High resolution satellite imagery for spatial data acquisition

High resolution satellite imagery for spatial data acquisition Wenzhong (John) Shi The Hong Kong Polytechnic University

Outline n n n Image fusion: Multi-band waveletbased method Feature extraction: Line segment match method Geometric correction: Line-based transformation model

High resolution satellite images

An IKONOS image

Several types of available high resolution satellite images Lunching Swath Width Time (km) Resolutio n（m) Satellite Company Quick Bird Earth Watch 1999 22 0. 6 Ikonos Space imaging 1999 11 1 orbview 3 Orbimmage 1999 8 1 orbview 4 Orbimmage 2000 8 0. 5 Eros B West Indian Space 1999 13. 5 1. 3 Spot 5 Spot Image 2001 60 5

Technologies for high resolution satellite image processing n n n n Georeferencing Orthorectification Image fusion DEM generation Classification Feature extraction High-resolution aerial photogrammetry

Our development n n n Image fusion: Multi-band wavelet-based method Feature extraction: Line segment match method Geometric correction: Line-based transformation model

Multi-band wavelet-based image fusion

Two-band multi-band wavelet transformation Multi-based wavelet: flexible in scale The original image The 2 -band wavelet transformed image The 3 -band wavelet transformed image

Image fusion for multi -scale satellite images n n n Images: panchromatic and multi-spectral images Spatial resolution Ratio of spatial resolutions: n n (a) 2 n (n = 1, 2, 3, …), for example 2, 4, 8, etc (b) 3, 5, 7 etc.

Two examples n n Multi-band wavelet for fusing SPOT panchromatic and multi-spectral image (10 m and 30 m) Multi-band wavelet for fusion of IKONOS Images (1 m and 4 m)

Fusion of IKONOS Images Four-band wavelet transformation

Test IKONOS Image 1 M 4 M

Result Assessment Method Original Images Image fused by 3 -band W. T. Image fused by 2 -band W. T. Image fused by IHS method Image M 1 M 2 M 3 F 1 F 2 F 3 C. E. : the combination entropy M. G. : the mean gradient W. T. : wavelet transformation C. C. : correlation coefficient C. E. 9. 7123 11. 7735 11. 2665 11. 4623 M. G. 10. 6102 5. 1062 3. 7069 17. 0242 10. 5206 8. 9659 16. 7243 9. 2284 6. 8934 16. 4425 8. 4133 6. 0456 C. C. 0. 9624 0. 8794 0. 9548 0. 8798 0. 8819 0. 7913 0. 8241 0. 7157 0. 8098

Line Segment Match method for road extraction

An example of road extraction A one-meter resolution satellite image of Valparaiso

- A road with a certain width can be considered as a set of straight-line segments. - To detect a road is to detect the corresponding straight-line segments with a certain length and direction. - Form the foundation of the road network detection method developed -- line segment match method. - A feature-based method for road network extraction from high-resolution satellite image.

The final extracted road network from the image Filling short small gaps, connecting line segments, deleting crude line segments Based on the knowledge about the roads

Accuracy of road extraction（Unit：%） Image Accuracy Image-1 90. 64 9. 36 0. 82 Image-2 91. 02 8. 98 0. 43 90. 42 9. 58 0. 36 90. 69 9. 31 0. 54 Image-3 Average Omission error Commission error

The Line Based Transformation Model

Our Research Objectives § To study the applicability and evaluate the accuracy of the results using existing point-based empirical mathematical models § To develop a new mathematical model for image rectification by using line features.

§ The LBTM developed in this research overcomes most of the problems encountered when using linear features with the present generation of rigorous mathematical models. § The model is applicable to various satellite imageries. § The model does not require any further information about the sensor model and satellite ephemeris data. § It does not need any initial approximation values.

Principle of modeling uncertainties in spatial data and analysis

Further contact: Wenzhong (John) Shi Dept. of Land Surveying and Geoinformatics The Hong Kong Polytechnic University Tel: +852 - 2766 5975 Fax: +852 – 2330 2994 Email LSWZSHI@POLYU. EDU. HK