Remote Sensing Data Compression Joan Serra-Sagrist` a Universitat - - PowerPoint PPT Presentation

▶
remote sensing data compression
SMART_READER_LITE
LIVE PREVIEW

Remote Sensing Data Compression Joan Serra-Sagrist` a Universitat - - PowerPoint PPT Presentation

Jan 20, 2023 340 likes •1.13k views

Title Detour Motivation Standards Proposals Summary Remote Sensing Data Compression Joan Serra-Sagrist` a Universitat Aut` onoma de Barcelona Joan.Serra@uab.cat Inpainting-Based Image Compression Schloss Dagstuhl, Leibniz-Zentrum f

slide-1
SLIDE 1

Title Detour Motivation Standards Proposals Summary

Remote Sensing Data Compression

Joan Serra-Sagrist` a

Universitat Aut`

  • noma de Barcelona

Joan.Serra@uab.cat

Inpainting-Based Image Compression Schloss Dagstuhl, Leibniz-Zentrum f¨ ur Informatik, Deutschland November 17, 2016

This work was supported in part by the Spanish Ministry of Economy and Competitiveness (MINECO) and the European Regional Development Fund (FEDER) under Grant TIN2015-71126-R, by the Catalan Government under Grant 2014SGR-691, and by the French Space Agency (CNES) under several contracts. 1 / 77

slide-2
SLIDE 2

Title Detour Motivation Standards Proposals Summary

Motivation for Still Image & Video Data Compression

2 / 77

slide-3
SLIDE 3

Title Detour Motivation Standards Proposals Summary

Motivation for Still Image & Video Data Compression

KPCB INTERNET TRENDS 2016 | PAGE 90

500 1,000 1,500 2,000 2,500 3,000 3,500 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 # of Photos Shared per Day (MM) Snapchat Facebook Messenger Instagram WhatsApp Facebook

Daily Number of Photos Shared on Select Platforms, Global, 2005 – 2015

(2013 onward only) (2015 only)

Image Growth Remains Strong

Facebook-

  • wned

properties

Source: Snapchat, Company disclosed information, KPCB estimates Note: Snapchat data includes images and video. Snapchat stories are a compilation of images and video. WhatsApp data estimated based on average of photos shared disclosed in Q1:15 and Q1:16. Instagram data per Instagram press release. Messenger data per Facebook (~9.5B photos per month). Facebook shares ~2B photos per day across Facebook, Instagram, Messenger, and WhatsApp (2015).

3 / 77

slide-4
SLIDE 4

Title Detour Motivation Standards Proposals Summary

Motivation for Still Image & Video Data Compression

KPCB INTERNET TRENDS 2016 | PAGE 78

Source: Facebook, Snapchat. Q2:15 Facebook video views data based on KPCB estimate. Facebook video views represent any video shown onscreen for >3 seconds (including autoplay). Snapchat video views counted instantaneously on load.

2 4 6 8 10 Q4:14 Q1:15 Q2:15 Q3:15 Q4:15 Q1:16 Video Views per Day (B) 2 4 6 8 10 Q3:14 Q4:14 Q1:15 Q2:15* Q3:15 Video Views per Day (B)

Facebook Daily Video Views, Global, Q3:14 – Q3:15 Snapchat Daily Video Views, Global, Q4:14 – Q1:16

User-Shared Video Views on Snapchat & Facebook = Growing Fast

4 / 77

slide-5
SLIDE 5

Title Detour Motivation Standards Proposals Summary

Motivation for Still Image & Video Data Compression

KPCB INTERNET TRENDS 2016 | PAGE 74

Source: “Engaging and Cultivating Millennials and Gen Z,” Denison University and Ologie, 12/14. Note: Gen Z defined in this report as those born after 1995. In 2016 they are ages 1-20. Note that there may be different opinions on which years each generation begins and ends.

