CHAPTER (2)

LITERATURE REVIEW

In this chapter, literature review about the mango fruit and other fruit with color
detection and defect detection are presented. Their related theory and methods are also described.
The implementation of each literature review is briefly explained.

2.1. Uses of Image Processing

(Javamala K Patil and Rai Kumar. 2011) identified about five main uses of
image processing in respect to agricultural plants and their fruits. In their research, they
identified that image processing is very useful to
 Detect plant leaves, stems and fruits that are affected by diseases.
 Quantify the areas affected by disease in plant leaves, fruit and stems.
 Detect the shape of the area of the leaves, fruits and stem that has been affected.
 Determine the color of the affected areas and finally.
 Find out the size and shape of fruits.
(K.M.M Rao, 2014) also identified some uses of Image Processing as applied
and used in various applications such as Material Science, Agriculture, Remote Sensing,
Medicine, Document processing, engineering, Non-destructive Evaluation, Forensic Studies,
Textiles, Military, Printing Industry, Film industry and Graphic design.

2.2. Image Processing Methods

Rao (2014) presented two main methods of image processing. In this paper, he
presented Analog image processing as the first method. This method refers to the changes
modifications and adjustment of image through electrical means. A typical example of this
method is the image produced by the television. The television transmits signals in a form of
voltage which varies in amplitude to represent brightness through the image. The writer
continues with the second method which he identified as digital method of processing image. In
this case, supposed image will be change or converted to digital from through a device known as
scanner digitizer for further processing.
2.2. Related Works on Mangoes Color and Defect Detection

Dameshwari. S. and Ravindra. M. P, presented development of a defect identification

and maturity detection for mango fruits using image analysis. This system was used to detect the
defected mango fruit(such as scars, dark spots, etc.) and fruity maturity based on shape, size and
color features by image processing and features extraction. MATLAB had been used as the
programming tool for identification and classification of fruits using Image Processing toolbox.
In this method, the proposed algorithm was tested using 10 mango that are 8 mangoes are
defected and 2 are not defected.
Sulakshana. A. G., K. S. Deore., M. K. Waykar, P. R. Dudhane. and G. Sorate, proposed
a solution for the detection and classification of fruits diseases. An image processing based
solution is proposed and evaluates in this paper for the detection and classification of fruit
diseases.
Mrunali. D., Ashish. K. J., and Karan. J, presented an approach for detection and
classification depends on disease in fruit using image processing. This paper provided a detailed
literature review and gave an idea that what was the scope of current research for the image
categorization problems and presented a solution for detecting and classifying diseases in fruit
with procedures. For segmented image, there are many types of techniques such as clustering
based segmentation, edge based segmentation, region based segmentation, threshold based
segmentation etc. This proposed system used clustering technique for image segmentation which
is followed by extraction of some features from the segmented image and finally images are
classified into one of the classes
Bhavini. J. S. presented a survey on apple fruit disease detection and classification with
different methods in image processing. This paper surveys the methodologies utilized for apple
fruit disease detection, segmentation of infected apple fruit part and classification of disease by
using image processing. In this paper include only three common apple fruit diseases such as
scab, apple rot and apple blotch. The diseases detection of fruit is basically having three steps.
The first step is to prepare the training and testing data set of the apple fruit images. Secondly, it
is to segment the infected apple fruit part from the testing image by using segmentation
techniques. In this segmentation step is based on color component with K-mean clustering
algorithm. The third step is feature extraction . In this step, some color and texture feature are
extracted to identify the apple diseases. And this paper is discussed and compared about the
different segmentation techniques, the different color feature techniques and the different texture
feature techniques with their description, advantages and disadvantages
Bhavini. J. S, predented a survey on apple fruit disease detection and identification with
different methods in image processing. This paper surveys the methodologies utilized for apple
fruit disease detection, segmentation of infected apple fruit part and classification of diseases by
using image processing. In this paper include only three common apple fruit diseases such as
apple scab, apple rot and apple blotch. The disease detection of apple fruit is basically having
three steps. The first step is to prepare the training and testing data set of the apple fruit images.
Secondly, it is to segment the infected apple fruit part from the testing image by using
segmentation techniques. In this segmentation techniques is based on color component with K-
mean clustering algorithm. The third step is feature extraction. In this step, some color and
texture features are extracted to identify the apple diseases. And this paper is discussed and
compared about the different segmentation techniques, the different color feature techniques and
the different texture feature techniques with their description, advantages and disadvantages
[15Bha].
S.V. Phakade, introduced an automated computer recognition system for fruit defect
detection using color features. In this system used HSV (Hue, Saturatuion, and Value) color
model and RGB (Red, Green Blue) colore model for detection of brown. The method was
implemented and tested on some fruit images such as apple and mango. It made a comparatively
study between HSV and RGB color space. Input image was segmented using thresholding
technique to identify whether the portion of image was defective or non-defective. Three
morphological operations, closing, region filling and area opening, were used in order to identify
the region of interest. This system calculated amount of defect and sorted the fruit based on
amount of defect.
Tomas, U. G. J, introduced a development of an efficient algorithm for detecting and
sorting mango fruit using computer vision with image processing techniques. Many mangoes
were acquired with digital camera using natural light (between 9AM – 12PM). MATLAB had
been used as the programming tool for classification of mango using Image Processing toolbox.
The proposed system used extracted features of the mango such as perimeter, area. roudness and
percent defect for identifying the class of mango. Not only the roudness and percent defect were
used to determine the mango’s quality but also the area was used to determine the size of mango
based on threshold values.

