See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281118125

IMAGE CLASSIFICATION TOOL FOR LAND USE / LAND COVER ANALYSIS: A

COMPARATIVE STUDY OF MAXIMUM LIKELIHOOD AND MINIMUM DISTANCE
METHOD

Article · October 2012

CITATIONS READS
25 2,553

3 authors:

Manisha Patil Chitra Desai

Zensar Technologies Marathwada Institute of Technology
12 PUBLICATIONS   53 CITATIONS    4 PUBLICATIONS   55 CITATIONS   

SEE PROFILE SEE PROFILE

Bhavana Umrikar
Savitribai Phule Pune University
61 PUBLICATIONS   290 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Study of Surface and Subsurface Water Interaction using Remote Sensing, Geo-hydrological and Geophysical Techniques and its modelling View project

PhD work View project

All content following this page was uploaded by Bhavana Umrikar on 21 August 2015.

The user has requested enhancement of the downloaded file.

International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article
IMAGE CLASSIFICATION TOOL FOR LAND USE / LAND COVER
ANALYSIS: A COMPARATIVE STUDY OF MAXIMUM LIKELIHOOD
AND MINIMUM DISTANCE METHOD
Manisha B. Patil1, Chitra G. Desai2 and *Bhavana N. Umrikar3
1
Department of Computer Science, University of Pune, India
2
MIT Engineering College, Aurangabad, India
3
Department of Geology, University of Pune, India
*Author for Correspondence

ABSTRACT
Monitoring Land use / Land cover (LULC) changes and develop a model is utmost important for the
developing cities to avoid its haphazard growth, especially in fringe areas. With the help of satellite
imageries available for last few years or decades, with different spatial and spectral resolution, this task
has become easier and reliable. But, to give the model solution for any real world problem there is a need
to have reliable input data in appropriate format. Hence to provide solution for uncontrolled land use
changing pattern, it is mandatory to perform image analysis on raw images.
This study aims to review different image classification methods and its utility for Land use / Land cover
analysis. The paper includes demonstration of supervised image classification with both maximum
likelihood and minimum distance method.

Key Words: Supervised Image Classification, Land Use / Land Cover, Pune Municipal Corporation

INTRODUCTION
Remote Sensing is an advanced tool for the inventory and analysis of Land use pattern and also for city
planning at both small and large scales in undeveloped and developing parts of the world. Various types
of remote sensing data, such as MSS, LISS, TM, SPOT, MODIS and AVHRR have been used for natural
resources management and monitoring Land use pattern (Gregorio and Jansen, 1998 and Platt, 2004).
Image classification is one of the important phases of image analysis process (Anderson et al., 1976;
Adeniyi, 1985; Paul, 1991; Richards, 1993; Meyer, 1995 and Mather, 2004). The overall objective of
image classification procedures is to automatically categorize all pixels in an image into land cover
classes or themes. Normally multispectral data are used to perform the image classification and, the
spectral pattern depicted by each pixel is used as the numerical basis for categorization (Jensen, 1996).
Image analysis is a sequential process which includes many phases like: Image import, Georeferencing,
Image subsetting, Geometric correction, Image classification, design algorithm etc (Jensen, 2002;
Lillesand and Kiefer, 2004 Thomas, 2001). Present research work focuses on Image classification phase
of Image analysis process.
Types of Image Classification Process
Image classification process is divided into two: Hard and soft classification.
Hard Classification:
In hard classification, each pixel represents a homogeneous area on the ground and shows only land cover
type. Statistical methods are used to map each pixel by assigning it exclusively to one specific class. The
spectrally similar data is described thematically in similar objects and is a dominant scene component for
each pixel. Hard classification method is subdivided into three types, namely Unsupervised, Supervised
and Expert knowledge based classification. Out of these three methods, two-classification methods i.e.
unsupervised and supervised classifications are mostly used in many spatial problem solving techniques.

189
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article
Supervised classification: In supervised classification, the image analyst supervises the pixel
categorization process by specifying the computer algorithm and numerical descriptor representing
various land cover types in a scene (ERDAS Field Guide, 1999).
Unsupervised classification: In unsupervised classification, the image analysts need not to have
knowledge about the existing assignment or name of those classes. In short, image analysts do not
supervise the pixel categorization process in unsupervised classification method.
Expert knowledge based classification: Expert knowledge based classification is a hierarchy of rules, or a
decision tree which describes the conditions under which a set of low level constituent information gets
abstracted into a set of high level informational classes.