Generation Z (Ages 1-20) = Communicates with Images

Gen Z Tech Innate: 5 screens at once Communicate with images Creators and Collaborators Future-focused Realists Want to work for success vs Attributes – Millennials vs. Gen Z Millennials Tech Savvy: 2 screens at once Communicate with text Curators and Sharers Now-focused Optimists Want to be discovered 5 / 77

slide-6
SLIDE 6

Title Detour Motivation Standards Proposals Summary

LandSat 7, 2001 Orbit Height 705 km Operator US Geographical Survey Orbit Type sun-synchronous Repeat Cycle 16 days

6 / 77

slide-7
SLIDE 7

Title Detour Motivation Standards Proposals Summary

LandSat 8, 14 September 2015 Operational Land Imager: 8 multispectral bands, 30 m spatial 1 panchromatic band, 15 m spatial Thermal Infrared Sensor: 2 bands, 100 m spatial

7 / 77

slide-8
SLIDE 8

Title Detour Motivation Standards Proposals Summary

LandSat 8, 9 October 2016 Radiometric resolution: 12-bit dynamic range Delivered products: 16-bit images File size: 2 GB (uncompressed) / 1 GB (compressed)

8 / 77

slide-9
SLIDE 9

Title Detour Motivation Standards Proposals Summary

International Space Station, 2004 Orbit Height 400 km Globe Circle 90 minutes Speed 28,000 km/h or 7.71 km/s

9 / 77

slide-10
SLIDE 10

Title Detour Motivation Standards Proposals Summary

LandSat 5, 1984 – 2007

10 / 77

slide-11
SLIDE 11

Title Detour Motivation Standards Proposals Summary

1 August 2016 8 August 2016 Copernicus Sentinel 2 Orbit Height 786 km Orbit Type sun-synchronous Revisit Time 5 days

11 / 77

slide-12
SLIDE 12

Title Detour Motivation Standards Proposals Summary

Remote Sensing

12 / 77

slide-13
SLIDE 13

Title Detour Motivation Standards Proposals Summary

Remote Sensing

13 / 77

slide-14
SLIDE 14

Title Detour Motivation Standards Proposals Summary

Remote Sensing

14 / 77

slide-15
SLIDE 15

Title Detour Motivation Standards Proposals Summary

Remote Sensing Applications Based on Satellite Open Data (Landsat8 and Sentinel-2), Nuno Duro Santos, Gil Gon¸ calves Conferˆ encia Nacional de Geodecis˜ ao, Barreiro, 15 e 16 de maio de 2014

15 / 77

slide-16
SLIDE 16

Title Detour Motivation Standards Proposals Summary 16 / 77

slide-17
SLIDE 17

Title Detour Motivation Standards Proposals Summary

  • 1. October 1957: Sputnik 1, first artificial satellite. Russia.
  • 2. November 2016: 4,256 satellites in orbit.
  • 3. November 2016: 1,419 operational satellites
  • ≈ 400 satellites in Low-Earth Orbit.

Just a few hundred kilometers above the surface. Earth Observation.

  • ≈ 100 satellites in Medium-Earth Orbit.

Around 20,000 kilometers up. Global Positioning. Navigation.

  • ≈ 750 satellites in Geostationary Orbit.

Altitude of almost 36,000 kilometers. Communications.

17 / 77

slide-18
SLIDE 18

Title Detour Motivation Standards Proposals Summary 18 / 77

slide-19
SLIDE 19

Title Detour Motivation Standards Proposals Summary 19 / 77

slide-20
SLIDE 20

Title Detour Motivation Standards Proposals Summary 20 / 77

slide-21
SLIDE 21

Title Detour Motivation Standards Proposals Summary

EO Remote Sensing sensors capture hyperspectral images

Source: ESA/Institute for Environment Solutions Source: Softpedia

21 / 77

slide-22
SLIDE 22

Title Detour Motivation Standards Proposals Summary

On-board Limitation: Restricted Storage Capacity

22 / 77

slide-23
SLIDE 23

Title Detour Motivation Standards Proposals Summary

Chronological Evolution of EO Satellites and Sensors

23 / 77

slide-24
SLIDE 24

Title Detour Motivation Standards Proposals Summary

On-board Limitation: Downlink Channel Capacity

24 / 77

slide-25
SLIDE 25

Title Detour Motivation Standards Proposals Summary

On-board Limitation: Storage and Transmission

25 / 77

slide-26
SLIDE 26

Title Detour Motivation Standards Proposals Summary

Overview

Motivation for Still Image & Video Data Compression Motivation for Remote Sensing Data Compression Standards for Remote Sensing Data Compression CCSDS-122.0 CCSDS-123.0 Some of our recent proposals Temporal Exploitation Regression Wavelet Analysis Inpainting-based

