1. INTRODUCTIONTABLE OF CONTENTS

40

EXPERIMENTSLOSSY MODEPortrait image5 DWT levels9/7 filter-taps64x64 codeblocks

CODERDECODERKDU0.5 ~ actual0.5=0.50.5 ~ Dsig Dref0.5actual actual ~ Dsig Drefactualapprox approx ~ Dsig Drefapprox

41EXPERIMENTSLOSSLESS MODEPortrait image5 IWT levels5/3 filter-taps64x64 codeblocks

CODERDECODERKDU0.5 ~ actual0.5=0.50.5 ~ Dsig Dref0.5actual actual ~ Dsig Drefactualapprox approx ~ Dsig Drefapprox

42

3. EXPERIMENTS4. CONCLUSIONS2. PDF-BASED DISTORTION ESTIMATORS1. INTRODUCTIONTABLE OF CONTENTSDistortion estimation

Code-stream transcoding

Interactive image and video transmission

Applications

43CONCLUSIONSMID-POINT RECONSTRUCTION DOES NOTCORRESPOND WITH THE NATURE OF THE SIGNALPenalization on the performance of distortion estimatorsRESEARCH PURPOSE:DETERMINATION OF DISTORTION ESTIMATORSTHAT BETTER CAPTURE THE SIGNALS NATURE44CONCLUSIONSDsigP* = ( 2P* + P* 2P* )x -)2(x2 -[]p(x)2P*2P*+1dxAPPLICATION TO THE PCRD PROCESS OF JPEG2000When both the coder and decoder use pdf-based reconstruction, gains of 1 dB are achieved in the lossless mode centroid2P*2P*+1P*45CONCLUSIONSFUTURE RESEARCHApplications

Distortion estimation

Codestream transcoding

Interactive image and video transmission

46EXPERIMENTS

LOSSLESS MODEPortrait image5 IWT levels5/3 filter-taps64x64 codeblocks

47