Shivaram, D, introduced about the Infected Fruit Part Detection using K-Means
Clustering Segmentation Technique. This paper displays a novel defect segmentation of fruit
based on color component with K-means clustering unsupervised algorithm. They utilized color
image of fruit for defect segmentation. Defect segmentation is done into two stages. At to start
with, the pixels are clustered based on their color and spatial features, where the clustering
procedure is accomplished. At that point, the clustered blocks are converged to a particular
number of regions. Utilizing this two stage system, it is possible to increase the computational
efficiency avoiding feature extraction for each pixel in the image of fruit. Likewise expresses a
survey of the different segmentation techniques. The fundamental point of this methodology is to
segment colors automatically utilizing the K-means clustering techniques and L*a*b color cpace.

Meenu, D. and V. K. Banga, presented the ripeness and maturity level detection of fruits
and vegetables by using color segmentation with three different techniques like histogram
matching, clustering algorithms based image segmentation and relative value of parameter based
segmentation. In clustering algorithm, this method used for color image segmentation with K-
means , the fuzzy c means (FCM), the Gustafson Kessel improved by Babuska (GK-B), and the
Guatafon Kessel Possobilistic Fuzzy c Means (GKPFCM). When comparing above methods, the
last method can be showed the best result for fruit image. To determine the ripeness of given
fruit, color histogram is being used to base on RGB color model and threshold levels. Histogram
matching techniques provides the better result because it increases the global contrast of image
so that intensities can be better distributed on histogram. The parameter based segmentation
techniques was used to determine maturity levels of fruits and vegetables by changing the
parameter [12Mee].

(Ballard and Brown, 1982) defined image segmentation as the process of dividing or
braking an image into different parts based on certain characteristics. The parts usually conform
to something that human beings can easily separate, view and analyze as individual objects. The
digital computer as we know does not have the ability to recognize objects intelligently on their
own: this is why different researches have come out with different approaches and methods to
segment images. Images are usually segmented depending on the various characteristics and
features found in the image. These features may include color information which is used to
create histogram, information about pixels that indicate boundaries and texture information.

In the case of computers, image segmentation can be defined as the process of dividing
digital image into several parts or components (a set of pixels that can also be referred to as super
pixels). The purpose of segmenting any image is to simplify ( make the image easier or less
complicated) and change the representation of the image into something meaningful for the
purposes of easier analysis. Image segmentation is usually used to detect and discover
characteristics, objects and boundaries such as curves and lines in images. Image segmentation
can simply be defined as the process of assigning some descriptions to every pixels in an image
to identify pixels with similar and unusual characteristics. The outcome of the segmentations
process depend on a set of components that together the entire image or a set of outcomes
deduced from the image. Each of the pixels in a particular region is similar with respect to some
characteristics or computer property, including intensity to the similar characteristics. When
applied to a pile of images, typical in medical imaging, the resulting outline after image
segmentation can be used to create 3 Dimensional reconstructions with the help of a written
algorithm.

2.2. Cluster Based Approach

Clustering based techniques [10Jur] segmented images into clusters that contained pixels
with similar characteristics. A similarity criteria was defined between pixels, and then similar
pixels were grouped together to form clusters. The grouping of pixels into clusters was based on
the principle of maximizing the intra class similarity and maximizing the inter class similarity.
Clustering technique attempted to access the relationship among patterns of the set by organizing
the patterns into group such that the pattern within a cluster were more similar to each other than
patterns that belong to another cluster.