26 / 77

slide-27
SLIDE 27

Title Detour Motivation Standards Proposals Summary

Types of Compression: Lossless

27 / 77

slide-28
SLIDE 28

Title Detour Motivation Standards Proposals Summary

Types of Compression: Lossy

28 / 77

slide-29
SLIDE 29

Title Detour Motivation Standards Proposals Summary

Consultative Committee for Space Data Systems

  • Founded in 1982.
  • Major space agencies in the world.
  • 11 Member Agencies
  • 30 Observer Agencies
  • 14 Liaison Organizations
  • 99 Associates
  • Neither Portugal nor Spain are CCSDS Participants
  • More than 800 space missions.

29 / 77

slide-30
SLIDE 30

Title Detour Motivation Standards Proposals Summary 30 / 77

slide-31
SLIDE 31

Title Detour Motivation Standards Proposals Summary

Multispectral Hyperspectral Data Compression Working Group

Lossless Lossy Near-Lossless Mono- 1997 2005 component CCSDS-121.0 CCSDS-122.0 Multi- 2012 2017 2019 component CCSDS-123.0 CCSDS-122.1 CCSDS-123.1

31 / 77

slide-32
SLIDE 32

Title Detour Motivation Standards Proposals Summary

Multispectral Hyperspectral Data Compression Working Group

Lossless Lossy Near-Lossless Mono- 1997 2005 component CCSDS-121.0 CCSDS-122.0 → Prediction → Transform → Golomb-Rice → Variable-to-Fix Multi- 2012 2017 2019 component CCSDS-123.0 CCSDS-122.1 CCSDS-123.1 → Prediction → Transform → Prediction → Golomb-Rice → Variable-to-Fix → Interleaved

32 / 77

slide-33
SLIDE 33

Title Detour Motivation Standards Proposals Summary

  • 2011. We were invited to participate in MHDC-WG.

Lossless Lossy Near-Lossless Mono- CCSDS-121.0 CCSDS-122.0 component TER Software Multi- CCSDS-123.0 CCSDS-122.1 CCSDS-123.1 component EMPORDA Software POT technical contribution Software

  • Our TER and EMPORDA softwares are one of the only two
  • pen-source implementations available worldwide.
  • Our proposal Pairwise Orthogonal Transform is at the basis of

CCSDS-122.1.

33 / 77

slide-34
SLIDE 34

Title Detour Motivation Standards Proposals Summary

CCSDS-122.0. Mono-component progressive lossy-to-lossless

34 / 77

slide-35
SLIDE 35

Title Detour Motivation Standards Proposals Summary

CCSDS-122.0. Mono-component progressive lossy-to-lossless

35 / 77

slide-36
SLIDE 36

Title Detour Motivation Standards Proposals Summary

CCSDS-122.0. Mono-component progressive lossy-to-lossless

36 / 77

slide-37
SLIDE 37

Title Detour Motivation Standards Proposals Summary

CCSDS-123.0. Multi-component lossless

Predictor Entropy Coder input image mapped prediction residuals compressed image

37 / 77

slide-38
SLIDE 38

Title Detour Motivation Standards Proposals Summary

CCSDS-123.0. Multi-component lossless

Predictor Entropy Coder

Predict

4x

  • +

+ +

( )

local sum

portion of band z:

A B C D E

A B C D E

38 / 77

slide-39
SLIDE 39

Title Detour Motivation Standards Proposals Summary

CCSDS-123.0. Multi-component lossless

Predict

4x

  • +

+ +

( ) A B C D E

A B C D E with

(for band z)

4x

  • +

+ +

( )

A B C D E

  • 4x

+ + +

( )

A B C D E

4x

  • +

+ +

( )

A B C D E

[...]

