7/1/2014 1 Image segmentation Assoc. Prof. KadimTaşdemir Antalya International University [email protected]Image segmentation • Image segmentation is a partitioning of an image into parts using image attributes such as pixel intensity, spectral values, and/or textural properties. Image segmentation produces an image representation in terms of edges and regions of various shapes and interrelationships. • Segmentation algorithms: – region growing/merging, – watershed segmentation, – edge-based segmentation, – probability-based image segmentation, – fractal net evolution approach (FNEA), – and many more… • Multi-scale image segmentation Jensen, 2005
22
Embed
Image segmentation · Image segmentation • Image segmentation is a partitioning of an image into parts using image attributes such as pixel intensity, spectral values, and/or textural
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Image segmentation• Image segmentation is a partitioning of an
image into parts using image attributes such as pixel intensity, spectral values, and/or textural properties. Image segmentation produces an image representation in terms of edges and regions of various shapes and interrelationships.
• Segmentation algorithms: – region growing/merging,
– watershed segmentation,
– edge-based segmentation,
– probability-based image segmentation,
– fractal net evolution approach (FNEA),
– and many more…
• Multi-scale image segmentation
Jensen, 2005
7/1/2014
2
• Division of an image into regions so that– the whole scene is covered by regions (spatially continuous,
exhaustive segmentation)
– the regions do not overlap
– the regions are homogeneous within themselves
– the homogeneity criteria of neighboring regions differ
• Region (token):– aggregate of pixels grouped together (directly or indirectly)
• Homogeneity as overarching principle– ‘relatively’ homogeneous regions reflect better the
‘Neardecomposability’ of natural systems
– High heterogeneity creates boundary to neighboring patches, low remaining heterogeneity within patches
– Homogeneity criterion: grey value, color, texture, form, altitude, etc.
• Initially: image as one object � division into 4 parts, if H does
not apply
• Resulting quadtree structure
• Merge of homogenous quadtree areas
• Bottom up region merging technique– Starting with each pixel being a region
– A pair of regions is merged into one region, each merge having a merging cost (degree of fitting)
– Objects are merged into bigger objects as long as the cost is below a ‘least degree of fitting’(scale parameter)= the merge fulfils the homogeneity criterion
– Starting points for merging distributed with maximum distance
– Pair wise clustering process considering smallest growth of heterogeneity
• Establishing segmentation levels on several scales using different scale parameters (e.g. 2nd level based on 1st level: larger scale parameter results in larger objects consisting of the objects of the 1st level)
• The scale parameter is an abstract value to determine the maximum possible change of heterogeneity caused by fusing several objects.
– The scale parameter is indirectly related to the size of the created objects.
– The heterogeneity at a given scale parameter is directly linearly dependent on the object size. Homogeneous areas result in larger objects, and heterogeneous areas result in smaller objects.
– Small scale � small objects, large scale � large objects. This refers to Multiresolution image segmentation.
• Color is the spectral feature
• Shape includes compactness and smoothness which are two geometric features that can be used as "evidence."
– Smoothness describes the similarity between the image object borders and a perfect square.
– Compactness describes the "closeness" of pixels clustered in a object by comparing it to a circle
7/1/2014
6
Color and shape
These two criteria are used to create image objects (patches) of relatively
homogeneous pixels in the remote sensing dataset using the general
segmentation function (Sf):
where the user-defined weight for spectral color versus shape is 0 < wcolor < 1.
If the user wants to place greater emphasis on the spectral (color)
characteristics in the creation of homogeneous objects (patches) in the
dataset, then wcolor is weighted more heavily (e.g., wcolor = 0.8).
Conversely, if the spatial characteristics of the dataset are believed to be
more important in the creation of the homogeneous patches, then shape
should be weighted more heavily.
( ) shapecolorcolorcolorf hwhwS ⋅−+⋅= 1
So the color criterion is computed as the weighted mean of all
changes in standard deviation for each band k of the m bands of
remote sensing dataset. The standard deviation sk are weighted by
the object sizes nob (i.e. the number of pixels) (Definiens, 2003):
where mg means merge (total pixels in all objects 1 and 2 here).
( )[ ]2
2
1
1
1
ob
kob
ob
kob
mg
kmg
m
k
kcolor nnnwh σσσ ⋅+⋅−⋅=∑=
Spectral (i.e., color) heterogeneity (h) of an image object is computed
as the sum of the standard deviations of spectral values of each layer
(sk) (i.e., band) multiplied by the weights for each layer (wk):
kk
m
k
wh σ⋅=∑=1
Usually equal weight for all bands except you know certain band is really important
IKONOS/QuickBird/OrbView/GeoEye, and DMC/SPOT-1/-2 Imagery—Part
I: System Design and Implementation”, IEEE TGRS, 48 (3), 1299-1325, 2010
• A. Baraldi, T. Wassenaar, and S. Kay, “Operational Performance of an
Automatic Preliminary Spectral Rule-Based Decision-Tree Classifier of
Spaceborne Very High Resolution Optical Images”, IEEE TGRS, 48 (9), 3482-
3502, 2010
Satellite Image Automatic Mapper™
7/1/2014
19
Hedge detection
S. Aksoy, H. G. Akcay, T. Wassenaar, "Automatic Mapping of Linear Woody Vegetation Features in
Agricultural Landscapes Using Very High-Resolution Imagery," IEEE TGRS, 48(1), 511-522, 2010.
Hedge detection
7/1/2014
20
Orchard detection
S. Aksoy, I. Z. Yalniz, K. Taşdemir, "Automatic
Detection and Segmentation of Orchards Using Very
High-Resolution Imagery," IEEE TGRS, 50(8), 3117-
3131, 2012.
Unsupervised clustering
(CONN linkage)
k: # of clusters is set by the user
Clusters are labeled to 4 classes:
GAC, woodlands, urban/bare, water
OBIA of clustermap(for texture and spatial analysis)
Inland vegetation and urban vegetation are derived
Taşdemir et al.
COMPAG, 2012
GAC (Good agricultural condition) Assessment
K. Taşdemir, P. Milenov and B. Tapsall, “A hybrid method combining SOM-based clustering and object-based analysis for identifying land in good agricultural condition,” Computers and Electronics in Agriculture,83, 92-101, 2012.
K. Taşdemir, Pavel Milenov and Brooke Tapsall, “Topology-based hierarchical clustering of self-organizing maps,” IEEE Transactions on Neural Networks, 22 (3), 474-485, 2011.
7/1/2014
21
Taşdemir et al., COMPAG 2012
KARD
SOM + OBIA
GAC
Forests
Urban/Bare
Water
Inland veg.
Urban veg.
Table is from Taşdemir et al., COMPAG, 2012
GAC (Good agricultural condition) Assessment
7/1/2014
22
Automatic LPIS assessment• K. Taşdemir, C. Wirnhardt, “Neural network based clustering for agriculture
management”, To Appear in EURASIP Journal on Advances in Signal Processing, Special Issue on Neural Networks for Interpretation of Remotely Sensed Data, Invited Paper, 2012.
• K. Taşdemir, “Vector quantization based approximate spectral clustering of large datasets,” Pattern Recognition, 45 (8), 3034-3044, 2012.