general presentation of otb

IntroductionBack in 2006

Key characteristicsFunctions and algorithmsWhat’s new in OTB 5.0 ?

Conclusion

State of the Orfeo ToolBoxOpen source library for remote sensing image processing

Julien Michel (CNES), Manuel Grizonnet (CNES)

http://www.cnes.fr

http://creativecommons.org/licenses/by-sa/3.0/

http://www.orfeo-toolbox.org



Conclusion

Outline

Introduction

Back in 2006

Key characteristics

Functions and algorithms

What’s new in OTB 5.0 ?

Conclusion

http://www.cnes.fr





Conclusion

Introduction

Goal of this presentation :

I Insight of project components

I Good practises to help starters usingthe library

I to go further and advanced use

Orfeo ToolBox is not a black box. . .Let’s start by opening the box !

http://www.cnes.fr





Conclusion

Things to know about OTB. . .

The Orfeo ToolBox is :

I A (The :) image processing library dedicated to remote sensing

I Free and open source software under CeCILL-v2 license(equivalent to GPL)

I Funded and developed by CNES (French Space Agency) in the frame of theOrfeo Pleiades program (and beyond)

I Written in C++ on top of ITK (medical image processing)

I Interfaces seamlessly with other IP and RS open-source software, like GDAL,OSSIM, OpenCV. . .

I Develop to allow processing of large data thanks to parallel and on the flowprocessing

www.orfeo-toolbox.org

http://www.cnes.fr



www.itk.org




Conclusion

Outline

Introduction

Back in 2006

Key characteristics



Conclusion

http://www.cnes.fr





Conclusion

How it starts ?

CNES Orfeo accompaniment program (2006-2014)

I Pleiades : gap in resolution comparing with SPOT5 which leads to new usages

I Goals : prepare, accompany and promote the use and the exploitation of theimages derived from Pleiades/COSMO-SkyMed satellites

I Preparatory phase from 2006 to 2012

I Thematic Commissioning activities from 2012 to 2014

OTB in the Orfeo program

I Answer to ORFEO user groups needs

I Capitalize CNES R&D in Information Extraction

I Deliver generic tools for Pleiades users

http://www.cnes.fr





Conclusion

Why Open source ?

ShowcaseImage processing library dedicated to remote sensing for Pleiades users. Its widedissemination contributes to the mission promotion.

Quality and efficiencyOTB covers a vast panel of applications and thematic fields.Openness should :

I Facilitate appropriation and validation for users

I Encourage contributions and bug reports

I Available on multiple platforms

I “The Cathedral & the Bazaar” 1 : the more widely available the source code is forpublic testing experimentation, the more rapidly all forms of bugs will bediscovered

Reproducible researchOTB capitalizes a part of the CNES R&D in IP, open source contributes totransparent, reproducible and trans-disciplinary research.

1. http://www.catb.org/esr/writings/cathedral-bazaar/

http://www.cnes.fr



http://www.catb.org/esr/writings/cathedral-bazaar/



Conclusion

A bit of history

2008 2010 2012 2014

1.0.0

Key steps

1.0.0 Architecture, compilation and documentation, few functions andapplications

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Key steps


2.0.0 More functions (SVM learning, feature extraction, pre-processing,vizualization. . . )

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Key steps



3.0.0 Support for vector data, Markov Random Field, keypoints, Kohonenmap. . . ) and more applications for demonstration (with GUI)

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

3.2.

0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Key steps




3.2.0 First version of Monteverdi, continue to enrich the library

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Etapes cles





3.12.0 New applications mechanisms, complete support for Pleiadesimagery, new functions. . .

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Etapes cles





3.12.0 New applications mechanisms, complete support for Pleiadesimagery, new functions. . .

3.16.0 Revamp of Monteverdi in Monteverdi2

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

4.4.

04.