(of band z-1) (of band z-2) (of band z-P)

parameter P

4x

  • +

+ +

( )

A B C D

  • 4x

+ + +

( )

A B C D

4x

  • +

+ +

( )

A B C D

(of band z) (of band z) (of band z)

B D A

39 / 77

slide-40
SLIDE 40

Title Detour Motivation Standards Proposals Summary

Overview

Motivation for Still Image & Video Data Compression Motivation for Remote Sensing Data Compression Standards for Remote Sensing Data Compression CCSDS-122.0 CCSDS-123.0 Some of our recent proposals Temporal Exploitation Regression Wavelet Analysis Inpainting-based

40 / 77

slide-41
SLIDE 41

Title Detour Motivation Standards Proposals Summary

Temporal Exploitation

  • 1. F. Auli-Llinas, M. Marcellin, V. Sanchez, J. Serra-Sagrist`

a, J. Bartrina-Rapesta, I. Blanes, ”Coding Scheme for the Transmission of Satellite Imagery,” in Proc. IEEE Data Compression Conference (DCC), Apr. 2016.

41 / 77

slide-42
SLIDE 42

Title Detour Motivation Standards Proposals Summary 42 / 77

slide-43
SLIDE 43

Title Detour Motivation Standards Proposals Summary 43 / 77

slide-44
SLIDE 44

Title Detour Motivation Standards Proposals Summary 44 / 77

slide-45
SLIDE 45

Title Detour Motivation Standards Proposals Summary

Regression Wavelet Analysis

  • 1. N. Amrani, J. Serra-Sagrist`

a, V. Laparra, M.W. Marcellin and

  • J. Malo, ”Regression Wavelet Analysis for Lossless Coding of

Remote-Sensing Data,” IEEE Trans. Geoscience and Remote Sensing, vol. 54, no. 9, pp. 5616-5627, September 2016. Digital Object Identifier 10.1109/TGRS.2016.2569485.

  • 2. N. Amrani, J. Serra-Sagrist`

a, M. Hernandez-Cabronero, and

  • M. Marcellin, ”Regression Wavelet Analysis for

Progressive-Lossy-to-Lossless Coding of Remote-Sensing Data,” in Proc. IEEE Data Compression Conference (DCC),

  • Apr. 2016.

45 / 77

slide-46
SLIDE 46

Title Detour Motivation Standards Proposals Summary

General Lossy Compression Scheme

Decorrelation Decorrelation Quantization Entropy Coder Rate Allocation

x d q c y

1-D Spectral Transform 2-D Spatial Transform

x s d      KLT DWT RWA

46 / 77

slide-47
SLIDE 47

Title Detour Motivation Standards Proposals Summary

1-D Spectral Transforms

Original DWT KLT RWA

1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224

Decorrelation

Suboptimal Optimal

(Gaussian)

Competitive

Component-scalability

  • ✗
  • Avoid side information
  • ✗

/✗

Data independent

  • ✗

/✗

Complexity

Low Very High Low

47 / 77

slide-48
SLIDE 48

Title Detour Motivation Standards Proposals Summary

RWA Algorithm

1 Discrete Wavelet Transform (DWT) V0

G ↓ 2 H

Scale 1

↓ 2

W1 V1

G ↓ 2 H

Scale 2

↓ 2

W2 V2 . . .

G ↓ 2 H

Scale J

↓ 2

WJ VJ 2 Estimation Removal V0

G ↓ 2 H

Scale 1

↓ 2

W1 V1

R1

−f (V1)

G ↓ 2 H

Scale 2

↓ 2

W2 V2 . . . R2

−f (V2)

G ↓ 2 H

Scale J

↓ 2

WJ VJ RJ

−f (VJ)

48 / 77

slide-49
SLIDE 49

Title Detour Motivation Standards Proposals Summary

RWA - 1rt Operation: DWT

V0 − →

  • VJ, (Wj)1≤j≤J

Vj−1 →

  • Vj, Wj

V0

z y x 1

W1 V1

Image V0 ∈ Rz×m, m = x · y Scale 1: (V1, W1)

