Classification of Banana with using Digital Image Processing

7/23/2019 Classification of Banana with using Digital Image Processing

1/58

120280723002

i

Quality Analysis and Classification of Bananas UsingDigital Image Processing

By:

Mr. Sanjay Chaudhary

Enrolment Number: 120280723002

M.E. (Information Technology)

Guided By:

Prof. Bhavesh PrajapatiAssistant Professor,

Information Technology Department,

L. D. College of Engineering,

Ahmedabad-15

A Thesis Submitted to

Gujarat Technological University

In Partial Fulfillment of the Requirements for

The Degree of Master of Engineering

in Information Technology

MAY 2014

Computer Engineering Department,

L. D. College of Engineering,

Ahmedabad-15


2/58

120280723002

ii

CERTIFICATE

This is to certify that research work embodied in this thesis entitled Quality Analysis

and Classification of Bananas Using Digital Image Processing was carried out by

Mr. Sanjaykumar R. ChaudharyEnrollment No. 120280723002 at L. D. College of

Engineering for partial fulfillment of M.E. degree to be awarded by Gujarat

Technological University. This research work has been carried out under my supervision

and is to my satisfaction.

Date: 12/05/2014

Place: Ahmedabad

Prof. Bhavesh Prajapati Prof. Dhaval Parikh

Assistant Professor, Associate Professor and Head,

I.T. Department Computer/I.T. DepartmentL.D. College of Engineering L.D. College of Engineering

Ahmedabad. Ahmedabad.

Dr M. B. Dholakia

Principal,

L.D. College of Engineering

Ahmedabad.


3/58

120280723002

iii

COMPLIANCE CERTIFICATE




Engineering (028) for partial fulfillment of Master of Engineering in Information

Technology degree to be awarded by Gujarat Technological University. He has compiled

to the comments given by the Mid Semester Thesis Reviewer to my satisfaction.

Date: 12/05/2014

Place: Ahmedabad

Mr. Sanjaykumar R. Chaudhary Prof. Bhavesh Prajapati

(120280723002) Assistant Professor,

I.T. Department

L.D. College of Engineering

Ahmedabad.


4/58

120280723002

iv

THESIS APPROVAL



Mr. Sanjaykumar R. Chaudhary Enrollment No. 120280723002 at L. D. College of

Engineering (028) is approved for award of the degree of Master of Engineering in

Information Technologyby Gujarat Technological University.

Date:

Place:

Examiner(s):

---------------------------- -------------------------- --------------------------( ) ( ) ( )


5/58

120280723002

v

PAPER PUBLICATION CERTIFICATE




Engineering (028) for partial fulfillment of Master of Engineering in Information

Technology degree to be awarded by Gujarat Technological University, has been

published/presented at IJARCSSE- January 2014 and accepted at IJCSE-April 2014.

Date: 12/05/2014

Place: Ahmedabad

Mr. Sanjaykumar R. Chaudhary Prof. Bhavesh Prajapati

(120280723002) Assistant Professor

I.T. Department,L.D. College of Engineering

Ahmedabad.

Dr M. B. Dholakia

Principal

L.D.College of Engineering

Navrangpura, Ahmedabad


6/58

120280723002

vi

DECLARATION OF ORIGINALITY

I hereby certify that I am the sole author of this thesis and that neither any part of

this thesis nor the whole of this thesis has been submitted for degree of any other

University or Institution.

I certify that, to the best of my knowledge, my thesis does not infringe upon

anyones copyright nor violate any proprietary rights and that any ideas, techniques,

quotations, or any other material from the work of other people included in my thesis,

published or otherwise, are fully acknowledged in accordance with the standard

referencing practices.

I declare that this is a true copy of my thesis, including any final revisions, as

approved by my thesis review committee.

Date: 12/05/2014

Place: Ahmedabad

Mr. Sanjaykumar R. Chaudhary

(120280723002)

Verified

Prof. Bhavesh Prajapati

Assistant ProfessorI.T. Department,L.D. College of Engineering

Ahmedabad.


7/58

120280723002

vii

Acknowledgement

I wish to express my deep sense of gratitude to my Guide, Prof. BhaveshPrajapati, Assistant Professor, I.T. Department, L. D. College of Engineering for his

great efforts and instructive comments in the dissertation work.He has helped me plan the work from the beginning of the research work. He

allowed me a great deal of freedom to choose a research topic. Also, he took care to help

me narrow down the choices for the topic. Without his valuable support, it would have

been very difficult to do quality and directional work.

I would also like to extend my gratitude to Prof. Hiteishi Diwanji for their

invaluable and precious guidance and continuous encouragement and motivation.

I would also like to thank my family and my friends at L. D. College of

Engineering and Ashish Chaudhary for their support and help during the Research work.

I would also like to thank expert of bananas Nirav Chaudhary for their support and

help during the Research work.

Sanjay Chaudhary


8/58

120280723002

viii

Table of Contents

Title Page ........................................................................................................................i

Thesis Approval .............................................................................................................ii

Certificate .....................................................................................................................iii

Compliance Certificate................................................................................................. iv

Paper Publication Certificate ............................................................................................v

Declaration of Originality ............................................................................................vi

Acknowledgement .......................................................................................................vii

Table of Content .........................................................................................................viii

List of Figures ................................................................................................................x

List of Tables ................................................................................................................xi

Abstract .......................................................................................................................xii

1. Introduction.............................................................................................................. 1

1.1. Introduction.....................................................................................................2

1.2. Problem Statement ...........................................................................................3

1.3. Aim of Dissertation ..........................................................................................3

1.4 Organization of Thesis.......................................................................................3

2. Literature Review..................................................................................................... 4

2.1. Bananas Properties ...........................................................................................5

2.2. Consumer Characteristics ..................................................................................5

2.3. Color Spaces ...................................................................................................6

2.3.1. RGB Color Model ................................................................................. 6

2.3.2. CIELAB color space ..............................................................................6

2.3.3. HSV color space ...................................................................................7

2.4. Segmentations Methods ...................................................................................92.4. Study of Papers ................................................................................................9

3. Existing Systems ...................................................................................................... 11

4. Proposed Methodology............................................................................................. 14

4.1. Proposed Algorithm ...................................................................................... 15

4.2. Methodology ................................................................................................. 15


9/58

120280723002

ix

4.2.1. Image Acquisition...................................................................................... 16

4.2.2. Image Segmentation ..................................................................................18

4.2.3. Feature Extraction.................................................................................... 19

4.2.4. Classification ............................................................................................20

4.3. Discriminate Power of Selected Features.......................................................... 22

5. Results and Discussion.............................................................................................. 25

5.1. Classification Performance.............................................................................. 26

5.2. Comments Resolution..................................................................................... 26

Conclusion and Future Work ..................................................................................... 27

Publication ................................................................................................................... 28

References .................................................................................................................... 29


10/58

120280723002

x

List of Figures

Figure 1: Seven stages of Banana ....................... ................................ ......... ................ 2

Figure 2: CIELAB color space ....................... ......... ............. ....................................... 7

Figure 3: HSV color space ............................................................................................ 8

Figure 4: Mapping of function to stage of bananas ..................... ........... ............ .......... 12

Figure 5: Process of classification of banana ............ ......................................... .......... 16

Figure 6: Pop up message allows you to select way of input ....................................... 16

Figure 7: Pop up message allows you to select demo image ........... ............................. 16

Figure 8: Pop up window allows you to select input image ......................................... 17

Figure 9: Image given to algorithm for classification .................................................. 17

Figure 10: Pop up allow user to select background color ............ ................................. 18

Figure 11:Figure Window allow user to select region of background ............ ............... 19Figure 12:Figure window of segmented image ............................................................ 19

Figure 13: Classifier for Proposed Scheme ................................................................... 21

Figure 14: Two Bunch of bananas at different stages .................................................... 22

Figure 15: Segmented image of Green part .................................................................. 23

Figure 16: Segmented image of Yellow part ................................................................ 23

Figure 17: Segmented image of Brown part .................................................................. 24


11/58

120280723002

xi

LIST OF TABLE

Table 1:Relationship between stages of bananas and physical characteristics ................. 5

Table 2: Scale related to consumer characteristics ........................................................... 6Table 3: Mean value Segmentation Table ..................................................................... 18

Table 4: Matrix as input for classifier ........................................................................... 20Table 5: Classification Criteria Table ............................................................................ 20

Table 6:Performance Index Table ......... ....................... ....................... ......... ............... 26


12/58

120280723002

xii

Abstract

Traditionally quality inspection of banana can be done by two ways: either

instrumental tools or human inspectors. Instrumentation technique, in the case of bananas

these techniques are usually destructive requiring the removal and flattening of the peel

for the measurement where decisions taken by human inspectors may be affected byexternal factors like: tiredness, bias, revenge or human psychological limitations. An

interesting alternative is image processing can overcome limitations of these two

techniques. Digital Image processing can classify the banana fruit with speed and

accuracy. Good algorithms available in image processing which can classify banana

based on their color and texture characteristics, but limitation is that they can work well

on single banana. Bananas is sold in bunch of dozen and thats why it is important to

analyze quality in bunch. This paper proposes the technique of digital image processing to

classify the bananas hand. Features including CIELAB, and HSV colorspaces have been

used for classification of bananas images. Result shows better accuracy of proposed

algorithm.

