Object Based Image Segmentation

Author: O.Conrad (c) 2014
Menu: Imagery | Segmentation

Description

This Object Based Image Segmentation tool chain combines a number of tools for an easy derivation of geo-objects as polygons and is typically applied to satellite imagery. Segmentation is done using a 'Seeded Region Growing Algorithm'. Optionally the resulting polygons can be grouped by an unsupervised classification (k-means cluster analysis) or supervised classification (needs classified feature samples as additional input), both is performed on the basis of zonal feature grid statistics for each polygon object.

References

Adams, R. & Bischof, L. (1994): Seeded Region Growing. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.16, No.6, p.641-647.
Bechtel, B., Ringeler, A. & Boehner, J. (2008): Segmentation for Object Extraction of Trees using MATLAB and SAGA. In: Boehner, J., Blaschke, T., Montanarella, L. [Eds.]: SAGA - Seconds Out. Hamburger Beitraege zur Physischen Geographie und Landschaftsoekologie, 19:59-70. online.

Parameters

	Name	Type	Identifier	Description	Constraints
Input	Features	grid list, input	`FEATURES`	-	-
Input	Training Samples	table, input	`SAMPLES`	Training samples for supervised classification. Provide a class identifier in the first field followed by sample data corresponding to the selected feature attributes	-
Output	Segments	shapes, output	`OBJECTS`	-	-
Options	Grid System	grid system	`GRID_SYSTEM`	-	-
	Normalize	boolean	`NORMALIZE`	-	Default: 0
	Band Width for Seed Point Generation	floating point number	`SEEDS_BAND_WIDTH`	Increase band width to get less seed points.	Default: 2.000000
	Neighbourhood	choice	`RGA_NEIGHBOUR`	-	Available Choices: [0] 4 (Neumann) [1] 8 (Moore) Default: 0
	Distance	choice	`RGA_METHOD`	-	Available Choices: [0] feature space and position [1] feature space Default: 0
	Variance in Feature Space	floating point number	`RGA_SIG_1`	-	Minimum: 0.000000 Default: 1.000000
	Variance in Position Space	floating point number	`RGA_SIG_2`	-	Minimum: 0.000000 Default: 1.000000
	Similarity Threshold	floating point number	`RGA_SIMILARITY`	-	Minimum: 0.000000 Default: 0.000000
	Generalization	integer number	`MAJORITY_RADIUS`	Applies a majority filter with given search radius to the segments grid. Is skipped if set to zero.	Default: 1
	Classification	choice	`CLASSIFICATION`	-	Available Choices: [0] none [1] cluster analysis [2] supervised classification Default: 0
	Split Distinct Polygon Parts	choice	`SPLIT_POLYGONS`	-	Available Choices: [0] no [1] yes Default: 0
	Number of Clusters	integer number	`NCLUSTER`	-	Default: 10
	Method	choice	`CLASSIFIER`	-	Available Choices: [0] Binary Encoding [1] Parallelepiped [2] Minimum Distance [3] Mahalanobis Distance [4] Maximum Likelihood [5] Spectral Angle Mapping Default: 0

Command Line


Usage: saga_cmd imagery_segmentation obia [-FEATURES ] [-NORMALIZE ] [-OBJECTS ] [-SEEDS_BAND_WIDTH ] [-RGA_NEIGHBOUR ] [-RGA_METHOD ] [-RGA_SIG_1 ] [-RGA_SIG_2 ] [-RGA_SIMILARITY ] [-MAJORITY_RADIUS ] [-CLASSIFICATION ] [-SPLIT_POLYGONS ] [-NCLUSTER ] [-CLASSIFIER ] [-SAMPLES ]
  -FEATURES:           	Features
	grid list, input
  -NORMALIZE:          	Normalize
	boolean
	Default: 0
  -OBJECTS:            	Segments
	shapes, output
  -SEEDS_BAND_WIDTH:	Band Width for Seed Point Generation
	floating point number
	Default: 2.000000
  -RGA_NEIGHBOUR:      	Neighbourhood
	choice
	Available Choices:
	[0] 4 (Neumann)
	[1] 8 (Moore)
	Default: 0
  -RGA_METHOD:         	Distance
	choice
	Available Choices:
	[0] feature space and position
	[1] feature space
	Default: 0
  -RGA_SIG_1:       	Variance in Feature Space
	floating point number
	Minimum: 0.000000
	Default: 1.000000
  -RGA_SIG_2:       	Variance in Position Space
	floating point number
	Minimum: 0.000000
	Default: 1.000000
  -RGA_SIMILARITY:  	Similarity Threshold
	floating point number
	Minimum: 0.000000
	Default: 0.000000
  -MAJORITY_RADIUS:    	Generalization
	integer number
	Default: 1
  -CLASSIFICATION:     	Classification
	choice
	Available Choices:
	[0] none
	[1] cluster analysis
	[2] supervised classification
	Default: 0
  -SPLIT_POLYGONS:     	Split Distinct Polygon Parts
	choice
	Available Choices:
	[0] no
	[1] yes
	Default: 0
  -NCLUSTER:           	Number of Clusters
	integer number
	Default: 10
  -CLASSIFIER:         	Method
	choice
	Available Choices:
	[0] Binary Encoding
	[1] Parallelepiped
	[2] Minimum Distance
	[3] Mahalanobis Distance
	[4] Maximum Likelihood
	[5] Spectral Angle Mapping
	Default: 0
  -SAMPLES:            	Training Samples
	table, input