International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056

Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

Color, Size, Volume, Shape and Texture Feature Extraction

Techniques for Fruits: A Review
Manali R. Satpute1, Prof. Sumati M.Jagdale2
1PG Student, Department of Electronics & Telecommunication, Bharati Vidyapeeth’s College of Engineering for Women,
Pune-43, Maharashtra, India.
2Associate Professor, Department of Electronics & Telecommunication, Bharati Vidyapeeth’s College of Engineering for
Women, Pune-43, Maharashtra, India.

---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Grading and sorting of agricultural products factors[1] are vitamins, minerals, nutrients, carbohydrates
such as fruits and vegetables are very important in post- present in that fruit. The sorting and grading based on size is
harvesting process. In manual sorting efficiency totally easy process it can be done manually also but sorting and
depends upon the operators, so it is very time-consuming, grading based on other external and internal factors is not
costly, less efficient process. While exporting to market, quality possible with operators so there is need of automatic fruit
of fruits must be very good. Grading of fruits is very important
quality inspection system which gives accurate result.
because it fetches a high price in a market. Grading and
sorting are done based on the fruits size, color, volume, shape
and texture. In this review paper, different methods of size, 2. MOTIVATION
color, volume, and shape and texture detection are discussed.
For color detection direct color mapping method, for volume Sorting large of fruits manually is a time consuming
detection water displacement method and Monto Carlo process. As the sorting depend upon skill of particular
method, for texture detection Direct curvelet trans-form and
person, skill varies from person to person so it is not
Gabor wavelet transform, for shape detection Discrete wavelet
transform, Fourier transform of a body, scale space, chain code accurate process Probability of error occurrences is more in
and clustering methods are discussed. manual sorting and cost is also high. To fulfill International
market demand of fruits these agro industries must work
Key Words: Grading, DCT, DWT, RGB, HSV, Sorting, Image with greater accuracy, more consistently and very efficiently.
processing. So automation is necessary in post-harvest process such as
sorting and grading.

1 .INTRODUCTION 3. METHODS
Agriculture has an important role in socio-economic
Grading and sorting is done based on external and internal
development of India. Various types of fruits produced quality factors. The external factors are color, size, volume,
through-out the year. There are two parts pre-harvesting shape and texture. There various methods used for color
and post-harvesting. The post-harvesting process includes detection, size and volume calculation, shape, and for skin
sorting and grading of fruits. In India, mostly fruit grading defect detection. But main condition is that the grading and
did manually. But manual grading is costly, time-consuming, sorting method must be nondestructive.
inconsistent and less efficient. The export market demands
high quality graded fruits. So there needs to develop a 3.1 Feature Extraction Based on Color
system which overcomes the problem of labor shortage, save
time, and improve graded products quality. There are many
Color is the most important parameter in defining the
factors based on that the grading of fruits is done. These
quality of any fruit. Color determines maturity of fruit. Most
factors can be classified into two type’s external quality
of the times consumer decision to accept or reject a
factor and internal quality factor. The external quality factors
particular fruit is depends upon its color. Most of the existing
are size, shape, color, surface defects which can easily see
technology determines color of fruit by comparing the fruit
with open eyes. But the internal quality factors are like
color with the existing predefined reference colors [2].
aroma, test, flavors, sweetness, another internal quality
Y. Wang et al (2010) [5] designed a system in which Fruit

