ORFEO Toolbox Jordi Inglada - CNES TOoLs for Open Mul/- risk assessment using Earth Observa/on data
Outline ORFEO Toolbox : general characteris>cs Example of OTB features OTB Applica>ons & Processing Chains Uses for
General characteris>cs
Orfeo Toolbox Framework: ORFEO Accompaniment Program Goals : make easier the development of new algorithms, their valida>on and capitalisa>on, fill the gap between researchers and ORFEO users OTB= Image processing library with focus on feature extrac>on Open source soqware for Image Processing labs, users and the industry CNES is responsible for design and specifica>on. Contractor : C- S
Orfeo Toolbox C++ library based on exis>ng developments OTB Applications External libs Utilities Users Library Developers
Orfeo Toolbox l External libs : l l l l l ( registration ITK (segmentation, ( ortho OSSIM (carto, ( GUI ) FLTK ( classification LibSVM (supervised learning and ( formats GDAL (IO for remote sensing
Orfeo Toolbox Version 2 (current): IO (image, vector), ortho- rec>fica>on, registra>on Pansharpening, padiometric correc>ons, filtering, segmenta>on and classifica>on, feature extrac>on (texture, lines, végéta>on indices), basic change detec>on, supervised learning, object coun>ng spa>al reasoning. U>li>es Quick look, ROI extrac>on, meta- data access, Ortho- rec>fica>on, cartographic projec>ons, classifica>on
Higher level languages: bindings Not all users like C++ programming Higher level languages are more appealing Many users have legacy code in other languages Rapid prototyping with an interac>ve command line OTB provides Python + Java IDL/ENVI add- ons
The Data Pipeline File Reader Ima ge Filter Ima ge Writer File
Data types Images (any pixel type, any number of bands) Meshes Polylines VectorData: OGC geometries (point, line, polygon, mul>- polygons) LabelObjects PostGIS tables
OTB Goodies Streaming / Threading Transparent Image Format Handling Iterators Composite Filters Frameworks Registra>on Change Detec>on Classifica>on Level sets
Examples of OTB features
Line detec>on on SAR images
Meaningful alignment detec>on
Change detec>on
Denoising Original Blurring Edge preserving
Circle extrac>on
Line detec>on and extrac>on
Watershed segmenta>on
Object segmenta>on
Markov random fields segmenta>on
An>- speckle filtering
Object detec>on (correla>on+ region growing )
OTB Applica>ons & Processing Chains
Goals and perimeter OTB lib is now rich enough Ready to use tools (no need for OTB compila>on, etc.) for thema>c valida>on and opera>onal use Complementary with methodology developed by thema/c users: capitalize, generalize, automate, deploy To be made availabe as IDL/ENVI add- ons, QGIS plug- ins or Monteverdi modules
Exis>ng chains Ortho- registra>on (Pléiades, CSK, SPOT5, QB, Ikonos, TSX, etc.) Ortho+pansharpening (Pléiades, QB) Supervised pixel- based classifica>on (mul>spectral, mul>temporal) Object Segmenta>on Standard FAO, Corine land cover map produc>on (Pléiades, QB, Ikonos, SPOT5) KML conversion (display & share on Google Earth) 3D & stereo anaglyh viewer
Exis>ng chains Image registra>on Object coun>ng (Pléiades, QB, Ikonos) Road network extrac>on (Pléiades, QB, Ikonos, SPOT5) Hydrographic network extrac>on (Pléiades, QB, Ikonos, SPOT5) Urban area extrac>on (Pléaides, CSK, QB, Ikonos, TSX, SPOT5) Image to Data Base registra>on (Pléiades/QB to BDTopo) Radiometric calibra>on (Pléiades, SPOT5, QB, CSK, ASAR, ERS, RadarSAT)
Image Viewer
Stereo Viewer
Ortho- rec>fica>on
Interac>ve Change Detec>on
Road (and hydrographic) Extrac>on
Feature extrac>on
Feature extrac>on
Feature extrac>on
Feature extrac>on
Feature extrac>on
Feature extrac>on
Feature extrac>on
Supervised Classifica>on (using SVM)
Segmenta>on by Region growing
Segmenta>on using SVM
Export to GoogleEarth Building shadows extracted from QB image
Object coun>ng
Urban area extrac>on
