International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

RESEARCH ARTICLE OPEN ACCESS

A Review on Recognition of Plant Species Using Leaf Image

through Digital Image Processing
Gautam Saikia [1], Mutum Bidyarani Devi [2]
[1]
Computer Science Department, Manipur International University, Manipur
[2]
Electronics and Computer Science Department, Manipur International University, Manipur
ABSTRACT
Identification of new plant species is an important aspect for conservation on Earth. Among many morphological features of a
plant the leaf morphology is the most popular and effective way to identify a plant species using digital image processing. This
paper aims to review and discuss some of the significant literatures and their methodologies used for recognizing plant species
using leaf image. Also, the performance of the different techniques used for plant species extraction and classification has been
reviewed in this paper.
Keywords — Digital Signal Processing, Leaf Recognition, Feature Extraction, Neural Network.

I. INTRODUCTION
Plants play an important rule on maintaining the balance
on Earth’s atmosphere. Around 3,10,000–4,20,000 known and
unknown plant species are available all over the world [1].
Finding out the correct nomenclature plays a vital role in
recognizing where an unknown plant belongs to the ecosystem.
But the process of identifying the correct nomenclature is
limited among taxonomists. Due to global shortage of
taxonomists and the increasing ability of computer vision and
machine learning, the demand of automated software tool for
recognizing of correct nomenclature of plants has increased.
For identification of plants, morphological characters are
primary based before molecular diagnostic. Morphological
features of leaves regarding perfect nomenclature signified
permanent charter when it is digitalized. Therefore, among Fig. 1 A flowchart representation a typical leaf base plant identification
several traditional ways for recognizing a plant morphological system
character, most of the scholars and researchers applied
morphological based character as leaf in respect to its base, Hati, et al. [2] use Artificial Neural Network (ANN) for
margin, apex, blade, texture and venation system etc. This recognition of plant based on leaf shape, where new input
paper presents a brief overview and literature review on features and image processing approach has been introduced.
various plant recognition techniques based on leaf image. During data acquisition phase, for preparation of leaf database
of 534 leaves from 20 plants, each leaf images were captured
II. ANALYTICAL COMPUTER-BASED on a clear white background. Due to this new input features,
the colour to grayscale conversion of leaf image can be done
TECHNIQUES APPLIED IN PLANT only by processing the blue band instead of converting RGB
SPECIES RECOGNITION value of each pixel of the image to grayscale value by using
An enormous application has been done on leaf the following formula [3].
morphology based on plant classification and recognition, but grey = 0.2989×R + 0.5870×G + 0.1140×B
these are still limited approaches for investigation of plant Aspect ratio, width ratio, apex angle, apex ration, base
identification system. Relating to various applications a few angle, moment ration and circularity were the seven
methods perform satisfactory results in respect to specific parameters, which were extracted during feature extraction
samples of leaves. Sometimes it has been observed that two or phase, and constitute the input layer of the artificial neural
more plant have similar leaf natures that possess a critical network. A software tool namely LEAFilia has been
illness during reorganisation of correct nomenclature. In such developed in Java with three modules for image processing,
cases, shape and size of leaf may be rejected but the colour training of neural network and matching of unknown leaf
and margin character of the leaf cannot be neglected. The image with a recognition accuracy of 92%.
following Fig. 1 represents a typical model of leaf base plant Sandeep Kumar. E [4] proposed a simple approach where
identification system. indian medicinal plant has been identified using area, color
histogram and edge histogram of leaf. The leaf image was