Author: Sanjaykumar R. Chaudhary

Enrolment Number: 120280723002

Degree: Master of Engineering

Department: Information Technology

Guide: Prof. Prof. Bhavesh Prajapati

Designation: Assistant Professor

Department: I.T. Department


13/58

120280723002

1

Chapter 1

Introduction


14/58

120280723002

1.1 Introduction

Bananas are the forth most important staple crops in the world and India is the leadingcountry in the production of banana [1]. Classification of bananas is important task for

banana industry. Classification of bananas can be done by basically three methods: (1)Human visual inspection; (2) Instrumental techniques; (3) Computerized image analysis

techniques.Human visual inspection is a highly subjective, tedious, time-consuming and labor-intensive process. By contrast, instrumental techniques (i.e., colorimeters) allow accurateand reproducible measurements of the colors not influenced by the observer. However,their main disadvantages are that the surface color must be quite uniform and that manylocations on the sample must be measured to obtain a representative color profile. Inaddition, in the case of bananas these techniques are usually destructive requiring theremoval and flattening of the peel for the measurement. An interesting alternative iscomputerized image analysis techniques (also known as computer vision systems), whichovercome the deficiencies of visual and instrumental techniques and offer an objectivemeasure for color and other physical factors [8].

Development in the field of image processing especially in its field image segmentationwhich is used to extract regions of interest has proven wonders in various applications

like Signature verification, Face recognition, Thumb impression verification, Automaticcharacter recognition, Industrial machine vision for assembly and inspection etc. But the

potential of image segmentation in the field of agriculture is yet to be exploited for thedaily use. Image Processing can use to analyze the fruit quality on the basis of its color,

size and weight. [9]

A mainly color change in banana during ripening is based on the peel color rather than thepulp color and hence color of banana peel has been used in the assessment of the stages ofripeness of banana. Commercial standard color charts are available in which 7 stages of

peel color were reproduced and translated to a numerical scale where Stage 1=all green,

2= green with trace of yellow, 3= more green than yellow, 4= more yellow than green, 5=yellow with trace of green, 6= full yellow, 7= full yellow with brown spots [3].

Figure 1: Seven stages of Banana [3]


15/58

120280723002

3

1.2 Problem Statement:

Today quality inspection of bananas in industry is mostly done by human inspectors, due

to limitation of instrumental tool and image processing doesnt have algorithm to classify

bananas hand (bunch of bananas). Mostly image processing is used for classification of

other fruits and grain but bananas characteristics are different from other fruits.

-

Bananas always sold in group or bunch of dozen where are other fruits soldsingly.

-

Bananas are important at all level, means people purchase bananas which are at

any level from 1 to 7, where other fruits are important at mature level, like for

apple mature stage is red, for orange maturity stage is orange, similarly for other

fruits have maturity stage.

- Bananas have maturity stage at stage 6 and 7, but people can purchase bananas at

different stage for backing use.

- According to experts if single banana is changed their stage higher from other

bananas than we have to classify that bunch of bananas to higher stage.

-

Listed algorithms are work on average of whole image thats why it will missclassifying bananas bunch in the above cases.

- In case as discussed in section 4.3 others systems mostly miss classify stage of

bananas.

- Also other system didnt take consumer characteristics in mind for selection of

features or in the classification process.

1.3 Aim of Dissertation:

The objectives of this study were: (i) To implement a standardized computer visionsystem for classification of bananas by characterize quantitatively color changes duringripening using the sRGB, L*a*b* and HSV color spaces; (ii) To identify features of

interest which can be related with the ripening stages such as color and textural featuresof the images, and; (iii) To develop a statistical model using selected features to identifythe seven ripening stages of bananas from samples previously classified by expert visualinspection. We were taken consumer behavior in mind to selection of features, like:%BSA is factor which consumers mostly take in mind to purchase banana. To increaseimportance of algorithm we can provide extra information related to stages of bananaslike: chemical, mechanical properties of bananas, those properties are change with changein stage of bananas.

1.4 Organization of Thesis:

Thesis is organized in 5 chapters. 1stchapter gives introduction and brief idea about the

problem. 2ndchapter covers all the basic concepts of literature review. 3rdchapter covers

existing work. Proposed work and implementation are in 4th

chapter. In chapter 5th

results

collected in the research. After that conclusion, references, and publication are covered.


16/58

120280723002

4

Chapter 2

Literature Review


17/58

120280723002

5

For selection of features of image processing it is important to first we know little aboutbananas properties and consumer characteristics. We were analyzing bananascharacteristics using image processing color space and mapping to different stage of

bananas in this process we consider consumer characteristics to increase the importanceof system.

2.1 Bananas Properties

Bananas physical, chemical and mechanical properties are change from stage 1 to 7, andrelated to that color properties of bananas also change, so that if we can successfullymeasure the stage of bananas then we can get approximately right information of

physical, mechanical and chemical properties. Skin color changes from green to yellow,firmness is decreased, fruit gets softened and starch is converted into sugar. A mainlycolor change in banana during ripening is based on the peel color rather than the pulpcolor and hence color of banana peel has been used in the assessment of the stages ofripeness of banana [3].

ParametersStage of Ripening

5 6 7

Pulp/Peel Ratio 2.0 2.3 2.7

Peel ColorYellow with Trace

of GreenAll Yellow

Yellow with brownspots

Pulp Color White White Creamy Yellowish Creamy

Table 1:Relationship between stages of bananas and physical characteristics

As the ripening proceed, pulp to peel ratio was increased from 2.0 in stage 5 to 2.7 instage 7 when the fruits become fully ripened. (Table1). This could be due to the osmotictransfer of moisture from the peel to the pulp as sugar content of pulp increased. It has

been suggested that pulp to peel ratio can be considered as a coefficient of ripeness. Theintensity of greenness of the peel also decreased from stage 5 to stage 7 [3].

2.2 Consumer Characteristics:To examine consumer purchasing behavior five characteristics are important: damage(dents and breaks in skin), markings (includes russet, waxy build-up, etc), brilliance(degree of shine), maturity/color, and one less used is bruising. The peel color of bananasis considered as the first quality parameter evaluated by consumers. Fruit maturity/colorwas based on the ground color of banana. The ground color is referred to the base color ofthe fruit; a green ground color could reflect immaturity, as in bananas; and more brownspots reflect over-ripeness of bananas. Final classification after calculation of measure: Aquality measure was calculated by summing together the values for bruising, marking,and damage. All the quality measure values were summed together to create a totalquality value. [6]. Development of spots expressed as %BSA (brown spots as a

percentage of the total area) [8].

Once the quality characteristics to be examined were identified, an assessment processwas developed. For bruising, markings, and damage the same evaluation technique andscale values were used [6].


18/58

120280723002

6

Scale

Fruit Appearance (For

bruising, markings, anddamage)

Standard (brilliance)

4 Very shiny lookingLess than 10 % of the fruit on display has the

negative quality characteristic present

3 Shiny looking

10 to 30 % of the fruit on display has the


2 Glossy looking30 to 50 % of the fruit on display has the


1 Dull lookingGreater than 50 % of the fruit on display has the

negative quality characteristic

Table 2: Scale related to consumer characteristics

2.3 Color Spaces

Color plays a major role in the assessment of external quality in food industries and foodengineering research. Color is basically specified by the geometry and spectral

distributions of three elements: [i] the light source, [ii] the reflectivity of the sample, and

[iii] the visual sensitivity of observer.

