A comparison of commercial Pan-sharpening techniques for HR … · 2013-06-28 · 6/21/2013 Rakesh...

A comparison of commercial Pan-sharpening techniques for HR Satellite imagery

Rakesh Kumar Mishra and Yun ZhangDepartment of Geodesy and Geomatics Engineering

University of New Brunswick

ESRI International User Conference 2013

6/21/2013 2 Rakesh K. Mishra

Contents

Why pansharpening ?

Pansharpening comparison

FuzeGo Pansharpening

Conclusion

6/21/2013 3 Rakesh K. Mishra

Importance of pan-sharpening

• More than 70% of optical satellites simultaneously collect a low resolution MS image and a high resolution Pan image.

• Most remote sensing and GIS applications desire high resolution colour image.

• Effective and high quality pan-sharpening technique is crucial for the success of many GIS/remote sensing applications.

• Research on pan-sharpening has been conducted since the mid-1980s.

6/21/2013 4 Rakesh K. Mishra

6/21/2013 5 Rakesh K. Mishra

Kodak’s patent

Yun Zhang’s paper

SpaceImaging’s patent

6/21/2013 6 Rakesh K. Mishra

Tens of thousands papers on pan-sharpening haves been published,

only a few outstanding pan-sharpening algorithms have been adopted by industry.

6/21/2013 7 Rakesh K. Mishra

Widely used commercial pan-sharpening algorithms

ERDAS IMAGINE:

• Subtractive Resolution Merge• HPF Resolution Merge • Modified IHS Resolution Merge• Wavelet Resolution Merge• Ehlers Fusion• HCS Resolution Merge• Resolution Merge

ENVI:

• CN Spectral Sharpening• Color Normalized(Brovey) • Gram-Schmidt• HSV Sharpening• PC Spectral Sharpening

6/21/2013 8 Rakesh K. Mishra

Widely used commercial pan-sharpening algorithms

ESRI:

• Brovey • ESRI • Gram-Schmid• IHS• Simple Mean

FuzeGo:

• UNB Pansharpening

HighView:

• Advanced Global Optimization

6/21/2013 9 Rakesh K. Mishra

Comparison among different pan-sharpening algorithmsData used for the pan-sharpening:

• IKONOS, 2002, Fredericton, Canada (Pan 540 MB, MS 135 MB)

• QuickBird, 2007, Beijing, China (Pan 820 MB, MS 205MB)

• GeoEye-1, 2009, Hobart, Australia (Pan 816 MB, MS 204 MB)

• WorldView-2, 2010, Moncton, Canada (Pan 131 MB, MS 65 MB).

All the images were in 16-bit format. All the MS bands were used in the test.

6/21/2013 10Rakesh K. Mishra

Pan-sharpening evaluation

Visual Analysis

Quantitative Analysis

There is no consensus on quantitative evaluation methods

6/21/2013 11Rakesh K. Mishra

All the images are displayed under the same visualization condition, i.e.

• The same area of the images before and after pan-sharpening are displayed, and

• The same histogram stretching is applied to all the images.

Comparison Results

6/21/2013 12Rakesh K. Mishra

Can not process images with more then 4 bands.

23/07/2012 12

6/21/2013 13Rakesh K. Mishra23/07/2012

6/21/2013 14Rakesh K. Mishra23/07/2012

6/21/2013 15Rakesh K. Mishra

Can not process images with more than 4 bands

23/07/2012 15

6/21/2013 16Rakesh K. Mishra23/07/2012 16

6/21/2013 17Rakesh K. Mishra23/07/2012 17

23/07/2012 18C. Pacwick, M. Deskevich, F. Pacifici, and S. Smallwood (2010): WorldView-2 PanSharpening. ASPRS 2010 Annual Conference, San Diego, California, April 26-30, 2010

(Kodak fusion)

WorldView-2 WorldView-2 (natural colour) (traditional)

(widely used) (newly developed in 2010)

(launched in 2009)

23/07/201219

00000000000000000000000111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111112222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000077777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////2222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222200000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000

6/21/2013 20Rakesh K. Mishra

Comparison conclusions

• FuzeGo produces constant best results for all type of sensors and areas.

• Other techniques work well for some images or sensors, but not for others

• The second best technique is Gram-Schmidt (Kodak’s patent). But it produces poor results for WorldView-2 images.

• The third best technique is HCS, but still result is notconsistent.

6/21/2013 21Rakesh K. Mishra

Further examples of FuzeGO

IKONOS, July 2000San Diego, USA

FuzeGo

Original IKONOS Pan and MS images courtesy of Space Imaging Inc. (now GeoEye Inc.)

23/07/2012 22

23/07/2012 23

23/07/2012 24

23/07/2012 25

23/07/2012 26

23/07/2012 27

23/07/2012 28

23/07/2012 29

23/07/2012 30

23/07/2012 31

23/07/2012 32

23/07/2012 33

23/07/2012 34

23/07/2012 35

23/07/2012 36

23/07/2012 37

23/07/2012 38

23/07/2012 39

23/07/2012 40

QuickBird, May 2005Beijing, China

FuzeGo

Original QuickBird Pan and MS images courtesy of DigitalGlobe Inc.

23/07/2012 41

23/07/2012 42

23/07/2012 43

23/07/2012 44

23/07/2012 45

23/07/2012 46

23/07/2012 47

23/07/2012 48

23/07/2012 49

23/07/2012 50

23/07/2012 51

GeoEye-1, February 2009Hobart, Australia

FuzeGo

Original GeoEye-1 Pan and MS images courtesy of GeoEye Inc.

23/07/2012 52

23/07/2012 53

23/07/2012 54

23/07/2012 55

23/07/2012 56

23/07/2012 57

23/07/2012 58

GeoEye-1 satellite, original natural colour, 2mLaunched in 2008

23/07/2012 59

GeoEye-1, original B/W, 0.5m

23/07/2012 60

GeoEye-1, FuzeGo, 0.5m

23/07/2012 61

GeoEye-1, original natural colour, 2m8x enlarged

23/07/2012 62

GeoEye-1, original B/W, 0.5m2x enlarged

23/07/2012 63

GeoEye-1, FuzeGo, 0.5m2x enlarged

23/07/2012 64

WorldView-2, July 2010Moncton, Canada

FuzeGo

Original WorlView-2 Pan and MS images courtesy of DigitalGlobe Inc.

23/07/2012 65

www.DigitalGlobe.com

1 2 3 4 5 6 78

23/07/2012 66

23/07/2012 67

23/07/2012 68

23/07/2012 69

23/07/2012 70

23/07/2012 71

23/07/2012 72

23/07/2012 73

23/07/2012 74

23/07/2012 75

6/21/2013 76Rakesh K. Mishra

FuzeGo Software

http://www.fuzego.com/

6/21/2013 77Rakesh K. Mishra

FuzeGo in ENVI 5.1 Beta

FuzeGo achieves constant good fusion results regardless of the differences in sensors, areas and seasons.

Fully automatic

FuzeGo achieves:maximum detail increasing, minimum colour distortion, andoptimal colour and feature integration.

UNB-Pansharp has been widely recognized by industry

23/07/2012 78

Conclusion

23/07/2012 79

• DigitalGloble Inc. and GeoEye Inc. provided the original Pan and MS images.

• The comparative research was conducted at the University of New Brunswick, Canada.

• GEOIDE MDF program provided the funding for the research.

Acknowledgments

6/21/2013 80Rakesh K. Mishra