ISSN: 2347-8578 www.ijcstjournal.org Page 67

International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

obtained using 5 mega pixel camera where 15cms distance has classifier consist of convolutional layer, maxpooling layer,
been maintained between leaf and the camera and all the dropout layer and fully connected layer. The proposed model
images were captured with white background in natural day found a prediction accuracy of 97% on unknown leaf images.
light. While considering the edge features of the leaf image, Hossain J. et al. [9] proposed a plant species recognition
RGB to grayscale conversion has been done and by using technique based on leaf image using probabilistic neural
Canny edge detection algorithm the edge histogram has been network. During data acquisition phase only the plant with
calculated. Also, the area of the leaf was calculated using the broad flat leaves which were two dimensional in nature were
same algorithm as well as the colour of the image was also considered, which may be a limitation of the proposed method
extracted in form of histogram. Three different algorithms as it is unable to recognise all type of plant leaves. 91.41% of
have been proposed for image acquisition phase along with a average recognition accuracy has been found during testing
system algorithm which was implemented and tested on using ten-fold cross-validation technique.
MATLAB 7.0. Quach, Boi M. et al. [10] proposed a method to recognise
Satti, V. et al. [5] proposed a leaf recognition system leaf using convolutional neural network, where during pre-
where they considered the leaf color as well as the shape processing of leaf image, a refined color image, vein image,
feature for improving the efficiency of recognition which is xy-projection histogram, handcrafted shape, texture features
tested on Flavia dataset consist of 1907 RGB leaves of 33 and Fourier descriptors were extracted and using neural
plants. Neural Network and Euclidean Distance Method has network based encoder all the extracted attributes were
been used to classify the dataset where 93.3% and 85.9% of transformed into a better representation. Support vector
accuracy has been found, respectively. machine (SVM) has been used for classification of pre-
Kaur, G. et al. [3] introduce a system for classification of processed image. During testing on Flavia leaf dataset,
biological species based on leaf architecture which consist of 99.69% of recognition accuracy has been achieved by the
four modules namely – image acquisition, image proposed method.
preprocessing, image recognition and display result. In this Ehsanirad, A. et al. [11] conducted excremental research to
work author used the Gray-level co-occurrence matrix illustrate the performance on extraction of leaf texture features
(GLCM) for texture feature extraction as well as Artificial of two algorithm namely grey-level co-occurrence matrix
Neural Network (ANN) and Multilayer Perceptron Neural (GLCM) and principal component analysis (PCA). Both the
Network (MLPNN) for classification, which improved the algorithms were trained to classify 13 kinds of plant based on
system accuracy up to 97.9%. 390 leaves where GLCM got a 78.46% of accuracy while
Anubha Pearline S. et al. [6] has conducted a compression PCM accuracy achievement is 98.46%.
study by applying both traditional image processing method Selda, J. D. S. et al. [12] build a portable device with a
and deep learning methods for plant recognition. In traditional capability to identify the plant species based on leaf vein,
approach, local binary patter (LBP), color channel statistics where scale invariant feature transform (SIFT) algorithm has
etc. were used during features extraction and during been used for extracting the matrix from the leaf vein image
classification, logistic regression, k-nearest neighbour etc. and the support vectors machines (SVM) has been used for
classifiers were used. In deep learning approach, classification.
Convolutional Neural Network (CNN) architecture with Mouine, S. et al. [13] proposed a new multiscale shape-
logistic regression had been used. From the study it has been based approach for leaf image classification where they
observed that 82.38% of plant recognition accuracy for studied four types of multiscale triangular shape namely
Leaf12 dataset has been found while applying traditional Triangle side length representation (TSL), Tringle area
approaches whereas 97.14% of accuracy result has been representation (TAR), Tringle side lengths and angle
obtained from deep learning approach. representation (TSLA) and Tringle oriented angles (TOA).
Ji-Xiang Du et al. [7] proposed an efficient computer-aided While testing their system on Swedish image dataset the
plant species identification approach based on leaf image classification accuracy rate of 95,73%, 90.4%, 96.53% and
where Douglas-Peucker approximation algorithm has been 95.20% has been found in respect of Triangle side length
used to form the sequence of invariant attributes from the representation (TSL), Tringle area representation (TAR),
original leaf image input. Then for recognition of plant leaf, Tringle side lengths and angle representation (TSLA) and
unlike traditional methods, the research proposed a modified Tringle oriented angles (TOA) respectively.
dynamic programming (MDP) algorithm for image shape Aakif, A. et al. [14] used artificial neural network along
matching. From the experimental result the MDP algorithm with back propagation for develop an algorithm which can be
has been found more suitable not only for intact leaves but able to automatically classify the plant species based on leaf
also for distorted, partial and overlapped plant leaves over image. A vector of morphological features along with Fourier
traditional approaches. descriptors (FD) extracted during feature extraction phase
Bisen, D. et al. [8] use deep convolutional neural network were fed into the artificial neural network classifier for
for identification of plant species through leaf image. Three accurate classification of plant species. During testing on
main identification steps were undertaken in the proposed Flavia and ICL dataset, the proposed system achieved a
method namely image pre-processing, feature extraction and classification accuracy of 96% in respect of both datasets.
recognition where the convolutional neural network (CNN)