Figure 1: Image classification techniques

Soft Slassification
This classification method deals with mixed pixels problem by describing the spatially heterogeneous
character of land cover in terms of continuous surfaces. This method provides the user with a measure of
the degree (termed membership grade ) to which the given pixel belongs to some or all of the candidate
classes, and leaves on the investigator’s decision as to which category the pixel could be attributed.
Object oriented image analysis classification: This classification technique is based on the object-based
approach and it is called as soft classification method. In this method, image attribute information like
location, shape, tone, texture, relational and contextual information is used during categorization.
Accuracy Assessment
The final stage of image classification process usually involves an accuracy assessment. Accuracy
estimation is the quantification of mapping using remote sensing data to the group classification
conditions. This is useful in evaluation of classification techniques, and determining the level of error that
might be contributed by the image. Accuracy of each classification is expressed in the form of an error
matrix (Congalton, 1991; Congalton and Green, 1998 and Smith et al., 2003).
An error matrix is a square array of numbers in which the columns express the informational categories,
and the rows show the classes in which those informational categories have been classified. The overall
agreement of the classification is therefore expressed by the sum of main diagonal entries. An omission
error happens when a test area is not classified into its informational category. On the other hand, the

190
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article
commission occurs when a test area is classified in a class different from its true informational categories.
Information about these types of error is given by the user’s and producer’s accuracies respectively. Test
area of same size as the training areas were used to determine the classes in which a test area is classified.
The agreement of the classification with ground truth was measured by means of the overall accuracy,
and the Kappa Statistics.
Traditional accuracy assessment is done by generating random set of locations to visit the ground for
verification of the true land cover type. A simple value file is then made to record the true land cover
class (by integer index number) for each of these location. This value file is then used with the vector file
of point locations to create raster image of true classes found at the location examined. The raster image is
then compared to the classified map using error matrix. Error matrix tabulates the relationship between
true land cover classes and the classes as mapped. It also tabulates error of omission and errors of
commission as well as the overall proportional error. This information is used to assess the accuracy of
the classification procedure that was under taken and is used for the results of all supervised
classification.

MATERIALS AND METHODS

In unsupervised classification, there is a possibility that the maps and images that are produced through
this method do not match the reality and therefore the accuracy of the resulting maps and images always
need to be verified in the field before using them. Due to this known drawback of unsupervised
classification method, Maximum Likelihood Classification (MLC) of supervised classification method is
used for the present research work.
After this, another supervised classification was performed with Minimum Distance Classification
(MDC) option to compare the results. LANDSAT image of the year 1999 is classified into 4 major
classes: barren land, vegetation, water bodies, built-up area etc. The built up area is direct indicator of
urban land which includes most of the dense built-up (Patil et al, 2012).
Pune Municipal Corporation (PMC) area is selected as a case study for this research work. Hence PMC
boundary has been cropped from Survey of India Topographic sheet No. 47 F/10, 11, 14 and 15 and
considered as a base map for the Land use Land cover (LULC) classification. Figure 2 shows
Topographic sheet of Pune Municipal Corporation area.
Algorithm used for this present research work
Practical Demonstration
1. The digital remote sensing data of past decades has been procured.
2. Topographic sheet of PMC area of year 1972 was scanned and geo referenced. This topographic
map was further used as base map for the image registration.
3. The LANDSAT imagery for the year 1999 was imported, enhanced and geo referenced using ERDAS
9.2 software (Figure 3).
4. Using signature editor tool, various pixel colors were selected for different classes
5. The enhanced LANDSAT satellite image of the year 1999 was classified into different classes by
using ERDAS imagine and compared with city land use maps.
6. Maximum Likelihood Classification (MLC) of supervised classification method was firstly used
(Figure 5). After this, another supervised classification was performed with Minimum Distance
Classification (MDC) option to compare the results / output of earlier option.

191
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article

Figure 2: Topographic sheet of Pune Municipal Corporation area

192
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article

Figure 3: LANDSAT satellite imagery of Pune Municipal Corporation area

Figure 4: Selected pixel colors for different classes using signature editor

193
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article

Figure 5: Maximum Likelihood and Minimum Distance options of supervised classification

Figure 6: Output of MLC (1) and MDC (2) supervised classification

194
International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article

Evaluation of Classification Error Matrix

Table 1: shows the Error Matrix Resulting from Classifying Training Set Pixels using Maximum
Likelihood method
Training Set Data
B W V Bu Row Total
B 126 0 5 0 131
W 0 52 0 10 62
V 0 16 100 0 116
Bu 0 0 38 480 518
Column Total 126 68 143 490 827
Table 1: Accuracy assessment Using MLC (126+52+100+480)/827 = 0.91%