Urban area extrac>on
Image to DB registra>on
Image to DB registra>on
What is Monteverdi? Framework to simply and interac>vely build an image processing streaming pipeline Orfeo Toolbox components : Func>ons and filters Applica>ons Global framework w => Monteverdi
Monteverdi Main menu w File w SAR w Filtering w Learning w Geometry w Visualiza>on w Help
Monteverdi File w Open dataset w Save dataset : save results (no pipeline module created) w Extract ROI from dataset : extract ROI for pipeline computa>on w Save dataset (advanced) : save with choice of output format, channels, etc. w Cache dataset : execute previous pipelines elements and cache results w Concatenate images : concatenate several images w Quit
Monteverdi File > Extract ROI
Monteverdi File > Concatenate images
Monteverdi Viewer Visualiza>on w Viewer (1/4) 6 novembre 2009
Monteverdi Visualiza>on w Viewer (2/4) 6 novembre 2009
Monteverdi Visualiza>on w Viewer (3/4) 6 novembre 2009
Monteverdi Visualiza>on w Viewer (4/4) 6 novembre 2009
Monteverdi Geometry w Orthorec>fica>on w Reproject image : reprojec>on of ortho- images w Superimpose two images w Homologous points extrac>on : manual selec>on of homologous points, then re- sampling of the image on the fixed one. w GCP to Sensor Model : manual set of longitude / la>tude of geographical points, then re- sample of the image. 6 novembre 2009
Monteverdi Menu Geometry > Orthorec>fica>on (1/2)
Monteverdi Menu Geometry > Orthorec>fica>on (2/2)
Monteverdi Menu Geometry > Reproject image
Monteverdi Menu Geometry > Superimpose two images
Monteverdi Menu Geometry > Homologous points extrac>on
Monteverdi Menu Geometry > GCP to sensor model
Monteverdi Filtering w Feature Extrac>on w MeanshiQ clustering w Pan- sharpen an image w Band math w Change detec>on w Threshold
Monteverdi Filtering w Feature Extrac>on (1/3)
Monteverdi Filtering w Feature Extrac>on (2/3)
Monteverdi Filtering w Feature Extrac>on (3/3) More than 65 features Mean, variance, Gradient, spectral angle Textures (energy, entropy, contrast, etc) Morphological filters Radiometric indexes Vegeta>on (NDVI, ARVI, etc), Soil, Built up, Water Edge density Mean shiq Original data (=> no need to concatenate channels aqer filtering) Etc Use before SVM classifica>on for example
Monteverdi Filtering w MeanshiQ clustering
Monteverdi Filtering w Pan- sharpen an image Requires Panchroma>c and Mul>spectral input images at the same resolu>on
Monteverdi Filtering w Band math Addi>on, Substrac>on, Mul>plica>on, Ra>o, ShiQ- Scale (A*X+B)
Monteverdi Filtering w Change Detec>on
Monteverdi Filtering w Threshold
Monteverdi Learning w SVM classifica>on w K- Means clustering
Monteverdi Menu Learning > SVM classifica>on (1/3)
Monteverdi Menu Learning > SVM classifica>on (2/3)
Monteverdi Menu Learning > SVM classifica>on (3/3)
Monteverdi Menu Learning > K- means
Monteverdi SAR w Despeckle image : apply Frost / Lee filter w Compute intensity and log- intensity
Monteverdi Menu SAR > Despeckle image
Monteverdi Menu SAR > Compute intensity and log- intensity
Uses for
WP2 Tools for human exposure (HUEX) Urban area extrac>on Object- based segmenta>on Spa>al reasoning
WP3 Tools for deforesta>on risk analysis (DERA) Use of vegeta>on indices Radiometric image calibra>on Vegeta>on reflectance models (recently introduced in OTB) SVAT models (coming soon to OTB)
WP4 Tools for earthquake physical (ERVU) vulnerability Goal: fuse GIS layers OTB «speaks» PostGIS QGIS plug- ins Spa>al reasoning Conceptual graphs + Fuzzy Constraint Sa>sfac>on (recently added to OTB)
WP5 Tools for flood vulnerability (FLOV) characteriza>on Water indices DEM manipula>on Object recogni>on (dams, dykes, etc.)
WP6 Tools for post- event risk (POER) management Change detec>on Pixel based Object based Thorough experience through the Charter CNES Risk chain is based on OTB