ISSN: 2347-8578 www.ijcstjournal.org Page 68

International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

Carranza-Rojas J. et al. [15] proposed a remarkably species (75 images from each species) which was followed by
interesting method where they created a clean dataset as well image pre-processing where noise handling, image resizing
as a noise dataset and implement the histogram of curvature and image enhancement were done. Gaussian filtering has
over scale (HCoS) along with local binary pattern variance been used for handing of noise and the image has been resized
(LBPV) algorithm on both datasets to extract contour to 330*400 pixel. To extract the region of interest (ROI)
information and texture information of leaf image. It has also during image segmentation phase author apply k-means color
observed that the clean dataset performs only 7.3% better based clustering techniques by using MATLAB R2018b.
result than the noise dataset. Texture and colour feature were extracted on feature
Herdiyeni Y. et al. [16] developed a mobile base extraction phase using grey-level co-occurrence matrix
application for identification of medicinal plant of Indonesia (GLCM) and first order statistics from colour image
where they use fusion of fuzzy local binary pattern, fuzzy respectively. During classification phase of the proposed
colour histogram and a probabilistic neural network (PNN) system multiclass support vector machine (MSVM) classifier
classifier for identification of plant based on leaf features. The has been used. And aggregate accuracy of 93.26% has been
proposed application has been tested on 2448 Indonesian achieved by the system during accuracy testing phase.
medical plant’s leaf images, where each image was of Wang Z. et al. [20] proposed an approach for retrieval of
270*240 pixel, and a 74.5% of classification accuracy has image leaf based on shape of the leaf which includes two steps.
been found. To reduce the search space in the first step, all dissimilar leaf
Hernandez-Serna A. et al. [17] proposed a new images with the query image were filtered out using
identification system capable for identification of fish, plant as eccentricity. In the next step, a fine retrieval in the reduced
well as butterfly species where multiple image processing search space has been carried out using simple shape features
techniques were used for extracting 6 geometrical, 2 including centroid contour distance (CCD), eccentricity and
morphological and 8 texture features of the image, and these angle code histogram (ACH). During testing phase 1400 leaf
extracted features were feed into an artificial neural network images were taken from 140 plant species where a better
for recognising of pattern. The proposed system has been retrieval accuracy has been found.
tested on Flavia dataset where 92.87% of accuracy has been Ibrahim Z. et al. [21] conducted a comparative study on
obtained in respect of plant species identification. recognition of leaf images among three popular texture
Begue A. et al. [18] introduced an automated method for features namely histogram of oriental gradients (HOG), local
recognition of medicinal plant using computer vision and five binary pattern (LBP) and speeded-up robust features (SURF)
different machine learning classifier namely random forest, with one-against-one (OAO) approach multiclass support
multilayer perceptron neural network, support vector machine, vector machine (MSVM). The recognition accuracy of three
naïve Bayes and k-nearest neighbour. Among all the five texture features HOG, LBP and SURF with MSVM classifier
classifier the best classification accuracy of 90.1% has been has been tested on Flavia dataset where HOB and LBP
achieved while using random forest classifier. Authors also achieved the same accuracy rate of 99% which is found better
clarified that due to constraints on resource the function of then SURF.
multilayer perceptron neural network classifier could not be Chaki J. et al. [22] developed a method to detect 3 classes
fully exploited therefore the accuracy of this classifier may be of plant category by analysing leaf shape using moment-
more than they opiated in their research which is 88.2%. invariants (M-I) and centroid-radii (C-R) approach
Kaur S. et al. [19] proposed a plant species identification individually with some neural network algorithm for
system using computer vision and machine learning approach classification. Also, a study has been conducted by the author
based on leaf images of a plant. The operational flow chart of to improve the reorganization accuracy rate by introducing a
the proposed system is shown in Fig.2. hybrid representation with combination of M-I and C-R
features. 90%-100% range of accuracy has been reported
during testing phase of the study.
Wu Q. et al. [23] introduce an approach on 2006 for
features extraction and recognition of plant based on leaf
shape using artificial neural network, where four visual
features were extracted for representing the leaf shape that
were slimness, roundness, solidity and moment invariants of
the leaf. To represent the leaf margin and dent four features
namely coarseness, size, angle and sharpness of the leaf were
extracted. Moreover, ramification and camber were two
another features, which were also extracting to represent the
leaf vein. Further, leaf shape, dent and vein features that were
Fig. 2 Operational flowchart extracted during feature extraction phase were feed into the
During image acquisition phase the author used the trained neural network for recognition of the plant. 180 no of
Swedish dataset which consist of total 1125 images of 15 leaf images has been collected from 6 plant species (30 leaf
image from each) for identifying the extraction and