2.0

4.0.0

3.20

.0

3.18

.0

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Key steps

4.0.0 Compatible with ITK 4.0, and more functions.

4.2.0 Speed-up in Haralick textures calculation, enhancement of theoptical calibration framework, RPC coefficients for sensor modelingcan now be read and written from/to GeoTIFF RPC tags. . .

4.4.0 Vector band math calculator based on MuparserX, New set ofapplications for learning/classification of geometries in a shapefile. . .

http://www.cnes.fr





Conclusion

A bit of history

2008 2010 2012 2014

5.0.0

4.4.

04.

2.0

4.0.0

3.20

.0

3.18

.0

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Key steps

4.0.0 Compatible with ITK 4.0, and more functions

4.2.0 speed-up in Haralick textures calculation, enhancement of the opticalcalibration framework, RPC coefficients for sensor modeling can nowbe read and written from/to GeoTIFF RPC tags. . .

5.0.0 In-dept changes in the way Orfeo ToolBox is organized and builds

http://www.cnes.fr





Conclusion

Some numbers

2008 2010 2012 2014

5.0.0

4.2.

0

4.0.0

3.20

.0

3.18

.0

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Lines of code

http://www.cnes.fr





Conclusion

Some numbers

2008 2010 2012 2014

5.0.0

4.2.

0

4.0.0

3.20

.0

3.18

.0

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Commits per month

http://www.cnes.fr





Conclusion

Some numbers

2008 2010 2012 2014

5.0.0

4.2.

0

4.0.0

3.20

.0

3.18

.0

3.16

.0

3.14

.0

3.12

.0

3.10

.0

3.8.

03.

6.0

3.4.

03.

2.0

3.0.0

2.8

2.6.

02.

4.0

2.2.

0

2.0.0

1.6.

01.

4.0

1.2.

0

1.0.0

Sourceforge downloads

http://www.cnes.fr





Conclusion

Outline

Introduction

Back in 2006

Key characteristics



Conclusion

http://www.cnes.fr





Conclusion

Build on top of other open source IP software

Motivations

I Interfaces seamlessly with other IP and RS open-source software. . .

I Reuse is powerful

I Increase the number of functions

I Combine tools to create hybrid data pipeline

OTB backbone

I ITK : OTB data processing schema based on ITK pipeline

I GDAL to read/write raster/vector data

I OSSIM sensor modelling and metadata support

I OpenCV and LibSVM provide machine learning algorithms

I MuParser and MuParserX powerful parsing of mathematical expression(bandmath)

http://www.cnes.fr



www.itk.org

www.gdal.org

www.ossim.org

www.opencv.org

www.libsvm.org

www.muparser.org

www.muparserx.org



Conclusion

Compatible (and available) on multiple plateforms

GoalI Compile with recent versions of :

I GCCI ClangI MinGWI Visual Studio. . .

I Binary packages available :I Ubuntugis repository (GIS and IP

software for Ubuntu)I Experimental Debian packagesI Available in OSGeo4W (OSGeo tools

on Windows)I Binary installers and Port for Mac

OSX. . .

Number of OTB downloads on Sourceforge per Operating System

http://www.cnes.fr





Conclusion

Flexibility, scalability : Pipeline, Streaming and multithreading

Pipeline data model

Streaming

source : http://www.aosabook.org/en/itk.html

http://www.cnes.fr



http://www.aosabook.org/en/itk.html



Conclusion

Behind the scene

source : http://www.aosabook.org/en/itk.html

http://www.cnes.fr



http://www.aosabook.org/en/itk.html



Conclusion

(Near) bleeding-edge techniques

I Try to keep track of up-to-date information about the latest developments,exchanging ideas, identifying future trends, and making networking

I Reference implementation of algorithms based on publications

I e.g. : morphological profil,MeanShift segmentation,Haralick textures,SURFkeypoints. . .

I Reference implementation contributes by authors with their publications. e.g. :Large Scale MeanShift 2, bayesian fusion 3, object detection . . .