49 / 77

slide-50
SLIDE 50

Title Detour Motivation Standards Proposals Summary

RWA - 2nd Operation: Estimation Removal

Vj−1 →

  • Vj, Wj

f

  • Vj

= Wj

W1 V1 V1 W1

  • W1

f (V1)

Scale 1: (V1, W1)

50 / 77

slide-51
SLIDE 51

Title Detour Motivation Standards Proposals Summary

RWA - 2nd Operation: Estimation Removal

Vj−1 →

  • Vj, Wj

f

  • Vj

= Wj Rj = Wj − Wj

V1 V1

  • W1

f (V1) R1 R1

− W1

Scale 1: (V1, R1)

51 / 77

slide-52
SLIDE 52

Title Detour Motivation Standards Proposals Summary

RWA Output Image

R1 R2 RJ VJ

RWA Scale J: (VJ, RJ, ..., R1) Original

  • Avg. R.2 = 0.5718

J-RWA

  • Avg. R.2 = 0.0052

52 / 77

slide-53
SLIDE 53

Title Detour Motivation Standards Proposals Summary

Squared Correlation Coefficients

Original

  • Avg. R.2 = 0.5718

KLT

  • Avg. R.2 = 0.0045

1-DWT

  • Avg. R.2 = 0.4665

1-RWA

  • Avg. R.2 = 0.1486

J-DWT

  • Avg. R.2 = 0.4033

J-RWA

  • Avg. R.2 = 0.0052

53 / 77

slide-54
SLIDE 54

Title Detour Motivation Standards Proposals Summary

RWA Regression Models

Regression Models

  • Maximum →

Wj

i = fi

  • Vj

i∈A

  • , where A = {1, .., k}
  • Restricted →

Wj

i = fi

  • Vj

i∈A′

  • , where A′ ⊂ A

Regression Variations

Exogenous: β → β Fast: m → m’ = m ×ρ

54 / 77

slide-55
SLIDE 55

Title Detour Motivation Standards Proposals Summary

Dataset: Uncalibrated Images

  • z:

spectral components

  • y:

height

  • x:

width

Corpus Image AVIRIS Yellowstone, sc: 0, 3, 10, 11, 18 z=224, y=512, x=680 Hawaii (x=614) and Maine Hyperion ErtaAle (y=3187) z=242 Lake Monona (y=3176) x=256

  • Mt. St. Helens (y=3242)

IASI Level 0 L0 1: 20091007093900Z z=8359 L0 2: 20091007143900Z y=1528 L0 3: 20100319050300S6 x=60 L0 4: 20120718075700Z Yellowstone00 (band:100) Hawaii (band:100) Erta Ale (band:100) 55 / 77

slide-56
SLIDE 56

Title Detour Motivation Standards Proposals Summary

Results: AVIRIS Yellowstone Corpus (Avg. bpppc)

Spectral Transform Level Spectral Transform Type Maximum Fast,ρ =0.01 Restricted Exogenous KLT CCSDS 123.0

— 10.453 10.453 8.884 —

Haar

7.869 7.883 7.544 7.774 1

IWT 5/3

7.785 7.800 7.296 7.693

IWT 9/7M

7.812 7.827 7.298 7.693

Haar

5.876 5.896 6.440 5.880 5.812 6.150 5

IWT 5/3

5.873 5.894 6.000 5.873

IWT 9/7M

5.882 5.902 5.937 5.873

Haar

5.830 5.850 6.442 5.841 8

IWT 5/3

5.850 5.871 6.015 5.854

IWT 9/7M

5.856 5.876 5.944 5.854

56 / 77

slide-57
SLIDE 57

Title Detour Motivation Standards Proposals Summary

Results: Hyperion Corpus (Avg. bpppc)

Spectral Transform Level Spectral Transform Type Maximum Fast,ρ =0.01 Restricted Exogenous KLT CCSDS 123.0

