Slide 1

LOW COMPLEXITY EMBEDDED QUANTIZATION SCHEME COMPATIBLE WITH BITPLANE IMAGE CODING Department of Information and Communications Engineering Universitat Autnoma de Barcelona, Spain Francesc Aul-Llins Slide 2 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW TABLE OF CONTENTS Slide 3 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW TABLE OF CONTENTS Slide 4 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION Progressive transmission Interactive applications Codestream truncation Image transcoding compressed codestream QUALITY PROGRESSIVITY 1 Slide 5 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION CLASSIC SCHEME: USDQ+BPC 0 W = 2 4 10 1 Slide 6 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION CLASSIC SCHEME: USDQ+BPC 0 W = 2 4 2323 10 (10 XXXX (2 1XXX (2 10XX (2 101X (2 1010 (22 2121 2020 emit 0 emit 1 0 1 0 1 0 1 0 1 0 1 0 1 = 10 (10 2 Slide 7 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION CLASSIC SCHEME: USDQ+BPC 0 W = 2 4 2323 10 (10 XXXX (2 1XXX (2 10XX (2 101X (2 1010 (2 2 2121 2020 = 10 (10 2 Slide 8 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION CLASSIC SCHEME: USDQ+BPC 0 W = 2 4 2323 10 (10 XXXX (2 1XXX (2 10XX (2 101X (2 1010 (2 2 2121 2020 density IS USDQ+BPC OPTIMAL FOR WAVELET-BASED LOSSY IMAGE CODING? = 10 (10 2 Slide 9 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION GENERAL EMBEDDED QUANTIZATION (GEQ) 0 W = 2 4 emit 0 emit 1 0 1 emit 0 emit 1 0 1 0 1 0 1 1 > T 1 ? yes no 0 > T 4 ? yes no 10 (10 0 1 0 > T 6 ? yes no T 6 T 6 T 6 T 6 T 6 T6T6T6T6 T5T5 T4T4 T 4 T 4 T3T3 T2T2 T1T1 3 Slide 10 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS INTRODUCTION GENERAL EMBEDDED QUANTIZATION (GEQ) 10 (10 100 10 (10 USDQ+BPC is optimal in terms of coding performance GEQ schemes can achieve same coding performance as that of USDQ+BPC employing fewer quantization stages GEQ schemes can help to reduce the computational costs of the codec in 20% GEQ is not compatible with bitplane coding strategies 1 > T 1 ? yes no 0 > T 4 ? yes no 0 > T 6 ? yes no 0 W = 2 4 T 6 T 6 T 6 T 6 T 6 T6T6T6T6 T5T5 T4T4 T 4 T 4 T3T3 T2T2 T1T1 RESEARCH PURPOSE: ADAPT THE LOW-COMPLEXITY GEQ SCHEME TO BITPLANE CODING 3 Slide 11 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW TABLE OF CONTENTS 1 Slide 12 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS GEQ 2SDQ L H same number of subintervals USDQ+BPC 0 W WW 1) H = L (1 - ) 2) 3) Each quantization stage halves the previous subintervals except in the first stage CONDITIONS PROPOSED SCHEME 2-STEP SCALAR DEADZONE QUANTIZATION 4 Slide 13 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS 0 W = 2 4 emit 0 emit 1 0 1 0 1 0 1 0 1 WW PROPOSED SCHEME 0 1 0 1 L H XXXX 1XXX 11XX 110X 1100 10 (10 2SDQ(1010 (2 )=1100 2-STEP SCALAR DEADZONE QUANTIZATION 5 Slide 14 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 15 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 16 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 17 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 18 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 19 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 20 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 21 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 22 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION 6 Slide 23 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION 7 Slide 24 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 0 2M2M 2M2M 2 M-1 2-STEP SCALAR DEADZONE QUANTIZATION 7 Slide 25 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 0 2M2M 2 M-1 2M2M L H 2SDQ header bit 2-STEP SCALAR DEADZONE QUANTIZATION 7 Slide 26 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 0 2M2M 2 M-1 2M2M L H variable H = 4(2ln2 + 1 ln2) RD-opt 2-STEP SCALAR DEADZONE QUANTIZATION 2SDQ header bit constant L = 4 2 M-1 2 M-2 0 7 Slide 27 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION RD-opt 2SDQ header bit 8 Slide 28 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS IMPLEMENTATION IN JPEG2000 ORIGINAL IMAGE MULTI-COMPONENT TRANSFORM WAVELET TRANSFORM QUANTIZATION TIER-1 CODING TIER-2 CODING JP2 CODESTREAM 2-STEP SCALAR DEADZONE QUANTIZATION RD-opt 2SDQ header bit 8 Slide 29 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW TABLE OF CONTENTS 1 Slide 30 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 31 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 32 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 33 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 34 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 35 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Portrait image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 9 Slide 36 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS EXPERIMENTAL RESULTS Cafeteria image (ISO 12640-1 corpus) Codeblocks of 64x64 2SDQ is applied on codeblocks with 6 bitplanes or more removing 1 bitplane 10 Slide 37 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW TABLE OF CONTENTS 1 Slide 38 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS L H WW explore new quantization schemes for wavelet-based image coding compatible with bitplane coding quantization scheme with 2 step sizes adapted to the density of wavelet coefficients replacement of USDQ quantization indices by 2SDQ indices introduction of three easy-to-implement steps in the coding pipeline reduction of coding passes without penalizing coding performance Motivation 2SDQ Implementation Adaptation in JPEG2000 Results 11 Slide 39 INTRODUCTION/ 11 QUANTIZATION SCHEMEEXPERIMENTAL RESULTS CONCLUSIONS 11 CONCLUSIONS L H WW explore new quantization schemes for wavelet-based image coding compatible with bitplane coding quantization scheme with 2 step sizes adapted to the density of wavelet coefficients replacement of USDQ quantization indices by 2SDQ indices introduction of three easy-to-implement steps in the coding pipeline reduction of coding passes without penalizing coding performance Motivation 2SDQ Implementation Adaptation in JPEG2000 Results