ISSN: 2347-8578 www.ijcstjournal.org Page 69

International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

recognition performance of the system which was found to be Anshul Satya and shape Distrance
satisfactory according to the authors. and Shanu Method
Using leaf shape features and minimum Euclidean distance Sharma
Haque, F. et al. [24] proposed a plant recognition system Local Binary
Pattern (LBP),
consisting of three main steps namely image pre-processing, S. Anubha
Color Channel
feature extraction and matching. During image pre-processing Pearline, V.
Leaf color Statistics,
the following operation were performed on the input leaf 5 Sathiesh [6]
and shape Logistic
image: Kumar, S.
Regression, k-
Harni
• Resize the image, nearest
• Convert from RGB to grayscale, neighbour, CNN
• Convert from grayscale to binary, Ji-Xiang Du,
Douglass-
Peucker
• Swap between the white pixels to black pixels, De-Shuang
algorithm,
• Perform morphological opening, 6
Huang, Xiao- Leaf
Modified [7]
• Use average filter, Feng Shape
Dynamic
Wang,Xiao
• Extract the region of interest (ROI), Programme
Gu
• Obtain the properties of ROI, (MDP)
• Convert the image into convex hull image, Deep
Dhananjay Leaf Conversational
• Obtain the properties of convex hall image, 7 [8]
Bisen Shape Neural
• Obtain the length and width of bounding box. Network, CNN
The following ratios were extracted during feature Javed Hossain
extraction phase: and M. Leaf
8 PNN [9]
• Aspect Ratio Ashraful shape
• Rectangularity Amin
• Area convexity Boi M.
Quach, V.
• Perimeter convexity Leaf
Cuong Dinh,
• Circularity 9 Nhung Pham,
Shape,
CNN, SVM [10]
• Eccentricity Dang Huynh,
Colour
and Vein
• Form factor Bint T.
All the ratios extracted from feature extraction phase were Nguyen
considered as input for matching phase. On matching phase, Abdolvahab
the closest match to the leaf image has been computed by Ehsanirad and Leaf
10 GLCM, PCA [11]
Y.H. Sharath Shape
using minimum Euclidean distance. The system has been
Kumar
tested on 50 leaf images from 10 plant species which were
Jesse Dave S.
collected from flavia dataset, and 94% accuracy rate was Selda, Roi
achieved. Martin R. Scale-Invariant
The following Table-I illustrate an overview of various Ellera, Feature
11 Leaf Vein [12]
plant feature extraction and classification method used for Leandro C. Transform
plant recognition based on leaf image introduced by different Cajayon, (SIFT), SVM
researchers. Noel B.
Linsanga
TABLE I Triangle side
AN OVERVIEW OF VARIOUS PLANT FEATURE EXTRACTION AND
CLASSIFICATION METHOD USED FOR PLANT RECOGNITION
length
representation
(TSL), Tringle
Sl. Research area
Researchers Method Used Reference S. Mouine, I.
No. Based on representation
Yahiaoui, A. Leaf
Shayan Hati 12 (TAR), Tringle [13]
Leaf Verroust- shape
1 and Sajeevan ANN, LEAFilia [2] side lengths and
shape Blondet
G. angle
Gurpreet representation
Kaur and Leaf GLCM, ANN, (TSLA) and
2 [3] Tringle oriented
Gurpinder shape MLPNN
Kaur angles (TOA)
Leaf Canny Edge A. Arkif and Leaf ANN with back
Sandeeo 13 [14]
3 Shape and Detection, [4] M. F. Khan shape propagation
Kumar E. J. Carranza- Leaf Histogram of
colour MATLAB 14 [15]
4 Vijay Satti, Leaf color ANN, Euclidan [5] Rojas and E. shape curvature over