© 2016, IRJET ISO 9001:2008 Certified Journal Page 703

 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

quality inspection is done based on fruits surface color. It is Histogram matching: In this method first step is
nondestructive method of fruit inspection. Fruit image is conversion of color image to grey image. Then histogram of
taken first with camera; this image is in RGB color format. At grey image is taken. The database contain sample of
next stage this RGB image is converted into the HSI color histograms of ripe fruits. Then obtained histogram is
model. After that image is segmented based on hue value and compared with database samples and ripeness is detected
then divide the fruit and its background. Then histogram of [6].
Hue and Saturation of fruits surface color is calculated. Back Parameter based segmentation: According to [8] here RGB
propagation network is designed, for this input is given as color space is used all the process is done on this original
histogram output obtained earlier of Hue and Saturation of color image. First the value of Red Green and Blue are set for
surface color of fruit. The Back Propagation gives the output ripeness detection. For particular fruit particular color is
as quality description of given tested fruit. Back propagation selected for example if ripe tomato has red color and unripe
is three layer network used here. Training of Back have green color. So to detect ripeness of tomato only red
propagation network is performed. They performed
color is segmented from the RGB color space. As per
experiment on banana and result obtained is accurate.
requirement color decided and based on some rules ripeness
Dah-Jye Lee et al.(2011) [2]designed a Direct color map-
is detected.
ping technique for obtaining the color of fruit and based on
Clustering algorithm is useful if the quantity of
that grading of fruit is done. In existing technology there is
information is more it only depends upon type of distance.
no such arrangement of adjusting color preferences or Histogram matching provides better result. Clustering
grading parameters as per application. Direct color mapping algorithm and histogram matching both requires color
overcomes this problem, so it is user friendly technique. In transform. But parameter based segmentation not required
this method, the given image of color space RGB. The color of color transform so this is efficient method for ripeness
interest is decided first from given range in this method for detection.
example tomato maturity steps are taken as example it
varies from dark green to dark red, and in between that 3.2 Feature Extraction Based on Size and Volume
green, orange, light red .Then these preselected colors used
to calculate a unique set of coefficients. These coefficients The cost of many products is directly related to their size.
are used for color space conversion. The 3 dimensional RGB The size calculation for regular shapes is easy, but it
space is converted into a small set of color indices as per becomes more complex in fruits and vegetables with
irregular shapes. Parameters used for size calculation are
requirement of application.95% accuracy is achieved in
area, length, width and perimeter. Size can be calculated by
grading application in variety of color fruits because of various ways. In one-dimensional based on length and width.
adjustment of color preference. In two-dimensional based on area and perimeter and in
M. Dadwal and V.K. Banga (2010) [6] done a review on three-dimensional using volume and surface area [3]. Fruits
Fruit ripeness detection based on color image segmentation. sorting and grading based on volume is very complex part as
Various techniques are available out of that Clustering volume calculation is three-dimensional measurement [3].
algorithm, Histogram matching and parameter based H. Dang et al (2010)[17] developed a system for fruit size
segmentation are given in this paper. detection. This system is based on the image processing. The
Clustering algorithm: In that same elements are grouped basic steps performed in this are image processing, then
together is called as clustering. All this is done step by step edge detection ,then fruit size detection and based on size
process that is called as algorithm. Here Clusters of similar grading of fruit is done. In image processing main step is
color pattern are calculated. There are various partitioning filtering of image ,in this faster median filter algorithm is
clustering algorithm available [7] K- means; Fuzzy c means used for removing noise and because of that good edge is
(FCM), Gustafson Kessel improved by Babuska (GK-B), and detected. In edge detection, first the image taken is
Gustafson Kessel Possobilistic Fuzzy c Means (GKPFCM). The converted into grey image. Then OSTU (maximum classes
method for performing this image is partitioned, clustering is square error) is used to get binary image directly. For edge
performed of similar color pattern, and then distance is sequence detection 8-connected boundary method is used.
calculated in between a pixel and cluster center, it is squared Then fruit size detecting algorithm in order to calculate size
or absolute difference. If spatial information is added in this of fruit. In order to calculate the diameter fruits natural
then identification is more efficient [6]. symmetry is taken into consideration. From symmetry

© 2016, IRJET ISO 9001:2008 Certified Journal Page 704

 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

center coordinate of fruits shape and axis in fruits image taken of that fruit. The image is with RGB color space and in
obtained. As center coordinate and axis are found , a line dimensions 640*480 and resolution 96dpi in vertical and
through center point which is vertical to line from axis point horizontal direction [3]. In processing the image with RGB
to center point. From this line diameter is calculated. For color space is converted to HSV color space. Because image
accurate result two edge points are searched. If fruit is segmentation is easy in HSV, fruit image can easily separate
rotated then also diameter shown is same then the diameter out from the background. Then in image segmentation is
indicated true fruit size. Then based on accurate size grading done .In that image is segmented into interested region and
is done [17]. background. The image formed in image segmentation is
binary image.Thresholding is applied here, one threshold is
Water Displacement method uses principle of Archimedes
decided if the pixel grey scale value is greater that threshold
[3]. This method is used for food volume measurement in
then it assigns as binary 1(white) otherwise o(black)[3].
early years when no technology is available. Hardware
required is jar showing contents of water inside it. The fruit
is dropped into the jar and rise in water level is calculated.
The difference between rise level and original level
represents volume of that fruit. But drawbacks are, this
process is time consuming and no guarantee that obtained
result is accurate [3]. This method is destructive to a fragile
object and if the product is porous it absorbs the water when
dropped in the water so shown result is inaccurate. [3].
J.Siswantore et al (2014) [3] developed a monto carlo
method. This method is used for irregularly shaped food
volume measurement with heuristic adjustment. The
hardware used for this a multiple camera for image
acquisition, a computer, a light source and a black
background as shown in Fig.1.

