CLASSIFICATION OF PLANT LEAVES USING MORPHOLOGICAL

FEATURES AND ZERNIKE MOMENTS

B.S.Harish
Department of Information Science and Engineering
Sri Jayachamarajendra College of Engineering
Mysore, India
E-mail: [email protected]
Aditi Hedge, OmPriya Venkatesh, D.G.Spoorthy,
D.Sushma
Department of Information Science and Engineering
Sri Jayachamarajendra College of Engineering
Mysore, India
E-mail :{ aditihegde22, ompriya.2691, spoorthy.dg,
sushma9620}@gmail.com


Abstract Plants are an integral part of our ecosystems.
Their identification and classification has always been a matter of
interest for the botanists as well as for the laymen. With around 3
million recognized plant species, only a tiny part of the plants is
known. The leaves of the plant carry a lot of information about
the plant species. These features are extracted and are used as a
basis for an automated identification and classification. The
advancement in image processing has made this a quick and easy
process. In the proposed work, Morphological features and
Zernike moments are used to identify and classify the plant
leaves. The features extracted are independent of leaf growth and
image translation, rotation and scaling and are studied to develop
an approach that produces the best classification algorithm.
First, the developed algorithms are used to classify a training set
of images. Then, a testing set of images is used for verifying the
classification algorithms. The output displays the leaf classified
into its particular class.
Keywords Leaf classification, Morphological feature, Zernike
moments, Identification, Classifiers.
I. INTRODUCTION
Plants form an essential part of life on earth and provide
oxygen, food, medicine, fuel and much more. Plants play a
vital role in environmental protection. The modern practices
of medicine, culinary arts, agriculture, and botany all branch
from a basic science called Plant Identification. This science,
found its start in the sheer practicality of identifying plants. It
is generally a good practice to know which plants are edible
and useful, and which ones produce itchy rashes or even
death. There are about 3 million species of plants that have
been named and classified on earth. It is impossible for any
botanist to know more than a tiny fraction of the total number
of named species. As a result, it is an important but a difficult
task to identify all the plant species on earth. One way to
identify and classify the plant is from their leaves. Each plant
has a unique leaf and it serves as a tool to plant biologists and
botanists for distinguishing plant species and measure the
plant health. Due to the unstoppable growth in human
population and varying climate, there is an escalating threat to
many ecosystems. Many of the plants are still unknown and
most are on the verge of extinction. Hence, plant identification
and classification based on leaves, a practice that is valuable
for its interest and for the multitude of ways in which it can be
applied, is indispensable.

To identify a plant is to recognize and associate it with its
appropriate name. Plants are easily identified based on flowers
and fruits. However these are three dimensional objects and
increases complexity [18]. Whereas plant leaves are two-
dimensional in nature. Thus, they are most suitable for
machine processing. Classification of plants based on leaves is
the quickest and easiest way to identify a plant. These are
the most visible and variable aspect of a plant from trees to
weeds and they can be easily found and collected everywhere
at all seasons, while flowers can only be obtained at blooming
season. The traditional approach to identify and classify these
species is to train taxonomists to manually assign a semantic
label. With the advancement of Information Technology,
systems with more functionality such as automatic labeling
and flexible searching are required and these can be achieved
through image processing and machine learning
techniques. So it is very necessary to set up a database of
leaves that can be used for comparing. Sampling leaves and
photoing them is inexpensive and expedient. We can easily
transfer the leaf image to a computer and then the computer
can extract necessary features automatically using image
processing techniques and subsequently can recognize the
plant / leaf using machine learning techniques. Thus the first
step in this process is to train the computer to classify plants
based on their leaf features. The leaf that is characterized by
its color, texture, vein structure and shape, reveals a vast
amount of information about the plant. This is taken as the
basis for the development of the system which will
automatically classify the given plant leaf into its belonging
class using image processing techniques. The figure1 below
depicts the different steps in the classification system. Image
is acquired and is subjected to pre-processing for removal of
noise and enhancement. Features are extracted and are
compared with those stored in the database. The classifier then
labels it based on the similarity values. Automatic plant
classification systems are essential for a wide range of
applications including environmental protection, plant
resource survey as well as for botanists.
1827 978-1-4673-6217-7/13/$31.00 c 2013 IEEE
Feature
Extraction
Matching
Algorithms /
Classification
System
Image
Pre-processing
Image
Acquisition
Labeling
(Output)
Feature
Database

