Urban area detection and segmentation using OTB

2,317 views

Published on

Urban area detection and segmentation using OTB
Stéphane May; CNES
Jordi Inglada; CNES

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
2,317
On SlideShare
0
From Embeds
0
Number of Embeds
253
Actions
Shares
0
Downloads
125
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Urban area detection and segmentation using OTB

  1. 1. Urban area detection and segmentation using OTB Stéphane MAY Jordi INGLADA stephane.may@cnes.fr jordi.inglada@cnes.fr orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 1
  2. 2. Urban area detection Context Floodings due to Hurricane Ike over Gonaives (Haïti) – September 2009  Detect urban area to evaluate the impact zone Urban area detection applied to: Rapid mapping Risk management Urbanism planning, ... orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 2
  3. 3. Urban area detection and segmentation Contents of the presentation Introduction Algorithms description Parameter estimation and validation process Analysis of the results Conclusion orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 3
  4. 4. Introduction Key requirements of the proposed algorithms Low computation time Over detection preferred to misdetection Refined algorithms may be applied in further steps Algorithm independent of the image resolution Mixing radiometric techniques (NDVI, NDWI mask) and texture algorithms 3 algorithms based on texture : Edge density Pantex detector Gabor Texture Index filtering algorithm orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 4
  5. 5. Edge density algorithm Computation of a local density of edges around a pixel Sobel filter (edge detection) Application of a threshold Computation of density of detected edges One parameter : the filter radius orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 5
  6. 6. Pantex detector Computation of the Haralick contrast texture descriptor Considering 8 directions, computation of the co-occurence matrixes Optimal histogram bin defined by the Scott formula Computation of the Haralick contrast descriptors The output value of the filter is the min value of contrasts 1 parameter : radius of the filter orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 6
  7. 7. Gabor Texture Index filtering algorithm (1) A Gabor filter is a linear filter whose impulse response is defined by a harmonic function multiplied by a Gaussian function A=0.10 A=0.12 B=0.20 B=0.12 =0° =45° f=0.25 f=0.12 Interesting properties to detect textures and alignments orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 7
  8. 8. Gabor Texture Index filtering algorithm (2) Proposed algorithm Gabor filtering with N directions (Gabor filters bank) Computation of the local standard deviation with a sliding window Selection of the median value for the N directions (x M channels) Hard thresholding to build the binary urban mask orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 8
  9. 9. Gabor Texture Index filtering algorithm (3) Method applicable using each channel of the image or intensity image Many parameters... Gabor filter parameters Number of directions Radius of the standard deviation filter  Evaluation of the best parameters set orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 9
  10. 10. Urban area detection and segmentation Contents of the presentation Introduction Algorithms description Parameter estimation and validation process Analysis of the results Conclusion orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 10
  11. 11. Parameter estimation and validation process (1) Use of a Quickbird multispectral image Creation of a reference mask : urban / non urban NDVI, and thresholding of clouds, bare soils Reference mask built for several resolutions : 30m, 20m, 10m, 5m, 2.5m, 1.25m, 0.62m Monte-Carlo simulation used to evaluate several parameters sets for the Gabor Texture Index (GTI). The generated mask is compared to the reference mask : Computation of True Positive Ratio (TPR), False Positive Ration (FPR) Computation of TPR and FPR after application of the NDVI mask orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 11
  12. 12. Parameter estimation and validation process (2) ROC simulations : representation of TPR versus FPR Indication on processing time (depending of the filters radius) orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 12
  13. 13. Parameter estimation and validation process (3) Selection of interesting parameters for the GTI detector Robustness to the parameters choice orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 13
  14. 14. Urban area detection and segmentation Contents of the presentation Introduction Algorithms description Parameter estimation and validation process Analysis of the results Conclusion orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 14
  15. 15. Analysis of processing time Comparison of the 3 algorithms ED and GTI are compliant with preliminary specifications Pantex filter is not applicable at high resolution orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 15
  16. 16. Results – Edge Density Great improvement thanks to the NDVI orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 16
  17. 17. Results – Gabor Texture Index Slight performance increase for the GTI (flat curves for lower false alarm rates) : Gabor sensitive to alignments orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 17
  18. 18. Results – Pantex filter Pantex seems to give the relative worse results in this experiment but differences are small orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 18
  19. 19. Conclusion 3 interesting algorithms to detect urban areas Edge density filter : one parameter method, robust and fast Gabor : slight enhancement of performance among the 3 algorithms (sensitive to alignments) Pantex : able to detect urban areas, but penalized by the high computation complexity Application of the NDVI mask and NDWI enhances the results Mix of radiometry method and texture method is an excellent approach Next step : use algorithms based on geometric features to detect buildings... orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 19
  20. 20. OTB contributions Edge detector and pantex available in a packaged application otbUrbanAreaExtractionApplication Gabor Texture Index filter soon available in the OTB orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 20
  21. 21. Urban area detection Thank you for your attention orfeo-toolbox.org S. MAY – J. INGLADA – Urban area detection – IGARSS 2009 21

×