You are on page 1of 6

Rohit Maurya et al.

/ (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES


Vol No. 5, Issue No. 2, 290 - 295

Road Extraction Using K-Means Clustering and


Morphological Operations

Rohit Maurya, Dr. Shalini Singh, Dr. P.R Gupta, Manish Kumar Sharma
Centre for Development of Advance Computing
B-30, Institutional Area, Sector- 62, Noida, India
Email: Rohitmaurya87@gmail.com
Email: shalinisingh@cdacnoida.in
Email: poonam.r.gupta@gmail.com
Email: Manisharma2009@gmail.com

T
Abstract— In this paper we proposed the method for road region merging [9] for image segmentation and road network
extraction. The road extraction involves the two main steps: the is
detection of road that might have the other non road parts like
buildings and parking lots followed by morphological operations
to remove the non road parts based on their features. We used
ES
the K-Means clustering to detect the road area and may be some
non road area. Morphological operations are used to remove the
non road area based on the assumptions that road regions are
an elongated area that has largest connected component.

Index Terms— K-Means clustering,


operations, road extraction, segmentation.
morphological

I. INTRODUCTION

T HE road extraction from digital images has drawn a


special attention in the last few decades. Numerous
A
methods has been developed which includes semi automatic
and automatic road extraction. Road extraction plays a very
important role in vehicle navigation system, urban planning,
disaster management system and traffic management system.
Semi automatic road extraction required requires user
interaction in order to extract the road where automatic
IJ

Figure 1. Roads in developing suburban area


method requires no user interaction.
In existing method of road extraction various semi extracted based upon skeleton pruning method based on
automatic and automatic methods have been developed. Karin discrete curve evaluation. Yinghua He, Hong Wang, Bo
K. Hedman, U. Stilla, G. Lisini, P. Gamba (2010) [1] has Zhang (2003) [4] has developed an algorithm composed of
used two road extractors one for rural areas and another for the segmentation of the corresponding road area based upon
urban areas. They used two steps for road extraction: first is the detected vanishing point. Anil and Natarajan (2010) [3]
line extraction followed by a smoothening and splitting step have developed the method which uses statistical region
and in another step linear features are evaluated on their merging [9] for image segmentation and road is extracted
attributes using Bayesian probability theory. Hui Kong, J.-Y. using skeleton pruning which is based on contour
Audibert, J. Ponce (2010) [2] has developed the method based partitioning. Yinghua He, Hong Wang, Bo Zhang (2003) [4]
upon the vanishing point associated with main part of road, has developed an algorithm consists of two major points:
followed by the segmentation of the corresponding road area boundaries are estimated based on the intensity image and
based upon the detected vanishing point. Anil and Natarajan road areas are detected based on the full color image. Tomoko
(2010) [3] have developed the method based upon statistical Tateyama, Zensho Nakao, Xian Yan Zeng, Yen-Wei Chen

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 290


Rohit Maurya et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES
Vol No. 5, Issue No. 2, 290 - 295

(2004) [5] has proposed a method which combines color,


texture information and shape information for segmentation Extracted roads
of high resolution images and a new direction filter which
focuses on road features having information on specific Figure 2. Flowchart of road extraction method
directionality. The method uses color and texture information
for global segmentation and shape information for local A. Image segmentation using K-Means clustering:
analysis. Onur Tuncer (2007) [6] has proposed a method
which consists of preprocessing the image via the series of Clustering is method to group similar objects into one
wavelet based filter banks and reducing the yielding data into cluster. K-means clustering [11] finds clusters such that
a single image, then utilizing a fuzzy inference algorithm to objects within each cluster are as close to each other as
find road areas. H. Mayer, I. Laptev, A. Baumgartner and C. possible, and as far from objects in other clusters as possible.
Steger (1997) [7] has used ribbon snakes not only to extract K-means clustering requires you to specify the number of
the roads eficiently, but also used to bridge gaps in the clusters to be partitioned and a distance metric to specify how
extracted roads due to shadows created by building and trees. close two objects are to each other.
Q. Zhang and I. Couloigner (2006) [8] has proposed a We used K-Means to cluster the objects into clusters using
method which starts with an image segmentation using a the Euclidean distance metric and to segment the image in

T
spectral clustering algorithm. This step focuses on the various clusters.
exploitation of the spectral information for feature extraction.
The road clusters are automatically identified using a fuzzy K-means Clustering:
classifier based on a set of predefined membership function
for road surface and the corresponding normalized digital As described by Suman Tatiraju and Avi Mehta [14]
ES
numbers in each multi spectral band. A number of shape
descriptors from the refined angular texture signature are
defined and used to reduce the misclassification between
roads and other spectrally similar objects such as parking lots,
building or crop field. Jalal (2006) [10] has proposed a
method comprise of two stages: feature extraction and fuzzy
K-Means algorithm is an unsupervised clustering algorithm
that classifies the input set of data into multiple clusters based
on their distance from each other. Based on the distance
metric specified the algorithm tries to group the input data
into various clusters . The points are clustered around
centroids µi ∀ i = 1... k which are obtained by minimizing
modeling for road identification. In the first stage mean, the objective
standard deviation, skewness and kutosis. In the fuzzy k
modeling stage, the roads are identified based on converted V = Σ Σ (xj - µi)2
features to the specific fuzzy sets in the linguistic variables. i=1 xj ϵ Si
In this paper we use the K-Means clustering based
A
segmentation to find the road cluster followed by where there are k clusters Si, i = 1, 2, . . . , k.
morphological operations to filter the area which has similar µi is the mean point of all the points xj ϵ Si.
features as the road like buildings, parking lots and crop field.
The algorithm has following steps:

II. METHODOLOGY 1. Compute the intensity distribution of the intensities.


IJ

2. Choose k centroids randomly.


The flowchart of road extraction is given in figure 2. This 3. Repeat the following steps until the cluster does not
method includes four parts: segmentation using K-Means change anymore.
clustering, road cluster identification, road cluster filtering 4. Cluster the points based on distance of their intensities
and result evaluation. from the centroid intensities.

Input Image c(i) := arg min || x(i) - µj || 2


j
Segmentation using K- Means clustering
5. Compute the new centroid or mean point for each
Road cluster identification clusters.

m
Road cluster filtering
Σ 1{ c(i) = j } x(i)
i=1
µi :=
Result evaluation m

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 291


Rohit Maurya et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES
Vol No. 5, Issue No. 2, 290 - 295

Σ 1{ c(i) = j } C. Road cluster filtering:


i=1
The road cluster which is obtained from the classification is
consists of roads, buildings, parking lots and other objects
Where k is the number of clusters to be found, i iterates over which has similar feature like a road. Further processing is
the all the intensities, j iterates over all the centroids and µi needed in order to remove the non-road areas before road
are the centroid intensities. extraction. In this work the road cluster filtering approach

T
Figure 3. Segmented image with road cluster shown
in red ES
B. Road cluster identification:

After input image segmentation, we have three clusters as


Figure 5 Extracted Roads after refinement step
an output so to make the approach fully automatic we should
know which cluster is the road cluster. Road cluster is need to
based on the concept that roads usually appear as elongated
be identified because we need the right cluster in order to
areas while the object similar to the roads are usually open
extract the roads otherwise our result we cannot get the right
areas.
output.
To identify the road cluster we used the concept that road
In this step the following operations are performed.
usually appear as elongate regions. So the cluster which has
A
the longest connected component will be the road cluster.
1. Dilation- Dilation refers to expansion i.e. adds pixels to the
boundaries of objects in an image. We use dilation to fill the
any gaps between the road pixels so that we can perform the
next step efficiently.
IJ

2. Non road area removal - In this step we use the concept


that road regions are an elongated area that has largest
connected component and the building and other non road
has relatively lesser area and is not connected to road.
This step involves-

1. Find the connected components.


2. Compute the area of each component.
3. Remove non road parts (small objects like building,
parking lots and other similar objects ).

D. Result evaluation:

Figure 4. Road cluster after segmentation The basic method of result evaluation is to compare the
results against a manually plotted, high quality reference

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 292


Rohit Maurya et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES
Vol No. 5, Issue No. 2, 290 - 295

model, such as the one shown in figure 6.


The method for evaluating the results of road extraction Quality ( Heipke et al.,1997) [12] focuses on the two main
can be done by completeness, correctness and quality measure point: (1) How complete is the extracted road, and (2) How
evaluation. The quality of output depends on how much part correct the extracted road. The completeness corresponds to
of our derived output matches with the reference data. the user’s demands (“what is missing in the output I want”),
whereas the correctness is related to the probability of an
1. Completeness and correctness extracted linear piece to be indeed a road.

The accuracy of any road extraction technique can be Completeness


summarized by measuring the completeness and correctness
of the derived road network. In order to evaluate the results
the quality measures completeness, correctness and quality as Completeness = Length of matched reference
defined in (Heipke et al., 1997) [12] are examined. The Length of reference
classification of road network is done as true positive (TP),
false negative (FN), or false positive (FP). ≈ TP
TP + FN

T
Completeness ϵ [0; 1]

ES
Figure 7. Matched extraction ( Heipke et al.,1997) [12]
A
Figure 6 Reference image
IJ

As Harvey [13] described, first a buffer of predefined width


is constructed around the reference road data (Figure 7). The
Figure 8. Matched reference ( Heipke et al.,1997) [12]
parts of the extracted data lying in the buffer are considered
as matched. The matched extracted data are denoted as true
positive with length TP, the unmatched extracted data is The completeness is the percentage of the reference data
denoted as false positive with length FP. which is explained by the extracted data, i.e., the percentage
of the reference data which lies within the buffer around the
Second, matching is performed the other way round. The extracted data.
buffer is now constructed around the extracted data (Figure The optimum value for completeness is 1.
8), and the parts of the reference data within the buffer are
considered as matched. In this case of low redundancy their Correctness
length can be approximated with TP. The unmatched
reference data are denoted as false negative with length FN.
Correctness = Length of matched extraction
Length of extraction
2. Quality measures

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 293


Rohit Maurya et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES
Vol No. 5, Issue No. 2, 290 - 295