Figure1. Block diagram of classification system

They can be of great help in Ayurveda to identify and classify
different medicinal plants. The system can be further modified
for classification of flowers, terrain or identification of plant
diseases. It can be used in Precision Agriculture which is the
application of technology and principles to manage spatial and
temporal variability associated with all the aspects of
agricultural production, environmental quality and
sustainability.
This paper is organized as follows. In section II, we briefly
review the works on leaf classification based on different
techniques. The next section, III, discusses the morphological
features and Zernike moments that are implemented in our
proposed system. The experimental set up for the same is in
section IV. The concluding remarks are given in section V.
II. RELATED WORK
Many papers have been presented in International Journals
and conferences; researchers have worked on hierarchical,
neural networks and machine learning methods. The work on
classification of leaves started in the early 20
th
century. The
earlier dataset was the laboriously collected set of leaves,
gathered by those who were working on it. Chrysanthemum,
for instance, was the dataset created and used in [7] involved
user interaction to differentiate the 12 varieties, of the same
kind that included self-intersections. The Flavia dataset [17]
was originally created by Wu et al. It is a collection of 32
different species, a total of 1907 leaves. He extracted 12 leaf
features and orthogonalised them into 5 principle variables.
This was the input vector to the Probabilistic Neural Network
(PNN) that was trained by 1800 leaves. It earned an accuracy
of more than 90%. This flavia dataset, now available as the
standard dataset is accessible to all the researchers to work on.

Initially the works were carried out using neural networks.
Tzionas et al., in [15] implemented an artificial vision system
that extracted specific geometrical and morphological features.
Using a novel feature selection approach, a subset of
significant image features was identified. A feed forward
neural network was employed to perform the main
classification task and that was invariant to size and
orientation. It could successfully operate even with the
deformed leaves. It achieved a considerable high classification
ratio of 99%. Further, Lin and Peng in [10] attempted to
realize a computer automatic classification for 30 broad leaved
plants in a more convenient, rapid and efficient manner using
PNN achieving 98.3% of accuracy. Kadir et al., in [3][4] also
proposed a method for leaf classification which incorporates
shape and vein, color and texture features and used PNN as a
classifier. The experimental result gives an accuracy of
93.75% when tested on Flavia dataset. Later with the
advancement of Machine Learning, researches were conducted
comparing its techniques with the neural network approaches.
Anami et al., in [6] used Support Vector Machine based on
color and texture feature and neural network classifier to
identify and classify images of medicinal plants such as herbs,
shrubs and trees. The edge and color descriptors have low
dimension and are effective, simple and rotation -invariant.
Singh et al., in [11] carried out the experiment on 50 leaf
samples of 32 different classes of varying shapes; using three
different techniques; Support Vector Machine (SVM), Fourier
Moments and PNN based on the leaf shape and achieved an
accuracy of 96%, 62% and 91% respectively. ArunPriya et al.,
in [5] compared SVM and k-NN classifier on flavia and a real
dataset consisting of 15 tree classes and claimed SVM to be a
better classifier achieving more than 94% of accuracy for both
the dataset. Kumar et al., in [12] conducted a survey on
different classification techniques; k-Nearest Neighbor
classifier, Neural Network, Genetic Algorithm, SVM and
Principal Component Analysis and listed their advantages and
disadvantages. The drawbacks of SVM are that it is a binary
classifier, training is slow, and it is difficult to understand
structure of algorithm. It also has limitation with speed and
size, both in training and testing [12]. Owing to the drawbacks
of SVM, Valliammal and Geethalakshmi in [13] worked on
automatic recognition system taking a total of 500 plant and
flower images for identifying and recognizing them to their
respected categories using Preferential Image Segmentation
which is invariant to translation, rotation and scale
transformations. They further extended their work towards
leaf classification and recognition [14] using Hybrid image
segmentation that combines thresholding and H maxima
transformations. This method extracted more accurate values
of the leaf and involved minimum computational time in
comparison. A recent paper by Kaur and Kaur in [8] used
neural network based LM algorithm to train the classifier and
carried out the experiment on 12 kinds of leaves which yielded
an accuracy of 97.9 %.