ISSN: 2347-8578 www.ijcstjournal.org Page 70

International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

Mata- scale (HCoS), III. CONCLUSIONS

Montero local binary
pattern variance To address the environmental challenges such as loss of
(LBPV) plant habitats and efforts towards conservation of plant
Yeni Fuzzy local biodiversity, recognition of plant species became an important
Herdiyeni Leaf binary pattern, aspect. In the meanwhile, the significant growth of computer
15 [16]
and Ni Kadak Shape Fuzzy colour vision and machine learning has led to increasing demand for
Sri Wahyuni histogram, PNN automated systems for proper prediction of plant species. In
Andres this paper we have discussed and explained several techniques
Hernandez- introduced in recent years by different researchers on plant
Serna and Leaf
16 ANN [17] species recognition based on leaf image. While analysis the
Luz Fernanda Shape
Jimenez- accuracy rate of the proposed systems of different literatures it
Segura has been observed that system using neural network classifiers
Adams like probabilistic neural network (PNN), convolutional neural
Begue, Random forest, network (CNN), Deep Conversational Neural Network
Venitha Multilayer achieved a high classification accuracy rate comparing to
Kowlessur, perceptron other methods. Table-1 summarizes an overview of various
Fawzi Leaf neural network, plant feature extraction and classification method used for
17 [18]
Mahomoodall Shape Support vector plant recognition based on leaf image proposed in each
y, Upasana machine, Naïve literature that are reviewed in this paper.
singh, bayes, k-nearest
Sameerchand neighbour
Pudaruth REFERENCES
Gaussion
Surleen Kaur filtering, k- [1] Ji-Xiang Du, Xiao-Feng Wang and DS Huang,
and Leaf color means color “Automatic plant leaves recognition system based on
18 [19] image processing techniques,” Technical Report,
Prabhpreet and shape based clustering
Kaur techniques, Institute of Intelligent Machines, Chinese Academy of
GLCM, MSVM Science, 883-893, 2004.
Centroid [2] Shayan Hati and Sajeevan G., “Plant recognition from
Contour leaf image through artificial neural network,”
Distance International Journal of Computer Application, Vol 62,
Z. Wang and Leaf (CCD),
19 [20] No. 17, 2013.
Zheru Chi shape Eccentricity and
Angle Code [3] Gurpreet Kaur and Gurpinder Kaur, “Classification of
Histogram biological species based on leaf architecture,”
(ACH) International Journal of Engineering Research and
Histogram of Development, 2012.
Oriental [4] Sandeeo Kumar E., “Leaf color area and edge features
Gradients based approach for identification of indian medicinal
Zaidah
Ibrahim,
(HOG), Local plants,” Indian Journal of Computer Science and
Binary Pattern Engineering, 2012.
Nurbaity Leaf
20 (LBP) and [21] [5] Vijay Satti, Anshul Satya and Shanu Sharma, “An
Sabri, Nur Shape
Speeded-Up
Nabilah Abu
Robust Features
automatic leaf recognition system for plant identification
Mangshor
(SURF), one- using machine vision technology,” International Journal
against-one of Engineering Science and Technology, 2013.
MSVM [6] S. Anubha Pearline, V. Sathiesh Kumar, S. Harni, “A
Jyotismita Moment- study on plant recognition using conventional image
21
Chaki and Leaf Invariants (M-I)
[22]
processing and deep learning approaches,” Journal of
Ranjan Shape and Centroid- Intelligent and Fuzzy System 36(6):1-8, 2019.
Parekh Radii (C-R) [7] Ji-Xiang Du, De-Shuang Huang, Xiao-Feng Wang,Xiao
Qingfeng Wu, Gu, “Computer-aided plant species identification
Leaf
Changle
Shape, (CAPSI) based on leaf shape matching technique,”
22 Zhon, ANN [23]
Chaonan
Dent and Transaction of the Institute of Measurement and Control
Vein 28(3), 2006.
Wang
Farhana [8] Dhananjay Bisen, “Deep convolutional neural network
Minimum based plant species recognition through features of leaf,”
Haque and Leaf
23 Euclidean [24]
Safwana Shape Multimedia Tools and Applications 80(8):1-14, 2021.
Distance
Haque