≈ TP
TP + FP
Correctness ϵ [0; 1] REFERENCES
[1] Hedman, K.; Stilla, U.; Lisini, G.; Gamba, P.; , "Road Network Extraction
The correctness represents the percentage of correctly in VHR SAR Images of Urban and Suburban Areas by Means of Class-
extracted road data, i.e., the percentage of the extracted data Aided Feature-Level Fusion," Geoscience and Remote Sensing, IEEE
Transactions on , vol.48, no.3, pp.1294-1296, March 2010.
which lie within the buffer around the reference data. [2] Hui Kong; Audibert, J.-Y.; Ponce, J.; , "General Road Detection From a
The optimum value for correctness is 1. Single Image," Image Processing, IEEE Transactions on , vol.19, no.8,
pp.2211-2220, Aug. 2010.
[3] P.N ANIL;Dr. S. NATARAJAN;," Automatic Road Extraction from High
Quality Resolution Imagery Based On Statistical Region Merging and
Skeletonization" International Journal of Engineering Science and
Technology Vol. 2(3), 2010, 165-171.
[4] Yinghua He; Hong Wang; Bo Zhang; , "Color based road detection in
Quality = Length of matched extraction urban traffic scenes," Intelligent Transportation Systems, 2003.
qq Proceedings. 2003 IEEE , vol.1, no., pp. 730- 735 vol.1, 12-15 Oct. 2003.
[5] Tomoko Tateyama, Zensho Nakao, Xian Yan Zeng, Yen-Wei Chen,
"Segmentation of High Resolution Satellite Images by Direction and
≈ TP Morphological Filters," his, pp.482-487, Fourth International Conference
TP +FP+ FN

T
on Hybrid Intelligent Systems (HIS'04), 2004.
[6] Tuncer, O.; , "Fully Automatic Road Network Extraction from Satellite
Images," Recent Advances in Space Technologies, 2007. RAST '07. 3rd
International Conference on , vol., no., pp.708-714, 14-16 June 2007.
Quality ϵ [0; 1] [7] H. Mayer, I. Laptev, A. Baumgartner, and C. Steger, “Automatic road
extraction based on multi-scale modeling, context, and snakes”, IEEE
qq = length of extracted data + length of unmatched Trans. on Med. Img., vol. 32, pp. 47–56, 1997.
[8] Zhang, Q. and Couloigner, I., 2006a. Automated Road Network
reference ES Extraction from High Resolution Multi-Spectral Imagery. In: ASPRS 2006
Annual Conference, Reno, Nevada, 10 pages.
The quality is a measure of the “goodness” of the final result. [9] Nock, R.; Nielsen, F.; , "Statistical region merging," Pattern Analysis and
Machine Intelligence, IEEE Transactions on , vol.26, no.11, pp.1452-
It takes into account the completeness of the extracted data 1458, Nov. 2004.
well as its correctness. [10] Jalal, A. A Fuzzy Model for Road Identification in Satellite Images.
The optimum value for quality is 1. Proceedings of the 2006 International Conference on Image Processing,
Computer Vision, & Pattern Recognition, Las Vegas, Nevada, USA, 2006.
[11] Anil Z Chitade, Dr. S.K. Katiyar, “Color Based Image Segmentation
We have tested the various images for road extraction, the using K-Means Clustering”, International Journal of Engineering Science
result of ten test images in terms of completeness, correctness and Technology Vol. 2(10), 2010, 5319-5325
[12] C. Heipke, H. Mayer, C. Wiedemann, and O. Jamet. Evaluation of
and quality is given in the following table. automatic road extraction. In International Archives of Photogrammetry
and Remote Sensing, volume 323-4W2, pages 151-160, 1997.
[13] Harvey, W.A., 1999. Performance evaluation for road extraction.
A
Bull. Soc. Franc Photogrammet. Tele detection 153 (1999-1), 79–87.
[14] Suman Tatiraju, Avi Mehta “Image Segmentation using kmeans clustering
, EM and Normalized Cuts”, University Of California Irvine . .
Image Completeness Correctness Quality
1 0.934682 0.79289 0.751222
2 0.912124 0.683685 0.641435
3 0.884832 0.789277 0.715748
IJ

4 0.783529 0.772095 0.636354


5 0.734900 0.848870 0.649871
6 0.621824 0.570012 0.423276
7 0.580309 0.722598 0.479582
8 0.608722 0.723162 0.493681
9 0.595211 0.346973 0.280730
10 0.677801 0.666831 0.506331
Table 1. Evaluation results

III. CONCLUSION
This method of road extraction can extract roads very
rapidly and gives very satisfactory results. But urban area
with large number of complicated buildings and area with
darker road needs to be researched further.

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 294


Rohit Maurya et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES
Vol No. 5, Issue No. 2, 290 - 295

Test image 1 Output Reference image

T
Test image 5
ES Output Reference image
A
Test image 6 Output Reference image
IJ

Test image 8 Output Reference image

ISSN: 2230-7818 @ 2011 http://www.ijaest.iserp.org. All rights Reserved. Page 295

You might also like