The leaf classification is of great importance in the field of
medicine. S.E Kumar in [16] conducted an experimental
analysis with a few medicinal plant species such as Hibiscus,
Betel, Ocimum, Leucas, Vinca etc. and proved that the method
of identification based on leaf features such as area, color
histogram and edge histogram is an efficient attempt. A
similar attempt by Gopal et al., in [2] used different
parameters such as shape, venation, texture and color features
for a set of 10 different plant species using boundary based
features, moment features and color features. They developed
software to identify and display the class of a particular plant
1828 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)
leaf. The efficiency was found to be 92%. Laga et al., in [9]
classified based on the outer boundary of leaves and sample
50 points uniformly to obtain discrete representation of the
boundary curve and its performance evaluated using standard
databases showed that the state-of-art descriptor based
techniques were significantly outperformed. Yet another
feature called Zernike moments, a shape descriptor, was
implemented by Kulkarni et al., in [1]. In this method, the
extracted leaf features were combined with the Zernike
moments. The focal improvements were based on the feature
extraction techniques that included Zernike moments and the
dual stage learning algorithm for training the classifier using
Radial Basis Function neural network. The use of dual stage
learning algorithm, for training, of the RBP neural networks is
faster, than the other methods that were used to train
multilayer neural networks. Also the use of dual stage learning
method results in faster convergence during the training phase.
The only negative aspect of Zernike moments is their costly
computation, which makes them inappropriate for certain
problems.
III. PROPOSED WORK
The proposed work attempts to use all the morphological
features along with Zernike moments to classify the leaves.
The image is initially resized to fit into the screen. This color
image is converted to grayscale and then to binary image.
Further, smoothing filter is applied to the binary image to
reduce the noise and extract the required features. The steps
involved in pre-processing technique are illustrated in figure 2.




Figure2. Steps in image pre-processing

The basic geometrical features [17] of the leaf are
1. Diameter (D): The longest distance between any two
points in the leaf
2. Physiological length (L
P
): It is also called as major
axis length. It is the distance between two terminals
of the main vein.
3. Physiological width (W
P
): It is also called as minor
axis length. It is the longest distance between two
terminals of the leaf that is orthogonal to the main
vein.
4. Leaf Perimeter (P): It is the number of pixels at the
margin of the leaf
5. Leaf Area (A): It is the number of pixels forming the
leaf. It can be estimated by calculating the number of
1s in the binary image.

Using the above mentioned geometrical features, following
are the digital morphological features [11] [17] defined.
Aspect Ratio
Aspect ratio is defined as the ratio of physiological
length L
P
to physiological width W
P
. It is also called
as Eccentricity or Slimness.
Lp
AspectRatio
Wp
=
.

Form Factor
This feature is used to describe the difference
between a leaf and a circle. It is the ratio of the area
of the leaf A and the square of the perimeter of the
leaf margin P. It is also called as Roundness,
Compactness or Circularity.
2
A
FormFactor
P
= .
Smooth Factor
The effect of noise to image area is used to illustrate
the smoothness of the leaf image.
5 5
2 2
Area of the image obtained by x rectangular averaging filter
SmoothFactor
Area of the image obtained by x rectangular averaging filter
=

Narrow Factor
Narrow factor is the ratio of the diameter of the leaf
D and physiological length L
P
.
D
NarrowFactor
Lp
=
.

Rectangularity
Rectangularity is the similarity between a leaf and a
rectangle. It is the ratio of the product of
physiological length L
P
and physiological width W
P

and the leaf area A.
*
Rectangularity
Lp Wp
A
=
Perimeter Ratio of Diameter
It is the ratio of perimeter to the diameter denoting
the ratio of leaf perimeter P and leaf diameter D.

