Scribd
Upload
12 views9 pages

Region Segment

WLC

Uploaded by

D H Makwana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views9 pages

Region Segment

WLC

Uploaded by

D H Makwana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

REGION BASED SEGEMENTATION

The objective of Segmentation is to partition an image into


regions. The region-based segmentation techniques find the
regions directly.

Extract those regions in the image whose pixel's have some


common property in terms of any one of these:

* Pixel Intensity
* Pixel Colour
* Texture
* Range or Depth (for laser images)
* Temperature (for thermal images)
* Echo ( for ultrasound sound images)
* etc.

Generally There are three methods for Region Based


Segmentation.

* Region Growing
* Region Splitting
* Region Splitting and Merging
Region-based Segmentation: Preliminaries
Image Segmentation can be viewed as a process that
partitions an image R into n subregions R1, R2,
R3, ...... Rn.
such that
* Union of Ri (for i = 0 to n) is equal to image R.
Which means whole image is segmented or segmentation
is complete.

* Ri is a connected region.
The points in a region are connected either with four
connectivity or eight connectivity.

* The regions Ri (for i = 0 to n) are disjoint.


For any two regions Ri and Rj there is no common point

* A predicate (a common property) is true for all the Ri


(for i = 0 to n) regions.

* The predicate is false for the union of any two regions.


It means any two regions (combined together) are
different on the basis of the predicate.
REGION GROWING

The procedure is also known as Region Merging. It is a


process which groups adjacent image pixel's or sub-regions
into larger regions which meet one or more than one
properties.

These properties are known as region growing/merging


criteria. The most common criteria is similarity in some pixel
property (such as intensity, texture, colour etc.).

Simple Method: Pixel Aggregation


* Starts form a set of seed points
* These seeds are grown into regions that have similar
properties

Problems with Region Growing Process

* How to select seed points ?


* How many seed points to start with ?
* When and where to stop region growing ?
REGION SPLITTING

Region splitting is the opposite approach to region growing or


merging.

* Starts with the whole image as a single region

* Divide the image successively into smaller regions


until each smaller region meets the similarity criterion

* Splitting stops when the sub-regions satisfies the


criteria

Criterion for Similarity in Intensity


Region Uniformity

The Computation involved at the region level are the


calculation of intensity mean, variance or standard
Deviation etc. for each region and sub-regions from
pixel's to image level.
Implementation of Region Uniformity Criterion
A number of alternatives, depending on the amount of
computation and the effectiveness of the criterion.

* Involves only the Intensity Mean of sub-regions and the


parent regions.
Region Merging: Continue merging until the intensity
mean of sub-regions and merged region are nearly equal.
Region Splitting: Stop
splitting when intensity mean of the parent and sub-
regions become nearly equal.

* Involves the intensity-based Standard Deviation of sub-


regions and parent regions. A uniformity threshold in terms of
standard deviation is used.
Region Merging: Continue merging if the intensity mean
of the sub-regions are nearly equal and standard deviation
of the parent (merged) region is below the uniformity
threshold. Region Splitting: Stop splitting
when the standard deviation of the regions falls below the
uniformity threshold.

* Involves intensity Variance of the sub-regions and parent


region. Recursive calculation of variance is economical than
direct pixel based calculation of standard deviation.
Region Merging and Splitting

For Merging Only:


* One needs to start from pixel's level and it requires lot
of computation.
* The most suitable seeds are difficult to find.
* The segmentation results depend on the number of seeds
and the location of seeds in the image.
* Useful when there are lot of uniform regions in the
image.

For Splitting only:


* One starts from the whole image and starts splitting.
* The segmentation results depends on how to split in
terms of the shape of the partitioned regions.
* The computation depends on the shape of uniform
regions and how one is splitting.
* Useful when there are only a few and regular uniform
regions in the image. Otherwise, lot of computation.

A combination of merging and splitting will enjoy the


benefits of both methods.
Region Merging and Splitting
(Cont.)

Quadtree Representation

A hierarchical data structure for representing images. The


quadtree representation is based on the successive sub-
division of an image into quadrants.

* It is a tree of out-degree four.

* The root node represents the whole image of 2n


2n pixel's where n is the maximum number of levels of
the quadtree.

* The four sons of the root represent the four quadrants


of an image.

* The terminal nodes either represent the uniform region


or pixel's.

When the shape of the partitioned regions can be taken as


squared, quadtrees are best suited for region splitting based
segmentation.
Region Merging and Splitting
(Cont.)

When a quadtree is employed for Region Splitting and if the


uniform region is neither a squared one nor perfectly aligned
with the image quadrants or sub-quadrants.
* The segmentation will not be complete
* Extra uniform regions will be extracted which requires
further merging.

Outcome is the Region Splitting and Merging Algorithm:


Step 1: Start from any quadtree based partitions (R1, R2,
R3, ........ Rn) of an image.

Step 2: Split a region Ri if it satisfies the splitting (non-


uniformity) criterion.

Step 3: Merge any adjacent regions Rj and Rk if they


satisfy the merging (uniformity) criterion.

Step 4: Continue until no further merging or splitting is


possible.
Region Merging and Splitting
(Cont.)

A number of variations of the Merging and Splitting


approach exist.
The following procedure only employs the quadtree structure
for both merging and splitting.

* Split the image initially into a set of squared blocks.

* Perform further splitting if the splitting criterion is


satisfied.

* Merge only the groups of four blocks which are


brothers/sisters and satisfy the merging criterion.

* If no further merging is possible, then perform a final


merge step for different sized regions satisfying the
merging criterion (similar to Step 3).

* Terminate the procedure.

You might also like