ISSN: 2347-8578 www.ijcstjournal.org Page 71

International Journal of Computer Science Trends and Technology (IJCST) – Volume 12 Issue 1, Jan - Feb 2024

[9] Javed Hossain and M. Ashraful Amin, “Leaf shape [18] Adams Begue, Venitha Kowlessur, Fawzi
identification based plant biometrics,” Mahomoodally, Upasana singh, Sameerchand Pudaruth,
DOI:10.1109/ICCITECHN.2010.5723901, 2010. “Automatic recognition of medicinal plants using
[10] Boi M. Quach, V. Cuong Dinh, Nhung Pham, Dang machine learning techniques,” International Journal of
Huynh, Bint T. Nguyen, “Leaf recognition using Advance Computer Science and Applications, Vol. 8, No.
convolutional neural networks based features,” 4, 2017.
Multimedia Tools and Applications 82(1):1-25, 2022. [19] Surleen Kaur and Prabhpreet Kaur, “Plant species
[11] Abdolvahab Ehsanirad and Y.H. Sharath Kumar, “Leaf identification based on plant leaf using computer vision
recognition for plant classification using GLCM and and machine learning techniques.” Journal of
PCA,” Oriental journal of computer science and Multimedia Information System, Vol. 6, No. 2, 2019.
technology 3-6, 2010. [20] Z. Wang and Zheru Chi, “Shape based leaf image
[12] Jesse Dave S. Selda, Roi Martin R. Ellera, Leandro C. retrieval,” IEE Proceedings – Vision Image and Signal
Cajayon, Noel B. Linsanga, “Plant identification by Processing, Vol. 150, 2003.
image processing of leaf veins,” The international [21] Zaidah Ibrahim, Nurbaity Sabri, Nur Nabilah Abu
conference, DOI:10.1145/3132300.3132315, 2017. Mangshor, “Leaf recognition using texture features for
[13] S. Mouine, I. Yahiaoui, A. Verroust-Blondet, “A shape- herbal plant identification,” Indonesian Journal of
based approach for leaf classification using multiscale Electrical Engineering and Computer Science, Vol. 9,
triangular representation,” In proceeding of the 3rd ACM No. 1, 2018.
conference on International Conference on multimedia [22] Jyotismita Chaki and Ranjan Parekh, “Plant leaf
retrieval, pp. 127-134, 2013. recognition using shape based features and neural
[14] A. Arkif and M. F. Khan, “Automatic classification of network classifiers,” International Journal of Advance
plants based on their leaves,” Biosystems Engineering, Computer Science and Applications, Vol. 2, No. 10,
Vol. 139, pp. 66-75, 2015. 2011.
[15] J. Carranza-Rojas and E. Mata-Montero, “Combining [23] Qingfeng Wu, Changle Zhon, Chaonan Wang, “Featuire
leaf shape and texture for costa rican plant species extraction and automatic recognition of plant leaf using
identification,” CLEI Electronic Journal, 19(1), paper 7, artificial neural network,” Advance en Ciencias de la
2016. Computacion, pp. 5-12, 2006.
[16] Yeni Herdiyeni and Ni Kadak Sri Wahyuni, “Mobile [24] Farhana Haque and Safwana Haque, “Plant recognition
application for indunesian medicinal plants identification system using leaf shape features and minimum
using fuzzy local binary pattern and fuzzy color Euclidean distance,” ICTACT Journel on Image and
histogram,” International Conference on Advance Video Processing, Vol: 09, Issue: 2, 2018
Computer Science and Information System, 2012.
[17] Andres Hernandez-Serna and Luz Fernanda Jimenez-
Segura, “Automatic identification of species with neural
networks,” PeerJ 2: e563, DOI 10.7717/peerj.563, 2014.

ISSN: 2347-8578 www.ijcstjournal.org Page 72