— 5.119 5.119 5.119 —

Haar

4.820 4.828 4.663 4.922 1

IWT 5/3

4.801 4.808 4.735 4.899

IWT 9/7M

4.814 4.821 4.757 4.912

Haar

4.543 4.552 4.384 4.755 4.550 4.293 5

IWT 5/3

4.566 4.575 4.475 4.771

IWT 9/7M

4.577 4.586 4.500 4.784

Haar

4.536 4.545 4.376 4.753 8

IWT 5/3

4.563 4.572 4.473 4.775

IWT 9/7M

4.573 4.582 4.495 4.786

57 / 77

slide-58
SLIDE 58

Title Detour Motivation Standards Proposals Summary

Results: IASI L0 Corpus (Avg. bpppc)

Spectral Transform Level Spectral Transform Type Maximum Exogenous KLT CCSDS-123.0

— 5.656 —

Haar

3.834 + 0.688 3.869 1

IWT 5/3

3.773 + 0.688 3.809

IWT 9/7M

3.800 + 0.688 3.835

Haar

2.499 + 0.917 2.547 — 2.897 5

IWT 5/3

2.550 + 0.917 2.600

IWT 9/7M

2.586 + 0.917 2.635

Haar

2.400 + 0.918 2.449 14

IWT 5/3

2.462 + 0.918 2.512

IWT 9/7M

2.498 + 0.918 2.548

58 / 77

slide-59
SLIDE 59

Title Detour Motivation Standards Proposals Summary

Results: Hyperion Corpus (R2, E(%), H)

Spectral Transform Level Spectral Transform RWA

  • Avg. R2

DWTs

  • Avg. R2

RWA

  • Avg. E(%)

DWTs

  • Avg. E(%)

RWA

  • Avg. H

DWTs

  • Avg. H

— 0.276 0.276 100 100 9.571 9.571

Haar

0.075 0.133 49.896 50.221 8.520 9.460 1

IWT 5/3

0.073 0.112 50.018 50.331 8.466 9.281

IWT 9/7M

0.074 0.112 49.894 50.211 8.489 9.286

Haar

0.005 0.085 3.014 5.265 5.749 8.435 5

IWT 5/3

0.005 0.052 3.262 4.829 5.787 8.036

IWT 9/7M

0.005 0.050 3.192 4.557 5.794 7.987

Haar

0.005 0.085 0.394 2.780 5.527 8.313 8

IWT 5/3

0.005 0.052 0.376 2.686 5.550 7.980

IWT 9/7M

0.005 0.050 0.379 2.382 5.552 7.921

59 / 77

slide-60
SLIDE 60

Title Detour Motivation Standards Proposals Summary

RWA: Precision and Entropy Decrease

4 5 6 7 8 9 10 11 12 13 14 15 50 100 150 200

bits Spectral components

Precision

4 5 6 7 8 9 10 11 12 13 50 100 150 200

bits Spectral components

Entropy

  • Avg. Precision

Original Haar-DWT Haar-RWA

  • Avg. Entropy

Original Haar-DWT Haar-RWA 60 / 77

slide-61
SLIDE 61

Title Detour Motivation Standards Proposals Summary

Discussion

  • Lossless RWA can outperform other widespread techniques:
  • CCSDS-123.0
  • KLT-based methods
  • DWTs

RWA (Max.) RWA (Exog.) Haar-DWT IWT 5/3 KLT CCSDS-123.0 Hawaii

2.55 2.82 3.37 3.27 3.20 2.70

IASI L0 2

3.24 2.39 2.99 3.02 2.86 2.88

  • RWA is suitable for different DWT filters with similar

performance.

  • The regression stage decorrelates almost all the remaining

spectral redundancy the DWT step can not remove.

61 / 77

slide-62
SLIDE 62

Title Detour Motivation Standards Proposals Summary

The lossy-RWA Inverse Process

Wj = Rj + Wj,

  • Wj = f [Vj]

(Vj, Wj) DWT −1 − − − − − → Vj−1

  • Error propagation.

