The Orfeo Toolbox provides tools for objects detection in images using multiple approaches including radiometry, textures and contours, and segments. It allows users to perform tasks like opening images, extracting features, applying filters, and connected component segmentation. The toolbox can also be used for object labelling through mean shift segmentation, selecting training samples, and reclassifying uncertain samples.
2. Objects detection
Type of objects searched
Buildings
Roads
Rivers
Tents
Boats, planes
Etc.
orfeo-toolbox.org
2
3. Objects detection
Multiples approaches
Radiometry (if the object has a homogenous specific radiometry)
➢ Thresholds on radiometry values
➢ Spectral angle
Textures and contours
➢ Local correlation, local mutal information
Contours
➢ HOG (Histogram of Gradient)
Segments (radiometry + contours)
➢ Computation of segments features and selection with SVM for example
orfeo-toolbox.org
3
4. Use case 1 : objects detection
Menu File > Open
./output_fus1.tif
Menu Filtering > Feature extraction
Original data
Angle spectral on 1 point of the object
Angle spectral on 2nd point of the object
Others as you want
Menu Filtering > Band Math
if(min(im1b2, im1b3)<0.15,im1b1,0)
Try yours...
orfeo-toolbox.org
4
5. Use case 2 : connected component segmentation (1/3)
Menu File > Open
./output_fus1.tif
Menu Filtering > Connected Segmentation Module
Mask Formulas
➢ ndvi(b3,b4)<0.5
➢ (b1>300) * (b2 >400)
Display : Mask output. Click on Update.
Display : Masked image. Click on Update.
Connected component formulas
➢ distance < 100 (Distance between adjacent pixels is less than 100)
➢ (p1b1>300) * (p1b2 >200)
Display : Segmentation output. Click on Update.
orfeo-toolbox.org
5
6. Use case 2 : connected component segmentation (2/3)
Objects min area
10
30
Display : Relabelling with small objects. Click on Update.
Object analysis formula
➢ (SHAPE_RegionRatio > 0.4) and (SHAPE_SizeOnBorder==0)
Display : Filter output. Click on Update.
In the viewer, right click on a segment : features are printed in the main
window
=> Now try your formulas !
orfeo-toolbox.org
6
7. Use case 2 : connected component segmentation (3/3)
Menu visualization > Viewer
Raster : reader0
Vector data to display : OuputVectorData
OutputVectorData
Right click > Export dataset
houses.shp (shapefile->SIG)
orfeo-toolbox.org
7
9. Use case 3 : object labelling (1/2)
Menu File > Open
./output_fus1.tif
Menu Filtering > Mean shift module
Spatial radius : 5, Spectral value : 30, Min region : 50
Menu learning > Object labelling
Image to classify : Reader0
Labeled Image : MeanShift->LabeledImage
Add 1 class for object. Right click on twice on segments to select them.
Add 2nd class for non object. Right click on twice on segments to select
them.
Tab Learning. Classify.
orfeo-toolbox.org
9
10. Use case 3 : object labelling (2/2)
Tab Objects.
➢ Click on Sample (list of segments ID with features near the threshold
of SVM classification)
➢ Select one segment ID. Select the other class. Right click to
redefine.
Tab Learning. Classify.
Save/quit
orfeo-toolbox.org
10