Fig -2: Flowchart of the Monto Carlo method [3]

Then in volume measurement, bounding box construction is

first step in this determine 3D bounding box of measured
object . From 3 D bounding box upper and lower co-
ordinates of x, y, z direction, of real world coordinate system.
The binary image and camera parameters help to determine
bounding box. The main condition is that the measured
object must be fit in the bounding box [3]. Then 3 D random
points are generated. Then Monto Carlo integration is done
to obtained volume. Various formulas are present for volume
calculation then heuristic adjustment is done because five
cameras are present to capture image. But problem facing is
that the object point selected may be wrong it may present
Fig -1: Camera Calibration systems [3][4] outside of object. If we consider wrong object point in
volume calculation then obtain result is wrong so here
Fig.2. shows the flow of Monto Carlo method. Then heuristic adjustment method is applied for correct result of
camera calibration is done. The main aim of camera volume [3]. In validation each sample is measured three
calibration is to obtain extrinsic and intrinsic camera times and mean is calculated. Absolute relative error (ARE)
parameters from taken 2-D image. Intrinsic and extrinsic and coefficient of variation were used to measure accuracy
camera parameters transform a point in real world co- and precision [3].
ordinate system to image co-ordinate system. Then in image Another method for volume calculation is color image
acquisition the fruit is located on plane z=0 as shown in fig.1. segmentation method [4] in that different methods are
As five camera placed at different position five images are available. In Region base technique image is divided into
© 2016, IRJET ISO 9001:2008 Certified Journal Page 705
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

different region based on some rule such that each region copied first and then pasted as row element in new
contains pixels with same grey scale value. In edge detection matrix [13]. This all process repeated up to 360
is generally used for detection of discontinuities boundary degree [13].
such as corners, lines and curves. In split and merge IV. Scale space: - Scale space means dealing with image
technique, split is performed when image is with too much structure at different scale. Scale space is very good
method for rep-resenting important object features.
variety. Split is performed with thresholding and In merge
Low pass Gaussian filter with variable widths is
technique, two regions are joined together if they are similar applied to track the position of infection point in
and adjacent. There are different algorithms are also representation. The infection point remained in
available for volume calculation such as image analysis representation were significant object
algorithm and Canny edge detection algorithm [4]. characteristics. Multiscale filtering helps to separate
detail of shape properties [13].
3.3 Feature Extraction Based on Shape V. Recognize boundaries of objects: - The image is two
dimensional arrays. The array contains binary
elements. Boundaries of object recognized based on
Shape is also most important parameter for selection of starting point which is considered as single pixel on
healthy fruit. There are various method available for Fruit the object background interface. First interest point
shape detection. is selected like this and then for finding other pixels,
I. Discrete Wavelet Transform (DWT):- Image is made moved in clockwise and anti-clockwise direction.
up of pixels. These pixel are arranged in two The pixels searched diagonally. The object
dimensional matrix form. Each pixel location boundary can be recognized by hunting other pixels
represents digital equivalent value of image in a fixed direction [13].
VI. Clustering: - Clustering means grouping similar
intensity. Wavelet is a function in Digital image
things together. The pixels in one group are more
processing and is mostly used for image
similar to each other than other group. Clusters
compression [13]. contain group of pixels values with small variation
II. Dipalee Gupta et al (2015) [9] developed a image [15].
compression technique by using Discrete wavelet VII. Chain code: - Different shapes such as line ,
transform. Here Discrete wavelets are used such as drawings, planer curves can be identified with chain
Haar Wavelet and Daubechies wavelet. It is used for codes. It is lossless compression algorithm. For each
compression in still image. Wavelets perform region, the coordinated of any point on the
compression of image so that wavelets are formed boundary are transmitted. The encoder moves
and space required for image storage is reduced along the boundary, at each stage encoder transmits
[13]. Wavelets have not uniform edges. They are any symbol that represents direction of movement.
able to provide better picture by removing noise This process continues up to when encoders return
pixels. In spatial domain the neighboring pixel value to starting position. Because of this blob is
highly depends upon each other and therefore completely detected, and process continues for next
redundant. To compress image these redundancies blob detection [16].
between pixels must remove. Discrete wavelet
transform converts spatial domain pixels into the
3.4 Feature Extraction Based on Texture
frequency do-main, representation is in multiple
sub-band format in different time scale and
frequency points [13]. The color recognition process is very important for
III. Fourier Transform of Boundary: - For shape ripeness detection. But instead of that fruits external surface
description method uses Fourier transform of one defect detection is also an important process. For that
dimensional boundary representation and Region texture analysis is useful.
based Fourier transform. In one dimensional S. A. khoje et al [10] developed a system of fruit grading
Fourier descriptors, Fourier transform is applied on for automated skin defect identification using Discrete
shape signature which is one dimensional function.
Curvelet Transform. Texture analysis is mainly used for fruit
Shape signature is derived from shape boundary
coordinates. Fourier descriptors of the shape are quality detection. Texture analysis is done by Discrete
normalized Fourier transformed coefficients [13]. In Curvelet transform. Discrete curvelet transform uses multi
Region based Fourier descriptors performed in two resolution approach, based on that local and global features
steps, in first step the approximated normalized of that fruit with the help of low and high resolution
image counter clockwise rotated with very small capability of DCT are taken .Feature such as Energy, entropy,
angular step. Then in second step from the image
mean and standard deviation of each good and defected fruit
center to the positive X-axis all pixel values are
is calculated. This four characters are used to characterize