δRj ± Rj = ⇒ δj ± ( Wj)1≤j≤J ∈ Rm×z.2−j, j > 1

R1 R1 VJ VJ R2 R2 RJ RJ

  • W1
  • W1
  • W2
  • W2
  • WJ
  • WJ

Estimates: ( Wj)1≤j≤J

62 / 77

slide-63
SLIDE 63

Title Detour Motivation Standards Proposals Summary

Error Propagation

R1 R1 VJ VJ R2 R2 RJ RJ

  • W1
  • W1
  • W2
  • W2
  • WJ
  • WJ

Estimates: ( Wj)1≤j≤J

˜ Wj= Rj + Wj+δRj

  • Ri1≤i<j

Vj δRj ±Rj δj ±

  • Wi1≤i<j
  • Wj

63 / 77

slide-64
SLIDE 64

Title Detour Motivation Standards Proposals Summary

Predictive Weighting Scheme (PWS)

N(Rj) = 1

z

  • z · j−1

i=1 2−i

, if j > 1 , if j = 1

W(Rj) =

1 N(Rj)−N(Rj−1) = 2j−1

W(Vj) = 2j

R1 20 VJ 2J R2 21 RJ 2J−1

64 / 77

slide-65
SLIDE 65

Title Detour Motivation Standards Proposals Summary

Rate-Distortion Performance. Different Regression Models

40 50 60 70 80 90 100 110 120 1 2 3 4 5 6 7

PSNR (dB) bpppc

PLL-RWA

50 60 70 80 90 100 110 120 1 2 3 4 5 6 7

PSNR (dB) bpppc

PLL-PWS-RWA

Yellowstone03

Maximum Fast Restricted Exogenous 65 / 77

slide-66
SLIDE 66

Title Detour Motivation Standards Proposals Summary

Rate-Distortion Performance. Different Images

50 60 70 80 90 100 110 1 2 3 4 5 6

PSNR (dB) bpppc

PLL-RWA

50 60 70 80 90 100 110 120 1 2 3 4 5 6

PSNR (dB) bpppc

PLL-PWS-RWA

Images

Yellowstone03 Hawaii Erta Ale IASI LO 1 66 / 77

slide-67
SLIDE 67

Title Detour Motivation Standards Proposals Summary

Rate-Distortion Performance. Different Transforms

30 40 50 60 70 80 90 100 110 120 2 4 6 8 10

PSNR (dB) bpppc

PLL

30 40 50 60 70 80 90 100 110 120 2 4 6 8 10

PSNR (dB) bpppc

PLL-PWS

Yellowstone03

Original IWT 5/3 IWT 5/3-RWA 67 / 77

slide-68
SLIDE 68

Title Detour Motivation Standards Proposals Summary

PWS: Bit Per Pixel Per Component (bpppc) distribution

1 2 3 4 5 6 7 8 9 10 50 100 150 200

bits Spectral components

PLL

2 4 6 8 10 12 50 100 150 200

bits Spectral components

PLL-PWS

Yellowstone03

1 bpppc - Lossy 3 bpppc - Lossy 5 bpppc - Lossy 5.9 bpppc - Lossless 68 / 77

slide-69
SLIDE 69

Title Detour Motivation Standards Proposals Summary

Discussion

  • Progressive Lossy-to-Lossless RWA can outperform other

widespread techniques:

  • KLT-based methods
  • DWTs

Rate-Distortion Computational Complexity

1 2 3 4 5 6

Bit-rate (bpppc)

30 40 50 60 70 80

SNR (dB)

RWA Maximum (k= 112) RWA Restricted (Neighbors Sel. k′ = 11) PCA DWT5/3

1.0 1.1 1.2 1.3 1.4 1.51e11

PCA reversible RWA Maximum k = 112

  • Exo. RWA

Maximum k = 112 RWA Restricted k′ = 11

  • Exo. RWA

Restricted. k′ = 11 DWT 5/3 reversible

1 2 3 4 51e10

69 / 77

slide-70
SLIDE 70

Title Detour Motivation Standards Proposals Summary

Inpainting-based

  • 1. N. Amrani, J. Serra-Sagrist`

a, P. Peter and J. Weickert, ”Diffusion-Based Inpainting for Coding Remote-Sensing Data,” IEEE Geoscience and Remote Sensing Letters, Submitted, November 2016.