To measure characteristics of bananas to classification factors following three colorspaces model of image processing are important.

2.3.1 RGB Color Model

The RGB (Red, Green, Blue) color model is an especially important one in digital imageprocessing because it is used by most digital imaging devices (e.g., monitors and colorcameras). In the RGB model, a color is expressed in terms that define the amounts ofRed, Green and Blue light it contains.

Defines the transformation from floating point nonlinear R`G`B` values to sRGB: Thenonlinear R`G`B` values are transformed to linear sRGB values by

If R`, G`, B` 0.04045

sR =`.,sG = `.,sB= `.

else if R`, G`, B`>0.04045

sR = -( .. )2.4,

sG = -( .. )2.4,

sB = -( .. )2.4

2.3.2 CIELAB color spaceCIE specified color space characterized as being less illumination-dependent and thecommonly usedL*a*b* or CIELAB [7].L* is the luminance or lightness component thatgoes from 0 (black) to 100 (white), and a* (from green to red) and b* (from blue toyellow) are the two chromatic components, varying from 120 to +120.


19/58

120280723002

7

Figure 2:CIELAB colorspace [14][15]

The definition of L*a*b* is based on the intermediate system CIE XYZ which simulatesthe human perception. The knowledge of these effects, such as the variations of L*, a*,and b* for a particular shape of the sample could be useful for developing image

processing correction algorithms which can permit a better correlation among product

quality. [7]Defines the transformation from sRGB values to CIEXYZ:

= 0.4124 0.3576 0.18050.2126 0.7152 0.07220.0193 0.1192 0.9505

sRsGsB

Defines the transformation from CIEXYZ to CIELAB :

L*= 116 f() 16

a*= 500 [f(

) - f(

)]

b*= 200 [f() - f()]

Where

f(q) = qif q > 0.0088567.787q+

otherwise

Xn, Yn, and Zn correspond to the XYZ values of a reference whitechart (q{X/Xn, Y/Yn, Z/Zn}).

The total color difference between two color in L*, A* and b*coordinates may be evaluated as

E*ab= [(L*)2+ (a*)2+ (b*)2]1/2

2.3.3 HSV color space:

The HSV (Hue, Saturation, and Value) color model describes a color in terms of how it isperceived by the human eye. This is useful when processing images to compare twocolors, or for changing a color from one to another. The HSV model is also a more useful


20/58

120280723002

model for evaluating or measuring an object's color characteristics, such as the"yellowness" of a banana.HSV separates color into three components varying from 0 to 1(when it is calculated using the function rgb2hsv available in Matlab);

The hue(H) of a color refers to which pure color it resembles. All tints, tones and

shades of red have the same hue. Hues are described by a number that specifiesthe position of the corresponding pure color on the color wheel, as a fraction

between 0 and 1. Value 0 refers to red; 1/6 is yellow; 1/3 is green; and so fortharound the color wheel.

The saturation (S) of a color describes how white the color is. A pure red is fullysaturated, with a saturation of 1; tints of red have saturations less than 1; and whitehas a saturation of 0.

The value(V) of a color, also called its lightness, describes how dark the color is.

A value of 0 is black, with increasing lightness moving away from black.

Figure 3:HSV color space[16][17]

This diagram, called the single-hexcone model of color space, can help you visualize themeaning of the H, S, and V parameters.

The outer edge of the top of the cone is the color wheel, with all the pure colors.The H parameter describes the angle around the wheel.

The S (saturation) is zero for any color on the axis of the cone; the center of the

top circle is white. An increase in the value of S corresponds to a movement awayfrom the axis.

The V (value or lightness) is zero for black. An increase in the value of V

corresponds to a movement away from black and toward the top of the cone.

The Ostwald diagram corresponds to a slice of this cone. For example, the trianglebetween red, white, and black is the Ostwald diagram for the varieties of red.


21/58

120280723002

9

Defines the transformation from sRGB to HSV :

V = max(R, G, B)

S =(,,)

H =

1+ (,,) for V = R2+

(,,) for V = G3+

(,,) for V = B

2.4 Segmentation Methods

Image segmentation is the process of dividing image into multiple parts. There are manydifferent ways to perform image segmentation, like : Thresholding methods such asOtsus method, Color-based Segmentation such as K-means clustering, Transformmethods such as watershed segmentation, Texture methods such as texture filters, etc.

Thresholding is most commonly used techniques, determining a threshold value as acriterion to select required region of interest. Selection of threshold value manuallyrequires trial and error method and is time consuming and chances of bias is also possible.For selection of segmentation method we consider first criteria is color, means we have tosegment color image thats why we cannot use thresholoding. Second criteria is surface offruits, or bananas. Surface of bananas are curved thats why it will possible change ofintensity of color. From those two criteria use of color based segmentation method usingk-means clustering is best suited, also use of CIELAB is important in fruits. CIELABcolorspace is less illuminant dependent so that it is widely used in fruits quality analysisor classification.

In this paper we were used color-based segmentation methods. This method segmentcolors in an automated fashion using the L*a*b* color space and K-means clustering. K-

means clustering takes number of clusters as input and based on that it create distancematrix. It find cluster in such away so that objects within each cluster are as close to eachother as possible, and as far from objects in other clusters as possible.

2.5 Study of Papers

2.5.1 Title: Predicting Ripening Stages of Bananas (Musa cavendish) by Computer

Vision

Authors: F. Mendoza, P. Dejmek, J.M. Aguilera

Summary: They have implemented system to predict the ripening stages of bananas. Two

simple color features from each image (mean value and variance of the intensity

histogram of image) were extracted and analyzed using the RGB, HSV and CIELABcolor spaces with classification purposes. Results show that the three evaluated sets were

able to correctly predict with more than 94% the ripening stages of bananas as

professional visual perception [8].

Issues: As discussed in problem statement this system classify banana in to different

stage but if we apply this algorithm to group of bananas with different stage it were miss

classified the stage of average of whole image. If we apply this algorithm to example


22/58

120280723002

10

given in section 4.3 it classify it to stage 4 or stage 5 because of average of green and

yellow.

2.5.2 Title: Calibrated color measurements of agricultural foods using image analysis

Authors: Fernando Mendozaa,Petr Dejmek, Jose M. Aguilera

Summary: Implemented system to quantify standard color of fruit and vegetables in

sRGB, HSV and L*a*b* color spaces. The results show that sRGB standard (linearsignals) was efficient to define the mapping betweenR`G`B` (no-linear signals) from the

digital camera and a device-independent system such as CIE XYZ. L*a*b* system is

suggested as the best color space for quantification in foods with curved surfaces [7].

2.5.3 Title: Hybrid Segmentation of Peel Abnormalities in Banana Fruit

Authors: D.Surya Prabha and J.Satheesh Kumar

Summary: Automation of banana fruit analysis based on morphological features likes:Mean Square Error (MSE) and Similarity Measure (SSIM) will help banana industries for

better quality analysis. They have developed a new method for better segmentation andcategorization of banana fruits. Result shows better accuracy of proposed algorithm

compared to other segmentation methods like: Thresoloding, Clustering, color imagesegmentation. Hybrid segmentation method improves results by combing edge based andregion based segmentation [4].

2.5.4 Title: Identification and Classification of Bulk Fruits Images using Artificial NeuralNetworksAuthors: Dayanand Savakar

Summary: The study reveals that the Classification of Chikoo is about 94% and Mango

and Orange is 92% using color and texture feature sets. Classification using textureanalysis is better than classification using color analysis. Best results are obtained byusing the combination of both color and texture features [5].

2.5.5 Title: The Development of a Quality Scale to Measure the Impact of Quality on

Supermarket Fruit DemandAuthors: Catherine A. Durham, Marc V. McFetridge, and Aaron J. JohnsonSummary: This research examines how fluctuations in quality affect consumer

expenditures for fresh fruit at the retail level. A four-point scale was created and used toquantify four different quality characteristics: bruising, markings, brilliance, and maturity[6].