P
Perimeter Ratio of Diameter
D
= .
Perimeter Ratio of Physiological length & width
This feature is the ratio of the leaf perimeter P and
the sum of physiological length L
P
and the
physiological width W
P
.
&
(Lp Wp)
P
Perimeter Ratio of Physiological length width =
+



2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI) 1829
6ROLGLW\
6ROLGLW\LVWKHUDWLRRIWKHOHDIDUHD$DQGWKHDUHDRI
WKHFRQYH[KXOO$F
$
6ROLGLW\
$F
=
&RQYH[LW\
&RQYH[LW\ LV WKH UDWLR RI WKH DUHD RI WKH FRQYH[
SHULPHWHU3
&
DQGWKHSHULPHWHURIWKHOHDI3
3F
&RQYH[LW\
3
=
,UUHJXODULW\
,UUHJXODULW\ LV WKH UDWLR RI WKH UDGLXV RI PD[ FLUFOH
5
PD[
HQFORVLQJWKHUHJLRQDQGWKHUDGLXVRIPLQFLUFOH
5
PLQ
WKDWFDQEHFRQWDLQHGLQWKHUHJLRQ7KLVIHDWXUH
LVLPSRUWDQWZKHQWKHVKDSHRIWKHOHDILVLUUHJXODU
PD[
PLQ
5
,UUHJXODULW\
5
=

,9 (;3(5,0(17$/6(783
:H XVH WZR GLIIHUHQW NLQGV RI GDWDVHW )ODYLD GDWDVHW DQG D
GDWDVHWFUHDWHGE\XV)ODYLDGDWDVHWFRQVLVWVRIOHDYHVRI
GLIIHUHQW FODVVHV ZLWK DQ DSSUR[LPDWLRQ RI OHDYHV LQ
HDFK FODVV >@ 7KH OHDYHV DUH WDNHQ RQ ZKLWH EDFNJURXQG
(DFKFODVVRIOHDILQWKHGDWDVHWLVRIGLIIHUHQWVKDSHDQGVL]H

)LJXUH,PDJHDFTXLVLWLRQ
$V D FDVHVWXG\ ZH KDYH FUHDWHG D GDWDVHW ZKLFK LV D
FROOHFWLRQ RI GLIIHUHQW PHGLFLQDO OHDYHV ,W FRQWDLQV VL[
GLIIHUHQW VSHFLHV QDPHO\ 6\]\JLXP FXPLQ -DPXQ
$]DGLUDFKWDLQGLFD1HHP(XFDO\SWXV2EOLTXD(XFDO\SWXV
+LELVFXV URVDVLQHQVLV +LELVFXV 7LQRVSRUD &RUGLIROLD
*XGXFKL DQG 3LSHU %HWOH %HWHO OHDYHV RI HDFK VDPSOH
ZHUH FROOHFWHG DQG WKH LPDJH ZDV DFTXLUHG LQ D EULJKWO\ OLW
URRPE\SODFLQJ WKH OHDIRQD ZKLWHEDFNJURXQGDQGFDSWXUHG
XVLQJ D 03 FDPHUD SODFHG DW D KHLJKW RI FP 7KH DERYH
ILJXUHVKRZVWKHLOOXVWUDWLRQRIWKHH[SHULPHQWDOVHWXSRIWKH
LPDJH DFTXLVLWLRQ %DFNJURXQG RI WKH LPDJHV KDV EHHQ
FXUHWWHG DQG WKH FRORU DGMXVWPHQWV ZHUH WDNHQ FDUH RI E\
NHHSLQJWKHUJEFRORUFXUYHDWDQDYHUDJHRIRXWSXWYDOXH
DQGLQSXWDYHUDJHYDOXHRI