It proposed a novel method for enhancing watershed segmentation by using prior shape
and appearance knowledge. The method iteratively aligns a shape histogram with result of an
improved k-means clustering algorithm of the watershed segments. The method comprised a
training stage and a segmentation stage. In the training stage, a ‘shape histogram’ and image
intensity statistics were used to model prior shape and appearance and consists of 4 steps:
Classical Watershed Transformation, Improved K-means Clustering, shape Alignment, and
Refinement. Advantages: Automatically locates and segments the target, overcomed some
intrinsic problems with watershed transformation.

2.3. Image Thresholding

Image thresholding refers to the process of creating a binary image (bitonal) by setting a
starting point which serves as the base value of the pixel intensity of the original image.
Thresholding technique is usually performed on grayscale images, however, thresholding may be
applied to original (true color) image. The threshold value of every image can either be set
manually or automatically using a specific software or application. In this case, all pixels that fall
below that set threshold value are converted to black which represents bit value of zero whilst
any other pixels above the threshold value are changed to white representing a bit value of one.
The thresholding can also be described as the process of breaking an image apart in order to get
foreground values as well as background values (black and white).

Thresholding can be simple or complex depending on the threshold value. Thresholding

is said to be simple if there is only one threshold value set for all the pixels in the image for no
matter the difference or variations in contrast. On the other hand, complex and sophisticated
thresholding (adaptive thresholding) takes number of regions of the image and set the threshold
value accordingly. It should be noted that quality cannot be compromised in thresholding
especially when dealing with scanning images with Optical Characters Recognition (OCR)
systems. The simplest form of segmentation is by means of thresholding.

This is because you only need to define a threshold then examine all pixels in the image
by comparing them with the threshold value. All pixels found above the threshold value are
considered as foreground, whilst the pixels located below the threshold value are considered as
background. Often, the threshold represents the intensity or the color value of the image. Other
forms of thresholding permit different or variable threshold values throughout the image. In this
case, the threshold is permitted to undergo through several changers throughout the entire image.
Thresholding is said to be a primitive technique because it works for most operations that need
segmentation.

2.4. The use of Support Vector Machines (SVM)

Support vector machines are learning models associated with learning algorithm that
analyze data with the intention of classifying and analyzing objects. They are discriminative
classifiers based on decision planes that define decision boundaries.

(Youwen et al. 2008) used an algorithm based a statistic pattern recognition to identify
some disease that affect cucumber leaves. The algorithm was used to segment the leaves into
healthy and diseased regions. That is followed by extracting color, shape and texture features
from the image. The final classification is done by feeding the features into the SVM. They
concluded by stating that, Support Vector Machine produce better results than Neural networks
based on their experimentation. (Camargo and Smith 2009) also tried to detect and measure the
extent of defect in cotton plants. Images were captured from the stem, leaves and the fruits of the
cotton plant for the experiment. They segmented the images using a technique they had already
developed earlier which was described in this paper under Thresholding. They then extracted
several features from the diseased regions. Those features are then used to feed the SVM for
detection and estimation of the diseased portions of the images. (Jian and Wei 2010) presented a
paper that uses the SVM to detect cucumber leaf diseases. They used the simple thresholding
method to segment the healthy and diseased regions of the leaves. That was followed by
extracting features like color, shape and texture from the image for further processing. Those
features are fed into the SVM with Radial Basis Function (RBF) as kernel, which performs the
final classification.

2.5. Edge Detection

Edge detection attempted to resolve image segmentation by locating edges or pixels

between different regions that shows rapid change transition in intensity values. Edge detection
techniques located edges where either the 1st derivative of intensity was greater than a specified
threshold or the 2nd derivative had zero crossings. All the edges was identified and then
connected together to form the boundaries of the object that segmented. In other words, the result
was an outline of borders. There are two common edge-detection method edge detection
operators can be used such as Sobel operator, canny operator. [86Spa] proposed a method that
utilizes simple circular shifting for manipulating the whole image at a time rather than template
based. As pre-processing step, real complement of each channel was taken and the circular shift
operations were applied in all directions in order to determine the edge pixels on the resultant
image. This process was considered to be attractive in a computational perspective since the
underlying operations at edge pint effectively reduce to shift and subtraction.

Gyunsun presents an edge detection algorithm based on calculations of the difference in

values of two clusters. For every pixel, a window is partitioned into two sub-regions respectively
in four different by placing the pixel at the center. This window is partitioned into two sub-
regions respectively in 4 different directions. An appropriate function is then chosen to estimate
the difference between each pair of two adjacent regions and also to calculate edge information
in terms of edge strength and direction by maximizing the difference value. Finally, the non-
maxima suppression is adopted to derive the output edge map. Advantages: more robust under
noisy conditions. It is consistent and reliable even when image quality is significantly degraded
by noise.