2.5.6 Title: Study of advanced maturity stages of bananaAuthors: Tapre A.R. and Jain R.K.Summary: Banana of three advanced stages of maturity i.e. stage 5, 6 and 7 wereanalyzed for their physical, chemical and mechanical properties. As the ripening

progressed, various physical changes observed in fruit such as increased in pulp to peel

ratio, decreased in intensity of greenness of peel from stage 5 to stage 7. As the ripeningproceed, pulp to peel ratio was increased from 2.0 in stage 5 to 2.7 in stage 7 when thefruits become fully ripened. It has been suggested that pulp to peel ratio can be consideredas a coefficient of ripeness [3].


23/58

120280723002

11

Chapter 3

Existing Systems


24/58

120280723002

12

3.1Classification of bananas during ripening by computer vision by Fernando Mendozaand Jos M. Aguilera [8]

A computer vision system was implemented to identify the ripening stages of bananasbased on color, development of brown spots, and image texture information. Simplefeatures of appearance like: L*a*b* values, brown area percentage, number of brownspots per cm2, extracted from images of bananas were used for classification purposes.Preliminary tests were performed to calibrate the performance of the selected parameters(i.e., a*band and threshold of 130) in the identification and quantification of brown spotsfrom images. Selection of features with the method of Sequential Forward Selection(SFS), SelectingL*, a*, b*, %BSA and contrast permitted the correct classification of the49 samples in different ripening stages with an accuracy of 98%.

Discriminant functions used for classification of bananas

Function 1 = -a* - 0.65 b* + 0.09L* + 0.08 %BSA + 0.15 ContrastFunction 2 = -0.07 a* - 1.35 b* + 0.82L* + 0.72 %BSA + 0.48 Contrast

Figure 4: Mapping of function to stage of bananas

3.2 Identification and Classification of Bulk Fruits Images using Artificial NeuralNetworks by Dayanand Savakar [5]

This paper presents an identification and classification of different types of bulk fruitimages using artificial neural networks. The color and texture features are extractedconsidering the whole image for feature extraction. The extracted features are stored inthe form of knowledge base. When a new image is encountered features are extractedfrom fruit image sample. The extracted features are used to identify and classify using

Neural Network.

Algorithm:

Algorithm 1: Identification and classification of fruit image samples

Input: Original 24-bit color image

Output: Classified fruit image of different types


25/58

120280723002

13

Start

Step1: Read the fruit images.Step2: Extract color and texture features.Step3: Use these features to identify and classify the fruit image samples

Stop

Algorithm 1 is takes 24-bit color image as input and extract color and texture features offruits based on that classifies fruits.

Algorithm 2: Color feature extraction

Input: Original 24-bit color image.

Output: 18 color features.

Start

Step 1: Separate the RGB components from the original 24-bit input color image.Step 2: Obtain the HSI components from RGB components.Step 3: Find the mean, variance, and range for each RGB and HSI components.

Stop.

Algorithm 2 is takes 24-bit color image of fruits as input, and extracts RGB and HIS

components and mean, variance, and range or each RGB and HIS component.

In this particularly topic large amount of research happened but normally all uses

common process but the selection factors are different, mostly L*a*b* and HSV color

space are used.


26/58

120280723002

14

Chapter 4

Proposed Methodology


27/58

120280723002

15

Proposed system is discussed in two different parts: proposed algorithm andmethodology. Proposed algorithm contains basic steps of proposed system where ismethodology containing detail of implementation with some screenshot.

4.1 Proposed Algorithm

Following is a suggested algorithm, which measures the different parameters andcompares them with standard values of parameters for classification of banana.

Algorithm:

Algorithm: Classification of bananas

Input: RGB color image of bananas

Output: Stage of bananas image

Start

Step 1: Read the fruit images.

Step 2: Apply segmentation methods to remove background from bananas image.

Step 3: Divides segmented image into 8 equal parts.

Step 4: Apply below steps A to D on each part of segmented image.

A.

Obtain the L*a*b* components for part of the input RGB color image.B. Obtain the HSV components for part of the input RGB color image.C.

Find the mean value of each components of L*a*b* and HSV.D. Find %Green, %Yellow, and %Brown (%BSA) colors in each part of input color

image based on above components.

Step 5: Classify a bananas image part in to specific stage of bananas using %Green,

%Yellow, and %Brown components values.

Step 6: Find maximum stage from 8 parts classified by previous step.

Stop

Above algorithm takes an image with proper size and resolution as input. Bananas aresegmented from background of image for good accuracy. After that divides image into 8equal parts and then measures parameter for every parts and compare with scale. Finallyit classifies bananas based on results of all parts into different categories. Graphicalrepresentation of algorithm and methodology are discussed in next section.

4.2 MethodologyAn algorithm discussed above is represented graphically in below figure [4]. Graphicalrepresentation also contain five step as per given in above algorithm. Step 1 and step 2given in above algorithm are represented in left box. Step 3 is represented in bottom box.Step 4 and step 5 are represented in right side box. Flow of data is given with directionalarrow from one step to another step.


28/58

120280723002

Figure 5: Process of classification of banana

4.2.1 Image acquisition

A digital color image is given as input to the algorithm. In the implementation weprovided three ways to gives input to the algorithm [Figure 6].

Figure 6: Pop up message allow you to select way of input

First way is to select color image from demo, it is allow you to select image from stage1,stage5, stage6 and by default is stage7 [Figure 7].

Figure 7: Pop up message allow you to select demo image

Second way is to select image from stored image from PC on which given program is

running. It will pop up a brows window to select an image [Figure 8].

Input

RGB color

Image

Image1 Image2 Image3 Image4 Image5 Image6 Image7 Image8

Calculate mean values

of L*, a*, b*, H, S, and V

and BSA

Classification:

Classify Bananas in to

different 7 stages

Divide ima e in to 8 e ual arts

Image

Segmentation

Input

RGB color

Image


29/58

120280723002

17

Figure 8: Pop up window allow you to select input image

Third way is allow you to capture image from camera connected to PC. An imagecaptured under following criteria will improve result of algorithm. A camera connected toPC was located vertically over the background at a distance of 40cm. A black cover wasused over the sample illuminators and the Color Digital Camera (CDC) to avoid theexternal light and reflections. Samples were illuminated using two parallel lamps weresituated 45 cm above the sample and at an angle of 45to the sample. The angle betweenthe camera lens and the lighting source axis was approximately 45. An image has aresolution of 1024 x 768 pixels and storage in JPEG format. The CDC was connected tothe USB port of a PC to acquire the images directly from the computer.

Image given to algorithm was displayed on figure window as shown in figure 8.

Figure 9: Image given to algorithm for classification


30/58

120280723002

18

4.2.2 Image Segmentation

The techniques that are used to find the region of interest (ROI) are usually referred to assegmentation techniques. We were extracting ROI of yellow, green and brown part fromoriginal image. ROI was extracted from the color image using combination of a* and b*components of CIELAB colorspace. Mean value of a* and b* for each color yellow,green and brown was calculated from some testing image. We were performingsegmentation of those colors using mean value of respective colors. We were convertinginput RGB image into L*a*b* image and then calculate distance matrix for each pixel ofimage using identical mean value of a* and b*. Based on distance from those colors wewere extracting ROI.Mean value of a* and b* used for segmentation are given in belowtable.

Color Mean value of a* Mean value of b*

Green 93.2115 178.3605Yellow 126.9776 172.5000Brown 141.5233 153.6703

White(Background) 128.00 128.00Table 3: Mean value Segmentation Table

Implementation allows you to select background, by default white color is taken asbackground color.

Figure 10:Pop up allow user to select background color

If you select white button then it takes mean value of a* and b* given in table for

segmentation of background otherwise it popup one window which allows you to selectregion of background. Based on Region of background selected by user system calculate

the mean value for selected background color and then use those values for segmentation.


31/58

120280723002

19

Figure 11:Figure Window allow user to select region of background

As shown in above image region selected on left-top side as background color region.System calculate means value of a* and b* for that selected region and take those valuesin segmentation of background.

After segmentation of green, yellow, brown, and background part and save as differentimages. Result of segmentation shown in below figure.