:H FRQGXFWHG WKH H[SHULPHQW XVLQJ WZR VHWV RI GDWDVHW 7KH

ILUVW VHW FRQWDLQV RI WKH VDPSOHV RI HDFK FODVV WKDW DUH
XVHG DV WUDLQLQJ DQG WKH UHPDLQLQJ RI WKH VDPSOHV DUH
XVHG IRU WHVWLQJ 7KH VHFRQG VHW ZDV VSOLW DV IRU WUDLQLQJ
DQG IRU WHVWLQJ )RXU FODVVLILHUV ZHUH LPSOHPHQWHG
QDPHO\ 1DwYH %D\HV N11 690 DQG 311 %RWK WKH VHWV
ZHUH UXQ IRU ILYH LWHUDWLRQV 7KH PLQLPXP PD[LPXP DQG
DYHUDJH DFFXUDF\ DUH QRWHG GRZQ IRU DOO WKH LWHUDWLRQV 1DwYH
%D\HV DQG N11 FODVVLILHUV VKRZHG DQ DFFXUDF\ RI WR
WKDW ZDV IDU RXW UDWHG E\ WKH 690 DQG 311 FODVVLILHUV
ZLWKDQDFFXUDF\RIWR$OOWKHWHVWVZHUHFDUULHGRXW
RQ D SHUVRQDO FRPSXWHU ZLWK DQ ,QWHO SURFHVVRU *% 5$0
DQG PDWODE YHUVLRQ 7KH WDEOH EHORZ VKRZV WKH UHVXOWV
REWDLQHG

0)0RUSKRORJLFDO)HDWXUHV =0=HUQLNH0RPHQWV

7DEOH5HVXOWVWDEXODWHGIRUIODYLDGDWDVHWWHVWHGRQIRXUFODVVLILHUV

9 &21&/86,21
7KH SURSRVHG WHFKQLTXH LV WHVWHG RQ WZR GLIIHUHQW GDWDVHW
)ODYLD GDWDVHW FRQVLVWLQJ RI FODVVHV DQG D GDWDVHW RI
PHGLFLQDOOHDYHVRIFODVVHV FUHDWHGIRURXUH[SHULPHQWDWLRQ
2XU DSSURDFK LQYROYHV LPDJH DFTXLVLWLRQ SUHSURFHVVLQJ
IHDWXUH H[WUDFWLRQ DQG FODVVLILFDWLRQ SKDVH 7HQ GLIIHUHQW
PRUSKRORJLFDO IHDWXUHV DQG =HUQLNH PRPHQWV DUH H[WUDFWHG
DQGIHG DV DQLQSXW WR IRXU GLIIHUHQW FODVVLILHUV QDPHO\ 1DwYH
%D\HV N11 690 DQG 311 :H REVHUYH WKDW 1DwYH %D\HV
DQG N11 EHLQJ OD]\ OHDUQHUV DOJRULWKP JLYH OHVV DFFXUDF\
ZKHUH DV 690 DQG 311 KDYH FRPSDUDWLYHO\ KLJK DFFXUDF\
$V IXWXUH HQKDQFHPHQW WR JHW EHWWHU UHVXOW RQ1DwYH %D\HV
DQG N11 D ODUJH QXPEHU RI VDPSOHV FDQ EH XVHG WR WUDLQ
WKHPDQGNHUQHOIXQFWLRQVFDQEHXVHGIRU690DQG311
'DWDVHW
&ODVVL
ILHU
7UDLQLQJ
YV
WHVWLQJ