© 2016, IRJET ISO 9001:2008 Certified Journal Page 706

 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

fruits surface texture. The obtained features are then further automation in fruit quality inspection.
given to Support vector machine and Probabilistic neural
network for classification of good fruits and defected fruits.
Based on obtained result SVM is more accurate than PNN ACKNOWLEDGEMENT
[10].
H. Alimohamadi et al (2013)[11] designed a system for I would like to thank my all the staff members of
BVCOEW,Pune for being moral support through the period of
skin defect detection in fruits. For this they use a robust
my project study in BVCOEW, Pune whose help and shared
algorithm called Gabor wavelet Filter. For analysis of skin knowledge was the main support to my project.
defect they first convert color image into texture image. And
then on that obtained texture image Bank of Gabor filter is REFERENCES
applied. Gabor Filter is linear filter and mostly used for edge
detection. A filter bank consisting of Gabor filters with 4 [1] M. Alfatni , A. Shariff , M. Abdullah , M Hamiruce B.
scales and 6 rotations. From the response of Gabor filters Marhaban , Osama M. Ben Saaed , The application of
internal grading system technologies for agricultural
image pixel is classifies as defected or normal skin. Optimal
products Review , Journal of Food Engineering 116(2013),
filter is chosen from bank of Gabor filters depending upon pp.703-725.
the response of Gabor filter. Last stage performed is doi:10.1016/j.jfoodeng.2013.01.001
thresholding the response of the optimal filter. Based on [2] D.Lee, J.Archibald and G.Xiong, Rapid Color Grading for
thresholding skin defect is detected. [11] Fruit Quality Evaluation Using Direct Color Mapping , IEEE
Transactions On Automation Science And Engineering, Vol
A. Majumdar et al (2009)[12] done a comparative study of
8, No.2 April 2011,pp.292-302.
curvelets , wavelets and contourlets. Curvelets, wavelets and doi-10.1109/TASE.2010.2087325
contourlets used as a feature sets. Mostly used in pattern [3] J.Siswantoro, A. Prabuwono, A. Abdullah and B. )drus,
recognition. In this curvelets and contourlets are multi- Monte Carlo Method with Heuristic Adjustment for
resolution multidirectional transforms, they are used in face )rregularly Shaped Food Product Volume Measurement ,
and character recognition [12]. Comparison result obtained Hindawi Publishing Corporation The Scientific World
is, for higher resolutions the wavelets used is a good option Journal Volume 2014, Article ID 683048,pp.1-10.
as a feature set for facial as well as character image. But as doi: http://dx.doi.org/10.1155/2014/683048
the resolution starts decreasing wavelets not work [4] H. Kachariya,K. Vasaniya, S. Dhameliya, R. Savant, A
efficiently as feature descriptors. But at lower resolution Review of Volume Estimation Techniques of Fruit ,
International Journal of Engineering Research &
curvelets work very efficiently. Contourlets not useful in
Technology (IJERT) , Vol. 4 Issue 01,January-2015,pp.217-
higher as well as lower resolution, it only works in very 221.
coarse resolution. Recognition capabilities of all these three [5] Y. Wang, Y. Cui, G. Huang, P. Zhang, S. Chen, Study on
are tested using KNN classification [12]. Fruit Quality Inspection Based on Its Surface Color in
Produce Logistics , )nternational Conference on
Manufacturing Automation,pp.107-111.
4. CONCLUSIONS Doi: 10.1109/ICMA.2010.47

