Department of Information and Communications Engineering Universitat Autònoma de Barcelona, Spain

1. Probability model for multi-component images through Prior-component Lookup Tables Francesc Aulí-Llinàs Department of Information and Communications Engineering UniversitatAutònoma de Barcelona, Spain

2. TABLE OF CONTENTS INTRODUCTION PCLUT METHOD EXPERIMENTAL RESULTS CONCLUSIONS • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis

3. TABLE OF CONTENTS INTRODUCTION PCLUT METHOD EXPERIMENTAL RESULTS CONCLUSIONS • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis

4. INTRODUCTION HYPERSPECTRAL IMAGES

5. INTRODUCTION TYPICAL CODING SCHEME

6. INTRODUCTION TYPICAL CODING SCHEME

7. INTRODUCTION TYPICAL CODING SCHEME

8. INTRODUCTION TYPICAL CODING SCHEME

9. INTRODUCTION TYPICAL CODING SCHEME

10. INTRODUCTION TYPICAL CODING SCHEME ARITHMETIC CODER CODESTREAM

11. INTRODUCTION TYPICAL CODING SCHEME • DISADVANTAGES • Highmemoryrequirements • Highcomputationalcosts • Poorcomponentscalability RESEARCH PURPOSE Lossycodingschemeforhyper-spectralimages withlowmemoryrequirements, lowcomputational costs, and highcomponentscalability ARITHMETIC CODER CODESTREAM

12. TABLE OF CONTENTS INTRODUCTION PCLUT METHOD EXPERIMENTAL RESULTS CONCLUSIONS • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis

13. PCLUT METHOD MAIN INSIGHTS 1) Wavelet transformonlyonthe spatialdimensions reduces computationalcosts reduces memoryrequirements increasescomponentscalability reduces coding performance

14. PCLUT METHOD MAIN INSIGHTS 1) Wavelet transformonlyonthe spatialdimensions reduces computationalcosts reduces memoryrequirements increasescomponentscalability reduces coding performance

15. PCLUT METHOD MAIN INSIGHTS 1) Wavelet transformonlyonthe spatialdimensions reduces computationalcosts reduces memoryrequirements increasescomponentscalability reduces coding performance 2) Enhancedprobabilitymodel foremitted symbols reduces computationalcosts increasescoding performance

16. PCLUT METHOD PRIOR COEFFICIENT LOOKUP TABLES METHOD LUT ARITHMETIC CODER CODESTREAM

17. PCLUT METHOD PRIOR COEFFICIENT LOOKUP TABLES METHOD REFINEMENT CODING SIGNIFICANCE CODING

18. PCLUT METHOD PRIOR COEFFICIENT LOOKUP TABLES METHOD REFINEMENT CODING SIGNIFICANCE CODING

19. TABLE OF CONTENTS INTRODUCTION PCLUT METHOD EXPERIMENTAL RESULTS CONCLUSIONS • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis

20. EXPERIMENTAL RESULTS CODING PERFORMANCE 3 AVIRIS images 512x512x224 (16bps) 1 usedto generateLUTs 3 Hyperionimages 768x256x242 (12bps) 1 usedto generateLUTs AVIRIS – yellowstone sc01 Hyperion – flooding

21. EXPERIMENTAL RESULTS COMPUTATIONAL COSTS 3 AVIRIS images 512x512x224 (16bps) 1 usedto generateLUTs 3 Hyperionimages 768x256x242 (12bps) 1 usedto generateLUTs THROUGHPUT (in seconds) MEMORY REQUIREMENTS (in MB)

22. TABLE OF CONTENTS INTRODUCTION PCLUT METHOD EXPERIMENTAL RESULTS CONCLUSIONS • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis

23. CONCLUSIONS PRIOR COEFFICIENT LOOKUP TABLES TYPICAL CODING SCHEME LUT • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis ARITHMETIC CODER ARITHMETIC CODER CODESTREAM CODESTREAM

24. CONCLUSIONS PRIOR COEFFICIENT LOOKUP TABLES TYPICAL CODING SCHEME LUT • PCLUT ADVANTAGES • Memoryrequirements: only 2 components in memory • Computationalcosts: reduction in 1/3 (mainlyduetotheprobabilitymodel) • Coding performance: equivalentto 1 level of Haartransformalongthedepth axis • Componentscalability: equivalentto 1 level of Haartransformalongthedepth axis ARITHMETIC CODER ARITHMETIC CODER CODESTREAM CODESTREAM