0LQLP
XP
$FFXUD
F\
0D[LP
XP
DFFXUDF
\
$YHUD
JH
DFFXU
DF\

)ODYLD

1DwYH
%D\HV
YV
0)
=0
YV
0)
=0

N11
YV
0)
=0
YV
0)
=0

690
YV
0)
=0
YV
0)
=0

311
YV
0)
=0
YV
0)
=0
1830 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)
REFERENCES

[1] A. H. Kulkarni, H. M. Rai, K. A. Jahagirdar and P. S. Upparamani,
(2013). A Leaf Recognition Technique for Plant Classification
Using RBPNN and Zernike Moments, International Journal of
Advanced Research in Computer and Communication
Engineering, Vol. 2, Issue 1, pp. 984-988.
[2] A. Gopal, S. P. Reddyn and V. Gayatri, (2012). Classification of
Selected Medicinal Plants Leaf Using Image Processing, IEEE
International Conference on Machine Vision and Image Processing
(MVIP), Taipei, pp. 5-8.
[3] A. Kadir, L. E. Nugroho, A. Susanto and P. I. Santosa, (2011),
Leaf Classification Using Shape, Color, and Texture Features ,
International Journal of Computer Trends and Technology, pp.
225-230.
[4] A. Kadir, L. E. Nugroho, A. Susanto and P. I. Santosa (2012),
Performance Improvement of Leaf Identification System Using
Principal Component Analysis, International Journal of Advanced
Science and Technology, Vol. 44, pp. 113-124.
[5] C. Arunpriya, T. Balassaravanan and A. S. Thanamani, (2012). An
Efficient Leaf Recognition Algorithm for Plant Classification
Using Support Vector Machine, IEEE International Conference on
Pattern Recognition, Information and medical Engineering, March,
Salem, Tamilnadu, pp. 428-432.
[6] B. S. Anami, S. S. Nandyal and A. Govardhan, (2010). A
Combined Color, Texture and Edge Features Based Approach for
Identification and Classification of Indian Medicinal Plants,
International Journal of Computer Applications, Vol. 6, No. 12, pp.
45-51.
[7] F. Mokhtarian and S. Abbasi, (2004). Matching Shapes With Self-
Intersections: Application to Leaf Classification, IEEE
Transactions on Image Processing, May, Vol. 13, No. 5, pp. 653-
661.
[8] G. Kaur and G. Kaur (2012), Classification of Biological Species
Based on Leaf Architecture, International Journal of Engineering
Research and Development, Vol 1, Issue 6, pp. 35-42.
[9] H. Laga, Kurtek, A. Srivastava, M. Golzarian and S. Miklavcic
(2012), A Riemannian Elastic Metric for Shape-based Plant Leaf
Classification, IEEE International Conference on Digital Image
Computing Techniques and Applications (DICTA), Fremantle,
WA, pp. 1-7.
[10] H. Lin and H. Peng, (2008). Machine Recognition for Broad-
Leaved Trees Based on Synthetic Features of Leaves Using
Probabilistic Neural Network, IEEE International Conference on
Computer Science and Software Engineering, Wuhan, Hubei, pp.
871-877.
[11] K. Singh, I. Gupta, and S. Gupta, (2010). SVM-BDT PNN and
Fourier Technique for Classification of Leaf Shape, (2010).
International Journal of Signal Processing, Image Processing and
Pattern Recognition, Vol. 3, No. 4, pp. 67-78.
[12] M. Kumar, M. Kamble, S. Pawar, P. Patil and N. Bonde , Survey
on Techniques for Plant Leaf Classification, International Journal
of Modern Engineering Research (IJMER), Vol.1, Issue.2, pp. 538-
544.
[13] N. Valliammal and S. N. Geethalakshmi, (2011). Automatic
Recognisition System Using Preferential Image Segmentation for
Leaf and Flower Images, Computer Science and Engineering: An
International Journal (CSEIJ), Vol. 1, No. 4, pp. 13-25.
[14] N. Valliammal and S. N. Geethalakshmi, (2011). Hybrid Image
Segmentation Algorithm for Leaf Recognisition and
Characerization, IEEE International Conference on Process
Automate Control and Computing (PACC), Coimbatore, pp. 1-6.
[15] P. Tzionas, S E. Papadakis and D. Manolakis, (2005). Plant leaves
classification based on morphological features and a fuzzy surface
selection technique, 5th International Conference on Technology
and Automation ICTA'05, Thessaloniki, Greece, 15-16 Oct. 2005,
pp. 365-370.
[16] S. E. Kumar, (2012). Leaf Color, Area and Edge Features Based
Approach for Identification of Indian Medicinal Plants, Indian
Journal of Computer Science and Engineering (IJCSE), Vol. 3, No.
3, pp. 436-442.
[17] S. G. Wu, F. S. Bao, E. Y. Xu, Y. X. Wang, Y. F. Chang and Q. L.
Xiang, (2007). A Leaf Recognition Algorithm for Plant
Classification Using Probablistic Neural Network, IEEE
International Symposium on Signal Processing and Information
Technology, Giza, pp. 11-16.
[18] T. Zaharia and F. Prteux, (2004). 3D versus 2D/3D Shape
Descriptors: A Comparative study, SPIE Conference on Image
Processing: Algorithms and Systems.

2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI) 1831