Table 2 shows the Error Matrix Resulting from Classifying Training Set Pixels using Minimum
Distance method
Training Set Data
B W V Bu Row Total
B 200 0 10 0 210
W 0 52 0 20 72
V 0 20 100 0 120
Bu 0 0 38 290 328
Column Total 200 72 148 310 730
Table 2: Accuracy assessment Using MDC (200+52+100+290)/730 = 0.87%

RESULT AND DISCUSSION

A variety of supervised classification methods has been applied and tested extensively for land use
planning and management worldwide. But, for this research study, conducted on Pune Municipal
Corporation area, a per-pixel maximum likelihood classification of LANDSAT-TM data offered the most
satisfactory results. From the results, it has been confirmed that supervised classification using maximum
likelihood method is most accurate as compared to minimum distance method.
Our goal was to compare the performance of MLC and MDC supervised classification algorithms in the
benchmark task of Land use / Land cover categorization of Pune Municipal Corporation area. Maximum
likelihood supervised classification method yielded higher accuracy in results of image classification even
though it is computationally intensive and time consuming. Hence for present research work, MLC
approach has been adopted for Land use / Land cover monitoring and analysis.

REFERENCES
Adeniyi PO (1985). Digital analysis of Multitemporal Landsat Data for Land-Use Land Cover
Classification in a Semi-Arid Area of Nigeria. In: Photogrammetry engineering and Remote sensing 51
(11) 1761-1774.
Anderson JR, Hardy EE, Reach IT and Witmer RW (1976). A land use and land cover classification
system for use with remote sensing data. U.S. Geological Survey. Reston, Virginia (USA). Professional
964 28.
Congalton RG (1991). A review of Assessing the Accuracy of Classifications of Remotely Sensed Data.
Remote Sensing of Environment 37 35-46.
Congalton RG and Green K (1998). Assessing the Accuracy of Remotely Sensed Data, Lewis
Publishers United States.

195
 International Journal of Geology, Earth and Environmental Sciences ISSN: 2277-2081 (Online)
An Online International Journal Available at http://www.cibtech.org/jgee.htm
2012 Vol. 2 (3) September - December, pp.189-196/Mondal
Research Article
Desai CG, Patil MB and Umrikar BN (2009). Application of Remote Sensing and Geographic
Information System to study land use / land cover changes: A case study of Pune Metropolis, Jour.
Advances in Computational Research 1(2) 10-13.
ERDAS Field Guide (1999). Earth Resources Data Analysis System. ERDAS Inc. Atlanta, Georgia 628.
Gregorio AD and Jansen LJM (1998). Land Cover classification System (LCCS); classification
concepts and user manual, SDRN, FAO, Rome.
Jensen JR (1996). Introductory digital image processing: A remote sensing perspective. Prentice-Hall
Inc, Eaglewood Cliff, UK.
Jensen JR (2002). Remote Sensing of the Environment. Prentice Hall Inc. UK.
Lillesand MT and Kiefer WR (2004). Remote sensing and image interpretation. 5th Edition John Wiley
and Sons, Inc New York.
Mather PM (2004). Computer processing of remotely sensed images. 3rd Edition John Wiley and Sons
Ltd.
Meyer WB (1995). Past and present land-use and land-cover in the USA consequences. The Nature and
Implications of Environmental Change 1(1) 24-33.
Patil MB, Desai CG and Umrikar BN (2012). Comparative study of nearest neighbour and bilinear
interpolation raster transformation techniques for predicting urbanization. Proceedings, National
Conference: INDIACom-2012, BVICAM, Delhi 509-512.
Paul M (1991). Computer processing of remotely sensed images: An introduction 352. Biddley Limited
Publication, UK.
Platt RV and Goetz AFH (2004). A Comparison of AVIRIS and Landsat for land use classification at
the urban fringe, Photogram Eng Remote Sensing 70 813-819.
Richards JA (1993). Remote sensing digital image analysis: an introduction.
Smith JH, Stehman SV, Wickham JD and Yang L (2003). Effects of landscape characteristics on land-
cover class accuracy. Remote Sensing of Environment 84 342-349.
Thomas H, Cormen Charles E, Leiserson RL, Rivest and Clifford S (2001). Introduction to
Algorithms, Second Edition MIT Press and McGraw-Hill Section 16.3 385-392.

196

View publication stats