2. Michel, J. ; Youssefi, D. ; Grizonnet, M., ”Stable Mean-Shift Algorithm and Its Application to the Segmentationof Arbitrarily Large Remote Sensing Images,” Geoscience and Remote Sensing, IEEE Transactions on , vol.53, no.2,pp.952,964, Feb. 2015

3. J. R. Dominique Fasbender and P. Bogaert. Bayesian data fusion for adaptable image pan- sharpening. IEEETransactions on Geoscience and Remote Sensing, 46(6) :1847–1857, 2007. 13.2

http://www.cnes.fr





Conclusion

How OTB is develop ?

I Distributed version control : Git (migration from Mercurial in July 2015)

I C++ and CMake(CTest, CDash)

I Test driven development (TDD)

I Agile (scrum)

I Continuous integration and packaging

Every day, almost 3000 tests are compiled, launched on 16 different configurations !

http://www.cnes.fr



http://dash.orfeo-toolbox.org/index.php?project=OTB



Conclusion

How to eat the OTB sandwich ?

Write your own codeFlexible, access to full API, requires C++ knowledge

Use the applicationsHigh level functions (e.g. segmentation), callable from CLI, Qt, Python, can beextended

Use Monteverdi2Visualization, data management, Access to all applications

http://www.cnes.fr





Conclusion

The applications : write it once, use everywhere

I 75 applications are shipped with OTB

I 1 application = 1 dynamic library(plugin)

I Applications are auto-descriptive andauto-documented

I Applications can be extended outsideof OTB

I Several plugins players :I Command-lineI Qt auto-generatedI Python

I Applications are meant for integrationin external systems

http://www.cnes.fr





Conclusion

Applications : command-line invocation$ otbcli_OrthoRectification

ERROR: Waiting for at least one parameter...

This is the OrthoRectification application, version 5.0.0

This application allows to ortho-rectify optical images from supported sensors.

Complete documentation: http://www.orfeo-toolbox.org/Applications/OrthoRectification.html

Parameters:

-progress <boolean> Report progress

MISSING -io.in <string> Input Image (mandatory)

MISSING -io.out <string> [pixel] Output Image [pixel=uint8/uint16/int16/uint32/int32/float/double] (default value is float) (mandatory)

-map <string> Output Cartographic Map Projection [utm/lambert2/lambert93/wgs/epsg] (mandatory, default value is utm)

-map.utm.zone <int32> Zone number (mandatory, default value is 31)

-map.utm.northhem <boolean> Northern Hemisphere (optional, off by default)

-map.epsg.code <int32> EPSG Code (mandatory, default value is 4326)

-outputs.mode <string> Parameters estimation modes [auto/autosize/autospacing/outputroi/orthofit] (mandatory, default value is auto)

MISSING -outputs.ulx <float> Upper Left X (mandatory)

MISSING -outputs.uly <float> Upper Left Y (mandatory)

MISSING -outputs.sizex <int32> Size X (mandatory)

MISSING -outputs.sizey <int32> Size Y (mandatory)

MISSING -outputs.spacingx <float> Pixel Size X (mandatory)

MISSING -outputs.spacingy <float> Pixel Size Y (mandatory)

-outputs.lrx <float> Lower right X (optional, off by default)

-outputs.lry <float> Lower right Y (optional, off by default)

-outputs.ortho <string> Model ortho-image (optional, off by default)

-outputs.isotropic <boolean> Force isotropic spacing by default (optional, on by default)

-outputs.default <float> Default pixel value (optional, on by default, default value is 0)

-elev.dem <string> DEM directory (optional, off by default)

-elev.geoid <string> Geoid File (optional, off by default)

-elev.default <float> Default elevation (mandatory, default value is 0)

-interpolator <string> Interpolation [bco/nn/linear] (mandatory, default value is bco)

http://www.cnes.fr





Conclusion

Applications : auto-generated Qt invocation (“Parameters tab”)

http://www.cnes.fr





Conclusion

Applications : auto-generated Qt invocation (“Documentation tab”)

http://www.cnes.fr





Conclusion

Applications : Python wrapping

#!/usr/bin/python

# Import the otb applications package

import otbApplication

# The following line creates an instance of the OrthoRectification application

OrthoRectification = otbApplication.Registry.CreateApplication("OrthoRectification")