Figure 12:Figure window of segmented image

4.2.3 Feature Extraction

For classification we are using following seven features: L*a*b* and HSV (Hue,Saturation and values) and %BSA. L*a*b* components extracted from RGB componentsusing the function applycform of Matlab. Mean value of a* and b* for each part of RGBcolor image are shown in above figure. HSV components are extracted from RGBcomponents using the function rgb2hsv of Matlab. The mean for all those 6 componentsare calculated and stored suitably for later usage. The brown spots on the peel of bananaswere segmented from input RGB images using the combination of a* and b* color bands


32/58

120280723002

20

of the CIELAB color space, since the combination of these two color bands bestrepresented the appearance and development of brown spots during ripening.

Once segmentation of each color performed we were divide each color image into eightequal parts and calculate %Green, %Yellow, and %Brown colors for each part of eachcolor partition or image using a* and b* component which calculated previously in samestep. In this step we create input for classifier, and classifier take matrix given below asinput. Feature extraction improve performance of classifier by providing information in

proper form before they needs. Below table show the matrix created at feature extractionstage and provided as input to classifier based on segmentation given in figure 12.

%Green %Yellow %Brown

0 0 0

94.70 5.20 0.0991.65 4.11 4.23

87.55 8.88 3.563.94 94.06 1.98

6.77 83.17 10.053.25 94.41 2.33

1.03 94.42 4.53

Table 4: Matrix as input for classifier

Above table contain matrix which given as input to classifier. This function identifies and

quantifies all features in the image and sends the data to a classification system.

4.2.4 Classification

Classification is the process of reducing images to usable information. This meant topredict the ripening stages of bananas previously classified by expert visual inspection

using the smaller number of best features extracted from the images. [8]. Classifier given

in below figure 13 takes matrix developed in previous step and represented as table 4, asinput and output is stage of bananas or part of bananas image. According to expert we

analyze bananas bunch of different stages and developed table 5 which contain %Green,%Yellow, and %Brown for different stages. Classifier is containing set of rules of typeIf..Then..Else.

Stage Green (G) Yellow (Y) Brown (B)

1 G 85% Y10%

2 85% > G 75% 10% < Y 15%

3 75% > G 50% 15% < Y 40%

4 20% G < 50% 40% < Y 50% B


33/58

120280723002

Representation of Classifier as flowchart is shown in figure 13. The root node is inputfrom previous step feature extraction, which is in form of matrix as represented in table 4.Each row of matrix contains %Green, %Yellow, and %Brown for each part of image.Classifier is work based on table 5, which contain %Green, %Yellow, and %Brown foreach stage of bananas. Classifier map each row of table 4 to table 5 and this will be donethrough set of rules of decision tree as given in figure 13.

G >Y

Input

Matrix

D1> 75

D1> 60

D2


34/58

120280723002

22

4.3 Discriminate Power of Selected Features

The comparison of relationships between the selected features in each set revealed thatthe average values ofL* and a* color bands and variance of a* color band, in all the sets,

presented the highest discriminating power in the predicting ripening stages. In fact,CIELAB is consider a perceptually uniform color space, and therefore more suitable for

direct comparison with sensory data [8]. To test the power of a* and b* component wewere apply it to the multi stage bananas bunch, result shown in figure 14.

Multi stage bananas two bunch, one of them is green and second is yellow. We wereapplied segmentation of green, yellow and brown component from below image given infigure 14. Segmented image of respective color are shown in figure 14 to figure 17.

Figure 14: Two Bunch of bananas at different stages


35/58

120280723002

23

Figure 15: Segmented image of Green part

Figure 16: Segmented image of Yellow part


36/58

120280723002

24

Figure 17: Segmented image of Brown part

It is of interest to point out that in the three evaluated sets, the most difficult stages todiscriminate were between stages 4 (more yellow than green) and 5 (green tip and yellow)due to the high variability of the color data at these stages. It is important to mention that

in the first stages the detection of spots in some images were due to defects on the surfaceof bananas. We observed that the appearance of brown spots was evident from stage 4onwards.


37/58

120280723002

25

Chapter 5

Results and Discussion


38/58

120280723002

26

5.1. Classification Performance

Bananas set are first classified by expert and based on that we create table which contain%green, %yellow, and %brown color for different stage and we compare it with inputcolor image of bananas.

Results show that using three features: L*,a*,b* we can classifies more than 92% bananas

hand. Results vary for different stages. Conclusion of performance is given in table.

StageNo. of

samplePerformance

1 20 94%

2 15 92%

3 10 91%

4 15 90%

5 20 91%

6 30 95%

7 25 94%

Table6: Performance Index Table

If analyse table 6 then we see that performance for middle stage are little bit decreasecompare to lower and higher stages and this is happen because variability of color fromgreen to yellow. The most difficult stages to discriminate were between stages 4 (moreyellow than green) and 5 (green tip and yellow) due to the high variability of the colordata at these stages.

5.2.

Comments Resolution

In Dissertation phase 1, experts gave some remarks on the Proposed Scheme. Here, I havetried to resolve those comments. Explanation to the comments is as below:

5.2.1.Justification of why 8 parts taken.

Justification: After study of more than 100 images of bananas hands and analyze that if

we taken hands of 12 fingers of bananas then its front part contain mostly 8 bananas and

4 bananas are covered by those 8 bananas. For improve results we were checked by

dividing each image into different parts, like divide image into 5, 6, 7, 8, 9, 10, 11, and

12parts respectively. After analysis we were conclude when we divides image of bananas

hands into 8 parts then each parts contain one finger of bananas. When we were going

from less number of part it were misclassified similarly as happened in reviewed

algorithm, also going to more number of division were divides hands of bananas in such

way so that single finger of hand of bananas were divided into two or more parts so that it

were degrade performance. Division of image of bananas hands into 8 equal parts is best

suited for algorithm.


39/58

120280723002

27

Conclusion and Future work

Digital image processing can improve the process of classification of bananas. Allreviewed and currently available algorithms can efficiently measure the quality of single

banana but when we apply on bunch of bananas, performances of all algorithms aredegrading. Proposed algorithm can overcome this limitation by division of image into

parts. If single banana from bunch of bananas is at different stage from other thanproposed algorithm can efficiently identify the stage of that banana. Selected features areable to predict change of color on curved surface of bananas. In this research some factor

play major role and they are like: image acquisition system and resolution of image whichis given as input to algorithm, selection of segmentation method, selection of classifier.

This system will be very useful for bananas classification industry. This system overcomethe limitation of alternatives methods and reviewed algorithm. In development of systemwe take industry requirement in mind, so that system can play major role in industry of

bananas classification. Sometimes objectives and industry you are targeting may help youto select parameter. Adding more parameter is not good for all schemes means selectionof right number of parameter is important task. Some limitations of system and

characteristics of object you are analyze play major role and performance of system alsodepends on it.

Thus, in the future, we can improve performance of proposed scheme by using differentsegmentation methods and different classifier. Segmentation of bananas is first importantthings for bunch of bananas so that we can improve performance by changing setting ofimage acquisition system and resolution of image. Every steps of system is depends on

previous step means every steps of system provide input to its successor steps. We canchange at any higher stage will affects on performance. We can make change in system

by changing image acquisition system, selecting different segmentation methods, usingdifferent classifier.