70 / 77

slide-71
SLIDE 71

Title Detour Motivation Standards Proposals Summary

Inpainting-based Lossy Compression Scheme

Image Spatial Inpainting Spectral Inpainting Prediction Entropy Coder Difference

f f u r w h y

  • 2D Homogeneous Diffusion

Probabilistic Sparsification

  • 1D Biharmonic Diffusion

NonLocal Pixel Exchange

71 / 77

slide-72
SLIDE 72

Title Detour Motivation Standards Proposals Summary

2D inpainting solution using 5% of the spatial pixels Correlation matrices for AVIRIS Yellowstone sc 00 Radiance (224 components)

1 56 112 224 1 56 112 224 1 56 112 224 1 56 112 224

0.5 1

Original (σ2=2.35×105) Residual (σ2=1.17×104)

72 / 77

slide-73
SLIDE 73

Title Detour Motivation Standards Proposals Summary

1D inpainting solution using 10% of the spectral pixels Correlation matrices for AVIRIS Yellowstone sc 10 Radiance (224 components)

e

1 56 112 224

h(w)

1 56 112 224

e′ =f−h(w)

0.5 1

Correlation: e vs. w. Correlation: e′ vs. h(w). MSE = 1

n

  • i e2

i = 653.7

MSE = 1

n

  • i e′2

i

= 49.9 Without prediction With prediction

73 / 77

slide-74
SLIDE 74

Title Detour Motivation Standards Proposals Summary

0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1

Bitrate (bpp)

34 36 38 40 42 44 46

SNR (dB)

2D 10% 1D 90% 2D 25% 1D 75% 2D 50% 1D 50% 2D 75% 1D 25% 0.0 0.5 1.0 1.5 2.0 2.5 Bitrate (bpp) 15 20 25 30 35 40 45 50 55 60 SNR (dB) Yellowstone: 2D+1D Yellowstone: 2D+1D+h( ·) Maine: 2D+1D Maine: 2D+1D+h( ·) Hawaii: 2D+1D Hawaii: 2D+1D+h( ·)

Percentage distribution Prediction impact

  • f the known data

74 / 77

slide-75
SLIDE 75

Title Detour Motivation Standards Proposals Summary

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Bitrate (bpp) 30 35 40 45 50 55 SNR (dB)

Inpainting 2D+1D+h( ·) JPEG2000 part II (DWT 97) JPEG2000 part II (PCA)

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Bitrate (bpp) 28 30 32 34 36 38 40 42 44 46 SNR (dB)

Inpainting 2D+1D+h( ·) JPEG2000 part II (DWT 97) JPEG2000 part II (PCA)

Maine (Uncalibrated) Hawaii (Uncalibrated)

0.0 0.5 1.0 1.5 2.0 2.5 Bitrate (bpp) 30 35 40 45 50 55 SNR (dB)

Inpainting 2D+1D+h( ·) JPEG2000 part II (DWT 97) JPEG2000 part II (PCA)

Yellowstone Sc 00 (Rad)

75 / 77

slide-76
SLIDE 76

Title Detour Motivation Standards Proposals Summary

Summary

  • 1. Remote sensing data provides a wealth of information worth

exploring and investigating.

  • 2. Remote sensing data need be compressed.
  • 3. We have proposed several approaches for data compression.

Open questions

  • 1. [Still Image & Video compression]

Multiple-reference inpainting-based compression ?

  • 2. [Remote sensing data compression]

Further improvements ? Inpainting-based compression ?

  • 3. Near-lossless / Visually-lossless inpainting-based compression ?

Looking forward to collaborate with you. Joan.Serra@uab.cat www.gici.uab.cat

76 / 77

slide-77
SLIDE 77

Title Detour Motivation Standards Proposals Summary

Inpainting vs RWA rate-distortion performance

0.0 0.5 1.0 1.5 2.0 2.5 Bitrate (bpp) 30 35 40 45 50 55 SNR (dB)

Inpainting 2D+1D+h( ·) JPEG2000 part II (DWT 97) JPEG2000 part II (PCA)

1 2 3 4 5 6

Bit-rate (bpppc)

30 40 50 60 70 80

SNR (dB)

RWA Maximum (k= 112) RWA Restricted (Neighbors Sel. k′ = 11) PCA DWT5/3

Inpainting RWA AVIRIS Yellowstone Sc 00 (Radiance)

77 / 77