This review paper analyses that, In India normally [6] M.Dadwal and V. Banga, Color )mage Segmentation for
grading is done manually. The grading and sorting is mainly Fruit Ripeness Detection: A Review , nd International
based on external and internal quality factors. The external Conference on Electrical Electronics and Civil Engineering
factors are color, size, volume, shape and texture. Among )CEECE’ Singapore, April 8-29, 2012, pp.190-193.
these color and size are mainly used for features for grading [7] B. Ojeda- Magana, R. Ruelas, J. Quintanilla-Domingvez and
of fruits. Grading based on Size is very easy method and less D. Andina, Colour )mage Segmentation by Partitional
expensive method used for sorting of apples, tomatoes etc. In Clustering Algorithms , pp. 8 8-2833.
[8] C. Lin , C. Su, (. (aung and K. fan , Colour )mage
color based grading Direct color mapping technique is
Segmentation Using Relative Values of RGB in Various
flexible and efficient method. For skin defect detection at
)llumination Circumstances , )nternational Journal of
higher resolution wavelets are used and at low resolution Computers, Issue 2, Vol-5, 2011,pp.252-261.
curvelets are best option. The grading based on size
manually can be performed but result obtain are not
accurate and grading and sorting based on other external
factor is not possible to done manually. So there is need of [9] D. Gupta, S. Choubey, Discrete wavelet transform

© 2016, IRJET ISO 9001:2008 Certified Journal Page 707

 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 04 | Apr-2016 www.irjet.net p-ISSN: 2395-0072

for )mage processing , )nternational journal of

Emerging Technology and Advanced Engineering,,
Issue 3, Vol.4, March 2015,pp.598-602.
[10] S. Khoje, S. Bodhe, A. Adsul, Automated Skin Defect
Identification System for Fruit Grading Based on
Discrete Curvelet Transform , )nternational Journal
of Engineering and Technology (IJET). Vol 5, No.4,
Aug-Sept 2013, pp.3251-3256.
[11] (. Alimohamadi, A Ahmadyfard, Detecting Skin
Defect of Fruits Using Optimal Gabor Wavelet
Filter , )nternational Conference on Digital )mage
Processing, pp.402-406.
[12] A. Majumdar , A. Bhattacharya, A Comparative
Study in Wavelets, Curvelets and Contourlets as
Feature sets for Patterns Recognition , The
International Arab journal of Information
Technology, Vol 6, No.1, January 2009, pp.47-51.
[13] D. Jasani, P. Patel , S. Patel, B. Ahir, K. Patel, and M.
Dixit, Review of Shape and Texture Feature
Extraction Techniques for Fruits , )nternational
Journal of Computer Science and Information
Technologies, Vol.6, 2015,pp.4851-4854.
[14] S. Loncaric, A Survey of Shape Analysis
Techniques , Pattern Recognition, Vol. , No.8,
1998, pp.983-1001.
[15] D. Kakadiya, N. Shah, M. Patel, R. Shah, C. Kachariya,
K. Sukhwani, Shape Extraction Methods for Fruits:
Technical Review , )nternational Journal of
Computer Applications(0975 8887) , Vol.111, No.1,
Feb.2015, pp.43-48.
[16] http://en.wikipedia.org/wiki/Chain_code
[17] (. Dang, J. Song and Q. Guo, A Fruit Size Detecting
and Grading System Based on )mage Processing
2010 Second International Conference on
Intelligent Human- Machine systems and
Cybernetics, pp.83-86.
DOI: 10.1109/IHMSC.2010.120

© 2016, IRJET ISO 9001:2008 Certified Journal Page 708