40/58

120280723002

28

Publication

Paper TitleNational/

InternationalConference/journal/institute Status

Quality Analysis andClassification of

Bananas

International

International Journal of Advanced

Research in Computer Science andSoftware Engineering (IJARCSSE)

ISSN: 2277 128XPg No. : 869-874

Published

Quality Analysis and

Classification of

Bananas using Digital

Image Processing

International

International Journal of ComputerScience and Engineering (IJCSE)

ISSN(Print): 2278-9960ISSN(Online): 2278-9979

Accepted


41/58

120280723002

29

References

[1]Sally SmithFairtrade Bananas: A Global Assessment Of Impact April 2010 page17-18

[2]

Bhavesh B. Prajapati Algorithmic Approach to Quality Analysis of Indian BasmatiRice Using Digital Image Processing IJETAE volume 3, Issue 3, March 2013

[3]Tapre A.R. and Jain R.K Study of Advanced Maturity Stages of Banana IJAERSJune 2012

[4]D.Surya Prabha and J.Satheesh Kumar Hybrid Segmentation of Peel Abnormalitiesin Banana Fruit ICRTCT and IJCA2013

[5]Dayanand Savakar Identification and Classification of Bulk Fruits Images usingArtificial Neural Networks IJEIT Volume 1, Issue 3, March 2012

[6]Catherine A. Durham, Marc V. McFetridge, and Aaron J. Johnson The Developmentof a Quality Scale to Measure the Impact of Quality on Supermarket Fruit Demand

2005

[7]Fernando Mendozaa,Petr Dejmek, Jose M. Aguilera Calibrated color measurementsof agricultural foods using image analysis 2006

[8]

Fernando Mendoza and Jos M. Aguilera Classification Of Bananas During

Ripening By Computer Vision February 2003

[9]Mamta Juneja, Parvinder Singh Sandhu Image Segmentation based Quality Analysisof Agricultural Products using Emboss Filter and Hough Transform in SpatialDomain, 2009

[10] Anup Vibhute and S K Bodhe Applications of Image Processing in AgricultureIJCA Volume 52 No.2, August 2012

[11]

Shiv Ram Dubey, Pushkar Dixit, Nishant Singh, and Jay Prakash Infected FruitPart Detection using K-Means Clustering Segmentation Technique IJAI IM, Vol. 2,

N 2 Feb 2013

[12]

Ms. Jyoti Atwal, Mr. Satyajit Sen Purkayastha Analysis of nutrient contents andquality feature extraction of clustered seeds using Digital Image Processing Vol. 2Issue 2, Feb.2012

[13]

Reza Fellegari 1, Hosein Navid Determining the orange volume using imageprocessing ,ICFEB, IACSIT, IPCBEE vol.9 2011

[14] John A. Molino, Jason F. Kennedy Daytime Color Appearance of Retroreflective

Traffic Control Sign Materials Citing Sources:[http://www.fhwa.dot.gov/publications/research/safety/13018/002.cfm]: Chapter. 2:[Oct 3, 2013]

[15] A Practical Guide and Tutorial to Digital Color Management for PhotographersSection I: ColorModels[http://www.booksmartstudio.com/color_tutorial/colortheory4.html] [Oct 13, 2013]


42/58

120280723002

30

[16] Introduction to color theory Section: 4.3 The hue-saturation-value(HSV) colormodel [http://infohost.nmt.edu/tcc/help/pubs/colortheory/web/hsv.html] [Oct 29,2013]

[17] Converting Color Data Between Color Spaces Section : HSV color space[http://www.mathworks.in/help/images/converting-color-data-between-color-

spaces.html] [Nov 12, 2013]


43/58

Volume 4, Issue 1, January 2014 ISSN: 2277 128X

International Journal of Advanced Research in

Computer Science and Software EngineeringResearch Paper

Available online at:www.ijarcsse.com

Quality Analysis and Classification of BananasSanjay Chaudhary

*, Bhavesh Prajapati

Department of CSE and IT

L.D.College of Engineering, India

Abstracta qual ity i nspection of banana can be done by mainl y two ways: either instrumental tools or human

inspectors. An interesting alternative is image processing can overcome limi tations of these two techn iques. Digital

Image processing can classif y the banana f ru it wi th speed and accuracy. Good algor ithms available in image

processing which can classif y banana based on their color and textur e characteri stics, but limi tation is that they can

work well on single banana. Bananas is sold in bunch of dozen and thats why it i s important to analyze quali ty in

bunch. Thi s paper reviews the techni que of digital image processing to classif y the banana in gr oup or bunch.

KeywordsInstrumental, inspectors, texture, digital image, dozen

I. INTRODUCTIONThis document is a template. An electronic copy can be downloaded from the Journal website. For questions on

paper guidelines, please contact the journal publications committee as indicated on the journal website. Information

about final paper submission is available from the conference website.

A. IntroductionBananas are the forth most important staple crops in the world and India is the leading country in the production of

banana [1]. With increased expectations for food products of high quality and safety standards, the need for accurate, fast

and objective quality determination of these characteristics in food products continues to grow. Classification of bananas

is important task for banana industry. Classification of bananas can be done by basically three methods: (1) Human

visual inspection; (2) Instrumental techniques; (3) Computerized image analysis techniques. A human inspection process

may be affected by external factors like: tiredness, bias, revenge or human psychological limitations where instrumental

techniques give accurate measurements of colors but requirement are that the surface color must be quite uniform andremoval of peel [8]. Image processing systems is good alternative for an automated, non-destructive and cost-effectivetechnique to accomplish these requirements and offer an objective measure for color and other physical factors.

B. Problem StatementToday quality inspection of bananas in industry is mostly done by human inspectors, due to limitations of alternative

methods. Mostly image processing is used for classification of other fruits and grain but bananas characteristics are

different from other fruits in such a way likes: bananas always sold in group or bunch of dozen where are other fruits sold

singly; bananas are important at all level, bananas have maturity stage at stage 6 and 7 but people can purchase bananas

at different stage for backing use. For effective classification we must have to concern equally on bananas characteristics

and consumer characteristics.

Presents Algorithm implemented on single banana thats why if we apply on bunch of bananas then we must have to

change image acquisition system like: we have to set camera to the different angle, distance from sample, etc; and if any

bunch contains some degraded bananas and remain are good then classifier will classify as a good quality due to averageof all bananas but consumer will never goes to purchase such bunch of bananas as good quality. So that it is important to

mind characteristics of consumer and analyze single bananas individually in the bunch without separating physically.

C. Aim of Di ssertationThe objectives of this study were: (i) To select important features which can be related with the characteristics of

bananas and characteristics of consumer, (iii) To develop a statistical model using selected features to identify the stagesof bananas from samples previously classified by expert, and (iv) To implement an image processing system for

classification of bananas using the sRGB, L*a*b* and HSVcolor spaces.

II. LITERATURE REVIEWFor effective classification of banana it is important to have good information about bananas properties, consumer

properties, image processing ability or methods, image processing limitations, etc.

A.

Bananas Propert iesCommercial standard color charts classifies bananas in following 7 different stages: Stage 1=all green, 2= green with

trace of yellow, 3= more green than yellow, 4= more yellow than green, 5= yellow with trace of green, 6= full yellow, 7=
http://www.ijarcsse.com/http://www.ijarcsse.com/http://www.ijarcsse.com/http://www.ijarcsse.com/


44/58

Chaudhary et al., I nternational Journal of Advanced Research in Computer Science and Software Engineering 4(1),

January - 2014, pp. 869-874

Fig. 1 Seven stage of banana [3]

Bananas physical, chemical and mechanical properties changes will change the color of bananas so that if we can

successfully measure the stage of bananas then we can get approximately right information of physical, mechanical

and chemical properties. Skin color changes from green to yellow, firmness is decreased, fruit gets softened andstarch is converted into sugar. A mainly color change in banana during ripening is based on the peel color rather

than the pulp color and hence color of banana peel has been used in the assessment of the stages of ripeness of

banana [3]. With experiment they suggested following table to get the right information about banana properties.

TABLE ICHANGES IN PHYSICAL CHARACTERISTICS IN BANANA FRUITS AT DIFFERENT STAGES OF MATURITY

Parameters Stage of Ripening

5 6 7

Pulp/Peel Ratio 2.0 2.3 2.7

Peel Color Yellow with Trace of

Green

All Yellow Yellow with brown

spots

Pulp Color White WhiteCreamy Yellowish Creamy

As the ripening proceed, pulp to peel ratio was increased from 2.0 in stage 5 to 2.7 in stage 7 when the fruits becomefully ripened. (Table1). This could be due to the osmotic transfer of moisture from the peel to the pulp as sugar content of

pulp increased. It has been suggested that pulp to peel ratio can be considered as a coefficient of ripeness. The intensity

of greenness of the peel also decreased from stage 5 to stage 7.

B. Consumer Characteri sti csTo examine consumer purchasing behavior five characteristics are important: damage (dents and breaks in skin),

markings (includes russet, waxy build-up, etc), brilliance (degree of shine), maturity/color, and one less used is bruising.

The peel color of bananas is considered as the first quality parameter evaluated by consumers. Final classification after

calculation of measure: All the quality measure values were summed together to create a total quality value. [6].

Development of spots expressed as %BSA (brown spots as a percentage of the total area) and NBS/cm2 (number of

brown spots per cm2 of surface), changes in percentage and number of brown during ripening, etc are measures which

are used in classification process [8].According to consumer characteristics following scale table is created [6].

