OBIA

Object based image analysis treats an image as a collection of regions rather than a grid of independent pixels. It matters as soon as your objects of interest span many pixels: on 30 centimetre aerial imagery, a roof is thousands of pixels, and classifying each of them separately produces a speckled map that no smoothing will really fix.

OBIA proceeds in three stages, and the 33 tools of this sub toolbox map onto them.

1. Segment. Group neighbouring pixels into homogeneous regions.
2. Describe. Compute features per region: spectral statistics, shape, texture, neighbours.
3. Classify. Label the regions, by rules you write or by a trained model.

Start here

Obia Pipeline Basic

The whole chain in one run: segmentation, cleanup of small regions, feature computation and classification. Start here to see what an OBIA result looks like on your imagery, then rebuild it stage by stage when you need control.

Segment Slic Superpixels

The segmentation to try first. SLIC groups pixels into compact, similarly sized superpixels by clustering on colour and position together. Fast, predictable, and it gives regions of a size you control.

Segment Graph Felzenszwalb takes a different approach, growing regions that follow the actual structure of the image, so their sizes vary a lot. Better at following real boundaries, harder to tune.

Segment Multiresolution Hierarchical produces several nested levels at once, coarse and fine, which is the classic eCognition style workflow: classify buildings at the coarse level and roof elements at the fine one.

Segment Scale Parameter Optimizer

The answer to the recurring question, which scale to segment at. It tries a range of scale parameters and reports which one gives the most coherent objects, so the choice comes from the data instead of from trial and error.

The Object Features tools

Once segmented, each object needs describing. Four tools, and you usually run several.

Object Features Spectral Basic gives the statistics of each band inside the object: mean, standard deviation, extremes. The core features.

Object Features Shape Basic gives geometry: area, perimeter, compactness, elongation. This is what separates a road from a field of the same colour.

Object Features Texture Glcm Basic gives texture from the grey level co occurrence matrix, which separates surfaces that share a colour but not a grain.

Object Features Context Neighbors and Object Features Topology Relations describe each object by its surroundings: how many neighbours, which classes, how much shared boundary. This is how you encode a rule like “a swimming pool is water surrounded by a garden”.

Classifying objects

Classify Objects Random Forest is the default, trained on labelled example segments, exactly like its pixel based counterpart.

Classify Objects Rules Basic takes the opposite approach: you write explicit rules over the features, such as brightness below a value and compactness above another. Transparent, auditable, and often preferred when the result has to be defended.

All the tools

Segmentation

Tool	What it does
Segment Slic Superpixels	Compact superpixels from clustering on colour and position.
Segment Graph Felzenszwalb	Graph based segmentation following the image structure.
Segment Multiresolution Hierarchical	Several nested segmentation levels, coarse to fine.
Segment Watershed Markers	Marker driven watershed segmentation, separating touching objects.
Image Segmentation	Seeded region growing over a multiband stack.
Segment Scale Parameter Optimizer	Searches the scale parameter giving the most coherent objects.
Build Object Hierarchy Multiscale	Builds a hierarchical network of objects by iterative aggregation.

Cleaning and editing segments

Tool	What it does
Segments Merge Small Regions	Merges undersized segments into their neighbours.
Segments Split Low Cohesion	Re segments the least homogeneous objects at a finer scale.
Objects Enforce Min Mapping Unit	Enforces a minimum mapping unit by merging segments below it.
Objects Boundary Refinement Pro	Refines object boundaries by iterative cleanup.
Segments To Polygons	Converts the segment raster into vector polygons, for editing or export.
Polygons To Segments	Converts edited polygons back into a segment raster, keeping the object ids.
Evaluate Segmentation Quality Pro	Diagnostics on object count, size distribution and homogeneity.

Object features

Tool	What it does
Object Features Spectral Basic	Per band statistics inside each object.
Object Features Shape Basic	Area, perimeter, compactness, elongation and other shape descriptors.
Object Features Texture Glcm Basic	Texture descriptors per object.
Object Features Context Neighbors	Neighbour counts, shared boundaries and dominant adjacent classes.
Object Features Topology Relations	Graph topology features, such as object degree.

Classification and evaluation

Tool	What it does
Classify Objects Random Forest	Trains a random forest on labelled segments.
Classify Objects Svm	Support vector machine classification of objects.
Classify Objects Ensemble Pro	An ensemble configuration tuned for accuracy.
Classify Objects Rules Basic	Explicit feature, operator and threshold rules.
Classify Objects Rules Hierarchical	Rule based classification across a segmentation hierarchy.
Propagate Labels Across Hierarchy	Passes coarse level labels down to the finer child objects.
Object Class Probability Maps	Per class probability rasters, for uncertainty analysis.
Object Uncertainty Diagnostics Pro	Aggregate uncertainty diagnostics from those probabilities.
Evaluate Object Classification Accuracy	Accuracy metrics against reference objects.

Workflow support

Tool	What it does
Obia Pipeline Basic	The full chain: segmentation, cleanup, features and classification.
Obia Batch Orchestrator Pro	Runs several OBIA jobs in one request.
Obia Audit Report Pro	Audit report over the artefacts of an OBIA workflow.
Image Stack Profile	Per point profiles across an ordered raster stack.
Image Slider	Interactive side by side comparison of two images.

Notes

Segmentation decides everything. No classifier recovers from objects that cut across real boundaries. Spend your time here, and check the segments visually before computing a single feature.

Objects carry many features. That is the point. A pixel has only its spectra, an object also has a shape, a texture, a size and neighbours, which is why OBIA distinguishes classes that pixel based methods confuse.

Vector output. Segments To Polygons gives you ordinary polygons, which you can then style, edit and export like any vector layer.