# The following lines set all the application parameters:

OrthoRectification.SetParameterString("io.in", "QB_TOULOUSE_MUL_Extract_500_500.tif")

OrthoRectification.SetParameterString("io.out", "QB_Toulouse_ortho.tif")

# The following line execute the application

OrthoRectification.ExecuteAndWriteOutput ()

http://www.cnes.fr





Conclusion

Monteverdi2 : visualization

http://www.cnes.fr





Conclusion

Monteverdi2 : processing

http://www.cnes.fr





Conclusion

OTB in Quantum GIS

http://www.cnes.fr





Conclusion

Outline

Introduction

Back in 2006

Key characteristics



Conclusion

http://www.cnes.fr





Conclusion

Incomplete list of OTB functions

Pre-processing

I Radiometric calibration, orthorectification, resampling (raster and vector),pan-sharpening, stereo rectification. . .

I Sensor supported : Pleiades, SPOT6, SPOT5, Digital Globe satellites

I Geometric models (thanks to OSSIM), support for DEM (SRTM or GeoTIFF)

Images and vector manipulation

I Formats supported by GDAL (raster and vector), conversion raster/vector

I Region of interest extraction, of spectral bands, concatenation or splitting. . .

I Band math, color mapping, contrast enhancement

I Linear filtering, Mathematical morphology

http://www.cnes.fr





Conclusion

(Incomplete) List of OTB functions

Feature extraction

I Edge detection, scale-invariant feature transform, lines, corners

I Radiometric indices, textures (Haralick, SFS, PanTex)

I Local statistics (Flusser moments, Histogram of Oriented Gradient)

I Keypoints matching (SIFT, SURF. . . )

Change detection

I Classic methods with image metrics comparison

I Multivariate Alteration Detector

Dimensionality reduction, hyperspectral processing

I PCA, NAPCA, ICA, MAF. . .

I Dimension estimation, endmembers extraction, Vertex Component Analysis(VCA)

http://www.cnes.fr





Conclusion

Incomplete list of OTB functions

Segmentation

I Segmentation algorithms : Connected Components, MeanShift,Watershed. . .

I Methods to apply those algorithms on large dataset

I Vector or raster representation which allow Object Based Image Analysis

Classification

I 9 supervised methods available (including SVM and Random Forest)

I Fusion and regularization of classifications

I K-Means clustering or Kohonen maps

I Object classification (from a segmentation)

http://www.cnes.fr





Conclusion

Outline

Introduction

Back in 2006

Key characteristics



Conclusion

http://www.cnes.fr





Conclusion

Modular architecture (inspired by ITK 4.x)

What has changed ?

I Organize the code into conceptual cohesive groups :I OTB 4.4 : 1672 files in 26 directoriesI OTB 5.0 : 1627 files in 124 modules divided in 16 groups

I Modules contain : tests, source code, applications are grouped

I Each module can be (de)activate, with automatic dependencies resolutions

Advantages ?

I Third part dependencies are integrated as modules and can be excluded

I Lots of CMake magic (less code for configuration, better support)

I Doxygen API documentation follows the new code organization (easier to findclass info)

I Facilitate external contributions with powerful mechanisms call remote module

http://www.cnes.fr



https://www.orfeo-toolbox.org/external-projects/



Conclusion

Contribute a module

ITK schema

How it works for us ?Good start ! Already 5 remote modules contributed seehttps://www.orfeo-toolbox.org/external-projects/

http://www.cnes.fr



https://www.orfeo-toolbox.org/external-projects/



Conclusion

Superbuild : installing OTB has never been so easy

Before in OTB 4.4

I Some of the OTB third part dependencies could be build internally

I External source code was integrated in OTB source tree (not a good idea)

In OTB 5.0, on Superbuild !