TABLE IISCALE RELATED TO CONSUMER PERSPECTIVE

Scale

Fruit Appearance(For

bruising, markings, and

damage)

Standard(brilliance)

4 Very shiny looking Less than 10 % of the fruit on display has the negative qualitycharacteristic present

3 Shiny looking 10 to 30 % of the fruit on display has the negative quality

characteristic present

2 Glossy looking 30 to 50 % of the fruit on display has the negative qualitycharacteristic present

1 Dull looking Greater than 50 % of the fruit on display has the negative

lit h t i ti


45/58


January - 2014, pp. 869-874

C. Color SpacesImage processing is able to measure characteristics of bananas and map to the consumer characteristics. In this section

we will discuss on image processing color spaces which will be useful in classification process. To measure

characteristics of bananas to classification factors following three color spaces model of image processing are important.

1) RGB Color Model: The RGB (Red, Green, Blue) color model is an especially important one in digital image

processing because it is used by most digital imaging devices (e.g., monitors and color cameras). In the RGB model,

a color is expressed in terms that define the amounts of Red, Green and Blue light it contains [7].

Defines the transformation from floating point nonlinearR`G`B`values to sRGB:The nonlinear R`G`B` values are transformed to linear sRGB values by

If R`, G`, B` 0.04045

sR =R`

12.92, sG =

G`

12.92, sB =

B`

12.92

else if R`, G`, B`>0.04045

sR = -(R`+0.055

1.055)

2.4,

sG = -(G`+0.055

1.055)

2.4,

sB = -(B`+0.055

1.055)

2.4

2) CIELAB and CIEXYZ color space: Color plays a major role in the assessment of external quality in food industries

and food engineering research (Segnini et al., 1999; Abdullah et al., 2001). Color is basically specified by the

geometry and spectral distributions of three elements: [i] the light source, [ii] the reflectivity of the sample, and [iii]

the visual sensitivity of observer. CIE specified color space characterized as being less illumination-dependent and

the commonly used L*a*b* or CIELAB (Robertson, 1976) [7].

L* is the luminance or lightness component that goes from 0 (black) to 100 (white), and a* (from green to red)

and b* (from blue to yellow) are the two chromatic components, varying from 120 to +120. The definition ofL*a*b* is based on the intermediate system CIE XYZ which simulates the human perception. The knowledge of

these effects, such as the variations of L*, a*, and b* for a particular shape of the sample could be useful for

developing image processing correction algorithms which can permit a better correlation among product quality [7].

Defines the transformation from sRGB values to CIEXYZ:

= 0.4124 0.3576 0.1805

0.2126 0.7152 0.0722

0.0193 0.1192 0.9505

sRsGsB

Defines the transformation from CIEXYZ to CIELAB :

L*= 116 f(

)16a

*= 500 [f(

) - f( )]b

*= 200 [f(

) - f(

)]

Where

f(q) = q13if q > 0.0088567.787q +

16

16 otherwise

Xn, Yn, and Zn correspond to the XYZ values of a reference white chart (q{X/Xn, Y/Yn, Z/Zn}).

The total color difference between two color in L*, a* and b* coordinates may be evaluated as

E*

ab= [(L*)2+ (a*)

2+ (b*)

2]

1/2

3)

HSV color space: The HSV (Hue, Saturation, and Value) color model describes a color in terms of how it isperceived by the human eye. This is useful when processing images to compare two colors, or for changing a color

from one to another. The HSV model is also a more useful model for evaluating or measuring an object's color


46/58


January - 2014, pp. 869-874HSV separates color into three components varying from 0 to 1; H (hue) refers to the dominant wavelength

perceived as different colors, such as red, yellow, green and blue, S (saturation) refers to how much such wavelength

is concentrated and it is equivalent to the concentration of a solute in a chemical solution; and V (value) represents

the total brightness, similar toL* [7]. Some author use intensity (I) instead of value.

Defines the transformation from sRGB to HSV :

V = max(R, G, B)

S =Vmin (R,G,B)

V

H =

1 +GB

Vmin (R,G,B) for V = R2 +

BRVmin (R,G,B) for V = G

3 +RG

Vmin (R,G,B) for V = B

Fernando Mendozaa,Petr Dejmek, and Jose M. Aguilera implemented system to quantify standard color of fruit

and vegetables in sRGB, HSV and L*a*b* color spaces. The results show that sRGB standard (linear signals) was

efficient to define the mapping between R`G`B` (no-linear signals) from the CCD camera and a device-independent

system such as CIEXYZ.L*a*b* system is suggested as the best color space for quantification in foods with curved

surfaces [7].

D. Di scriminant power of selected featuresThe comparison of relationships between the selected features in each set revealed that the average values of L* and

a* color bands and variance of a* color band, in all the sets, presented the highest discriminating power in the predicting

ripening stages. In fact, CIELAB is consider a perceptually uniform color space, and therefore more suitable for direct

comparison with sensory data (Segnini, 1999) [8].

It is of interest to point out that in the three evaluated sets, the most difficult stages to discriminate were betweenstages 4 (more yellow than green) and 5 (green tip and yellow) due to the high variability of the color data at these stages.

It is important to mention that in the first stages the detection of spots in some images were due to defects on the surface

of bananas. We observed that the appearance of brown spots was evident from stage 4 onwards [8].

III.

EXISTING SYSTEM

A. Fernando Mendoza and Jos M. Aguilera implemented a computer vision system to identify the ripening stagesof bananas based on color, development of brown spots, and image texture information. Nine simple features of

appearance like: L*a*b* values, brown area percentage, number of brown spots per cm2, extracted from images of

bananas were used for classification purposes. Preliminary tests were performed to calibrate the performance of the

selected parameters (i.e., a* band and threshold of 130) in the identification and quantification of brown spots from

images. Selection of features with the method of Sequential Forward Selection (SFS), Selecting L*, a*, b*, %BSA andcontrast permitted the correct classification of the 49 samples in different ripening stages with an accuracy of 98% [8].

Discriminant functions used for classification of bananas:

Function 1 = -a* - 0.65 b* + 0.09L* + 0.08 %BSA + 0.15 Contrast

Function 2 = -0.07 a* - 1.35 b* + 0.82L* + 0.72 %BSA + 0.48 Contrast

Fig. 3: Mapping of function to stage of bananas

B.

Paper presents an identification and classification of different types of bulk fruit images using artificial neuralnetworks. The color and texture features are extracted considering the whole image for feature extraction. The extractedfeatures are stored in the form of knowledge base When a new image is encountered features are extracted from fruit


47/58


January - 2014, pp. 869-874Algorithm:

Algorithm 1:Identification and classification of fruit image samples

Input: Original 24-bit color image

Output: Classified fruit image of different types

Start

Step1: Read the fruit images.

Step2: Extract color and texture features.

Step3: Use these features to identify and classify the fruit image samplesStop

Algorithm 1 is takes 24-bit color image as input and extract color and texture features of fruits based on that

classifies fruits.

Algorithm 2: Color feature extraction

Input: Original 24-bit color image.

Output:18 color features.Start

Step 1: Separate the RGB components from the original 24-bit input color image.

Step 2: Obtain the HSV components from RGB components.

Step 3: Find the mean, variance, and range for each RGB and HSV components.

Stop.Algorithm 2 is takes 24-bit color image of fruits as input, and extracts RGB and HSV components and mean,

variance, and range or each RGB and HSV component.The study reveals that the Classification process best results are obtained by using the combination of both color and

texture features [5].

IV. BANANAS CLASSIFICATION SYSTEMBananas classification process contains mainly five steps as follow.

Fig. 4 Process of classification of banana

A. Image acquisiti onBlock which contain camera, lights, and stand for banana. Images from one side of the bananas were taken and storage

in JPEG format. The camera was connected to the USB port of a PC to acquire the images directly from the computer.

Image acquisition system setup is shown in below figure.

Fig. 5 Image acquisition system [8]

A Color Digital Camera (CDC) was located vertically over the sample at a distance of 22 cm. The angle between the

camera lens and the lighting source axis was approximately45 [8]. If we use this image acquisition system for bunch of

bananas than because of camera was located vertically at a small distance over the sample and large size of bananas we

cannot get complete image of bunch of bananas So that it is important to change distance of camera from sample and

Image

Se mentati

Digital

Ima e

Image

Ac uisiti

Feature

ExtractionClassificati


48/58


January - 2014, pp. 869-874

B. Digital ImageDigital images are one of the most key medium of conveying information. Extracting the information from images and

understanding them such that the extracted information can be used for several tasks is an important characteristic.

