Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain
description
Transcript of Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain
![Page 1: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/1.jpg)
Probability model for multi-component images through Prior-component Lookup Tables
Department of Information and Communications EngineeringUniversitat Autònoma de Barcelona, Spain
Francesc Aulí-Llinàs
![Page 2: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/2.jpg)
TABLE OF CONTENTS
EXPERIMENTAL RESULTS
INTRODUCTION
PCLUT METHOD
CONCLUSIONSPCLUT ADVANTAGES
• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 3: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/3.jpg)
TABLE OF CONTENTS
INTRODUCTION
PCLUT METHOD
EXPERIMENTAL RESULTS
CONCLUSIONSPCLUT ADVANTAGES
• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 4: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/4.jpg)
INTRODUCTIONHYPERSPECTRAL IMAGES
![Page 5: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/5.jpg)
INTRODUCTIONTYPICAL CODING SCHEME
![Page 6: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/6.jpg)
INTRODUCTIONTYPICAL CODING SCHEME
![Page 7: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/7.jpg)
INTRODUCTIONTYPICAL CODING SCHEME
![Page 8: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/8.jpg)
INTRODUCTIONTYPICAL CODING SCHEME
![Page 9: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/9.jpg)
INTRODUCTIONTYPICAL CODING SCHEME
![Page 10: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/10.jpg)
INTRODUCTION
ARITHMETICCODER CODESTREAM
TYPICAL CODING SCHEME
![Page 11: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/11.jpg)
INTRODUCTION
ARITHMETICCODER CODESTREAM
TYPICAL CODING SCHEME
DISADVANTAGES• High memory requirements• High computational costs
• Poor component scalability
RESEARCH PURPOSELossy coding scheme for hyper-spectral images
with low memory requirements, low computational costs, and high component scalability
![Page 12: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/12.jpg)
TABLE OF CONTENTS
INTRODUCTION
PCLUT METHOD
EXPERIMENTAL RESULTS
CONCLUSIONSPCLUT ADVANTAGES
• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 13: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/13.jpg)
PCLUT METHODMAIN INSIGHTS
1) Wavelet transform only on the spatial dimensions
reduces computational costs reduces memory requirements increases component scalability reduces coding performance
![Page 14: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/14.jpg)
PCLUT METHODMAIN INSIGHTS
1) Wavelet transform only on the spatial dimensions
reduces computational costs reduces memory requirements increases component scalability reduces coding performance
![Page 15: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/15.jpg)
PCLUT METHODMAIN INSIGHTS
1) Wavelet transform only on the spatial dimensions
reduces computational costs reduces memory requirements increases component scalability reduces coding performance
2) Enhanced probability model for emitted symbols
reduces computational costs increases coding performance
![Page 16: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/16.jpg)
PCLUT METHODPRIOR COEFFICIENT LOOKUP TABLES METHOD
ARITHMETICCODER CODESTREAM
LUT
![Page 17: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/17.jpg)
PCLUT METHODPRIOR COEFFICIENT LOOKUP TABLES METHOD
SIGNIFICANCE CODING REFINEMENT CODING
![Page 18: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/18.jpg)
PCLUT METHODPRIOR COEFFICIENT LOOKUP TABLES METHOD
SIGNIFICANCE CODING REFINEMENT CODING
![Page 19: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/19.jpg)
TABLE OF CONTENTS
INTRODUCTION
PCLUT METHOD
EXPERIMENTAL RESULTS
CONCLUSIONSPCLUT ADVANTAGES
• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 20: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/20.jpg)
EXPERIMENTAL RESULTSCODING PERFORMANCE 3 AVIRIS images
512x512x224 (16bps)
1 used to generate LUTs
3 Hyperion images768x256x242 (12bps)
1 used to generate LUTs
AVIRIS – yellowstone sc01 Hyperion – flooding
![Page 21: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/21.jpg)
EXPERIMENTAL RESULTSCOMPUTATIONAL COSTS
THROUGHPUT (in seconds)
MEMORY REQUIREMENTS (in MB)
3 AVIRIS images512x512x224 (16bps)
1 used to generate LUTs
3 Hyperion images768x256x242 (12bps)
1 used to generate LUTs
![Page 22: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/22.jpg)
TABLE OF CONTENTS
INTRODUCTION
PCLUT METHOD
EXPERIMENTAL RESULTS
CONCLUSIONSPCLUT ADVANTAGES
• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 23: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/23.jpg)
CONCLUSIONS
ARITHMETICCODER CODESTREAM
TYPICAL CODING SCHEME PRIOR COEFFICIENT LOOKUP TABLES
ARITHMETICCODER CODESTREAM
LUT
PCLUT ADVANTAGES• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis
![Page 24: Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain](https://reader035.fdocuments.in/reader035/viewer/2022062305/56816515550346895dd79443/html5/thumbnails/24.jpg)
CONCLUSIONS
ARITHMETICCODER CODESTREAM
TYPICAL CODING SCHEME PRIOR COEFFICIENT LOOKUP TABLES
ARITHMETICCODER CODESTREAM
LUT
PCLUT ADVANTAGES• Memory requirements: only 2 components in memory• Computational costs: reduction in 1/3 (mainly due to the probability model)• Coding performance: equivalent to 1 level of Haar transform along the depth axis• Component scalability: equivalent to 1 level of Haar transform along the depth axis