I No more third party library sources integrated in OTB

I External project called superbuild which allows todownload/configure/build/install OTB and all dependencies in one pass !

I Allow to build OTB on an (almost) empty platform (CMake, gcc, zlib, curl), andeverything is automatic. . .

I There is also an offline mode which does not require Internet

http://www.cnes.fr





Conclusion

Project Steering Committee : OTB governance structure

I Open governance

I High level guidance and coordination for the ORFEO ToolBox

I Animation of OTB community, communication, orientation of the project

I Everyone can participate

I All members have equal standing and voice in the PSC (1 member = 1 vote)

I Proposals are written up and submitted on the otb-developers mailing list fordiscussion and voting

I Status and decision process are public 4

I Note that the PSC is not a legal entity !

4. http://wiki.orfeo-toolbox.org/index.php/Project_Steering_Committee

http://www.cnes.fr



http://wiki.orfeo-toolbox.org/index.php/Project_Steering_Committee



Conclusion

Outline

Introduction

Back in 2006

Key characteristics



Conclusion

http://www.cnes.fr





Conclusion

How many users ?

Hard to tell. . .

I ≈ 600 members on the otb-users list

I Between 100 and 150 mails by months

I ≈ 100 members on the developers list

I ≈ 118 user accounts on the bugtracker

I ≈ 50 contributors in thedocumentation

I ≈ 3400 downloads for OTB 5.0 onSourceForge(released June 1, 2015).

http://www.cnes.fr





Conclusion

Success storiesI OTB has been useful for (some) ORFEO/RTU

users !

I Several training courses (3/5-day courses)given in France, Belgium, Madagascar,UNESCO, Hawaii,Finland. . .

I OTB has successfully processed 619 Pleiadesimages on RTU web site

I OTB provides many useful RS functions in onesingle tool

I OTB is/was the only open-source supportingPHR images (thanks to OpenJPEG)

I OTB equals or beats state-of-the-art tools (osand maybe $$) on some points :

I band calculatorI tile-wise segmentation of full imageryI full scene classification with a range of machine

learning algorithmsI bridges between RS and SIG . . .

I Beyond Orfeo, OTB is already used in severalprojects and software

I OSGeo incubation in progress

Thematic map from OTB segmentation, B.Mougenot - IRD

http://www.cnes.fr





Conclusion

Projects and software using OTB

I OTB applications are availablethrough QGIS processing framework

I TerreImage : Educational software forsatellite image analysis

I Gnorasi Software (National TechnicalUniversity of Athens)

I Vahine project (hyperspectralprocessing of astrophysics),IPAG

I Geosud project(IRSTEA)

I OTB is part of some components forSentinel-2 and Venus ground segment(CNES and ESA)

I TCM research program (ETS Quebec)

I ESA Sentinel2 agri

The Gnorasi software

http://www.cnes.fr





Conclusion

Support/Help/Contribute

General resourcesSite web www.orfeo-toolbox.org

Wiki wiki.orfeo-toolbox.org

Blog blog.orfeo-toolbox.org

Documentation and helpDoxygen http://www.orfeo-toolbox.org/doxygen/

Guides Software Guide and CookBook (remote sensing recipes)

Users mailing list [email protected]

Developers mailing list [email protected]

Follow-upLook at the code ? git.orfeo-toolbox.org

Find a bug ? bugs.orfeo-toolbox.org

Agile ? scrum.orfeo-toolbox.org

Weather ? dash.orfeo-toolbox.org

http://www.cnes.fr




wiki.orfeo-toolbox.org

blog.orfeo-toolbox.org

http://www.orfeo-toolbox.org/doxygen/

git.orfeo-toolbox.org

bugs.orfeo-toolbox.org

scrum.orfeo-toolbox.org

dash.orfeo-toolbox.org



Conclusion

Thank you ! Any questions ?

http://www.cnes.fr



general presentation of otb

Technology