Convert image into digital form for the further processing. If we are use digital camera then the image is in digital form

and we did not need conversion otherwise this step is required.

C. Image SegmentationThe techniques that are used to find the objects of interest are usually referred to as segmentation techniques [9].

D.Surya Prabha and J.Satheesh Kumar describes in their research [4]. They have developed a new method for bettersegmentation and categorization of banana fruits. Result shows better accuracy of proposed algorithm compared to other

segmentation methods like: Thresoloding, Clustering, color image segmentation. Hybrid segmentation method improves

results by combing edge based and region based segmentation [4].

In this research we successfully segment banana different part like brown sport, dents and breaks in skin, etc, but at

industry level it has no advantage because these are end up with unimportant results means what to do after segment, we

will do not apply any treatment on its and its takes more times compare to Thresoloding method. Thats comparing touse complex hybrid system for segment; simple Thresoloding method is more suitable for bananas classification.

Fernando Mendoza and Jos M. Aguilera used combination of threshold of 50 with an edge detection technique based on

the Laplacian-of-Gauss (LoG) operator to remove background from grayscale image. The brown spots on the peel of

bananas were segmented from binary images using the combination of a* and b* color bands of the CIELAB color space.

Images were binarized using threshold values of a*156 (for a* and b* values ranging from 0 to 255) [8].

D. Feature extraction

The RGB components are separated from the original image, and the L* a* b* and HSV components are extractedfrom RGB components. The mean, variance and range for all these 6 components are calculated and stored suitably for

later usage [4]. This function identifies and quantifies all features in the image and sends the data to a control program.

E. ClassificationClassification is the process of reducing images to usable information. This meant to predict the ripening stages of

bananas previously classified by expert visual inspection using the smaller number of best features extracted from the

images. [8]. Based on all parts factors value bananas are classified into different class or grade. Classification process can

be improved by adding more factors in the process.

V. CONCLUSIONSDigital image processing can improve the process of classification of bananas. All reviewed and currently available

algorithms can efficiently measure the quality of single banana but when we apply on bunch of bananas then either image

acquisition setup are not suite or performances of algorithm are degrading. Based on review we can say that image

acquisition is important steps over all other step. Mostly fruit classification researcher used common color space and

even features are not change more. Color and texture features combination gives best result for classification process.Adding more factors for analysis will not always increase the result thats why use of good combination of factors is

important.

REFERENCES

[1] Sally Smith Fairtrade Bananas: A Global Assessment Of Impact April 2010 page 17-18[2] Bhavesh B. Prajapati Algorithmic Approach to Quality Analysis of Indian Basmati Rice Using Digital Image

Processing IJETAE volume 3, Issue 3,March 2013

[3] Tapre A.R. and Jain R.K Study of Advanced Maturity Stages of BananaIJAERSJune 2012[4]

D.Surya Prabha and J.Satheesh Kumar Hybrid Segmentation of Peel Abnormalities in Banana Fruit ICRTCT

and IJCA 2013

[5] Dayanand Savakar Identification and Classification of Bulk Fruits Images using Artificial Neural Networks

IJEIT Volume 1, Issue 3,March 2012

[6]

Catherine A. Durham, Marc V. McFetridge, and Aaron J. Johnson The Development of a Quality Scale toMeasure the Impact of Quality on Supermarket Fruit Demand 2005

[7] Fernando Mendozaa,Petr Dejmek, Jose M. Aguilera Calibrated color measurements of agricultural foods usingimage analysis 2006

[8] Fernando Mendoza and Jos M. Aguilera Classification Of Bananas During Ripening By Computer VisionFebruary 2003

[9] Mamta Juneja, Parvinder Singh Sandhu Image Segmentation based Quality Analysis of Agricultural Productsusing Emboss Filter and Hough Transform in Spatial Domain, 2009

[10] Anup Vibhute and S K Bodhe Applications of Image Processing in AgricultureIJCA Volume 52No.2, August

2012

[11] Shiv Ram Dubey, Pushkar Dixit, Nishant Singh, and Jay Prakash Infected Fruit Part Detection using K-MeansClustering Segmentation TechniqueIJAI IM, Vol. 2, N 2,Feb 2013

[12]

Ms. Jyoti Atwal, Mr. Satyajit Sen Purkayastha Analysis of nutrient contents and quality feature extraction ofclustered seeds using Digital Image Processing Vol. 2 Issue 2,Feb.2012

[13] Reza Fellegari 1, Hosein Navid Determining the orange volume using image processing , ICFEB, IACSIT, and


49/58

Quality Analysis and Classification of Bananas Using Digital Image Processing

Sanjay Chaudhary* Bhavesh Prajapati

Department of CSE and IT Department of CSE and ITL.D.College of Engineering L.D.College of [email protected] [email protected]

Abstract a quality inspection of banana can be done by mainly two ways: either instrumental tools or human

inspectors. An interesting alternative is image processing can overcome limitations of these two techniques.

Digital Image processing can classify the banana fruit with speed and accuracy. Good algorithms available in

image processing which can classify banana based on their color and texture characteristics, but limitation is

that they can work well on single banana. Bananas is sold in bunch of dozen and thats why it is important to

analyze quality in bunch. This paper proposes the technique of digital image processing to classify the banana

in group or bunch. Features including CIELAB, and HSV colorspaces, k-means clustering a segmentation

method, and decision tree classifier have been used for classification of bananas images. Result shows better

accuracy of proposed algorithm.

Keywords Bananas,Colorspaces, Clustering, Digital image, Instrumental, Segmentation

I. INTRODUCTION

Bananas are the forth most important staple crops in the world and India is the leading country in the production

of banana [1]. With increased expectations for food products of high quality and safety standards, the need for

accurate, fast and objective quality determination of these characteristics in food products continues to grow.

Classification of bananas is important task for banana industry. Classification of bananas can be done by

basically three methods: (1) Human visual inspection; (2) Instrumental techniques; (3) Computerized image

analysis techniques. A human inspection process may be affected by external factors like: tiredness, bias,

revenge or human psychological limitations where instrumental techniques give accurate measurements of

colors but requirement are that the surface color must be quite uniform and removal of peel [8]. Image

processing systems is good alternative for an automated, non-destructive and cost-effective technique to

accomplish these requirements and offer an objective measure for color and other physical factors.

For effective classification of banana it is important to have good information about bananas properties,

consumer properties, image processing ability or methods, image processing limitations, etc. Commercial

standard color charts classifies bananas in following 7 different stages: Stage 1=all green, 2= green with trace of

yellow, 3= more green than yellow, 4= more yellow than green, 5= yellow with trace of green, 6= full yellow,

7= full yellow with brown spots [3].


50/58

Fig. 1 Seven stage of banana [3]

Bananas physical, chemical and mechanical properties changes will change the color of bananas so that if we

can successfully measure the stage of bananas then we can get approximately right information of physical,mechanical and chemical properties. Skin color changes from green to yellow, firmness is decreased, fruit gets

softened and starch is converted into sugar. A mainly color change in banana during ripening is based on the

peel color rather than the pulp color and hence color of banana peel has been used in the assessment of the

stages of ripeness of banana [3].

Today quality inspection of bananas in industry is mostly done by human inspectors, due to limitations of

alternative methods. Mostly image processing is used for classification of other fruits and grain but bananas

characteristics are different from other fruits in such a way likes: bananas always sold in group or bunch of

dozen where are other fruits sold singly; bananas are important at all level, bananas have maturity stage at stage

6 and 7 but people can purchase bananas at different stage for backing use. For effective classification we must

have to concern equally on bananas characteristics and consumer characteristics.

Presents Algorithm implemented on single banana thats why if any bunch contains some degraded bananas and

remain are good then classifier will classify as a good quality due to average of all bananas but consumer will

never goes to purchase such bunch of bananas as good quality. So that it is important to mind characteristics of

consumer and analyse single bananas indivi

Classification of Banana with using Digital Image Processing

Documents