Documente Academic
Documente Profesional
Documente Cultură
Skin defect detection of Pomegranates using Color Texture Features and DWT
Meenakshi M. Pawar
Dept. of Electronics & Telecommunication
SVERIs College of Engineering Pandharpur,
Maharashtra, India
minakshee2000@gmail.com
Abstract Various Skin disorders lower the quality of fruits
due to environmental stress such as high temperature and
solar radiation some other skin disorders are induced by
chemical treatments and pathogens. Skin defect detection is
important in the development of automatic grading and sorting
system for pomegranate, because manual sorting process is
very expensive and time consuming to automate this process
skin defect can be identified with the help of color texture
feature and discrete wavelet transform. For color texture
feature analysis, acquired image is transformed into HSI color
space, which is further used for generating SGDM matrix.
Total 12 texture features were computed for hue (H),
saturation (S) and intensity (I) images from each image
samples. Then wavelet transform is used to compute statistical
features, Total 3 features were computed for R, G & B
components of each image samples. Best features were used as
an input to Support Vector Machine (SVM) classifier and tests
were performed to identify best classification model. Features
showing optimal results were mean (99%), variance (99.80),
cluster shade (99.88%), cluster prominence (99.88%), Mean
intensity (99.81%).
Keywords- pomegranate; color Co-occurrence method; texture
features; discrete wavelet transform; support vector machine
I.
INTRODUCTION
Meghana M. Deshpande
Dept. of Electronics & Telecommunication
SVERIs College of Engineering Pandharpur,
Maharashtra, India
meghana.deshpande24@gmail.com
to recognize a fruit rely on four basic features which
characterize the object on the basis of intensity, color, shape
and texture. Blasco et al.[6] recognized skin damage in
citrus fruit using multispectral data and morphological
features and fruit sorting is done to identify the defect.
Input Image
Cropping of image
Separating RGB
Component
Generation SGDM
Matrix
Computation of DWT
Coefficient
Calculation of
Texture Matrix
Calculation of
Statistical Matrix
Kim et al. [7] used the color co-occurrence method for citrus
peel fruit classification. Pydipati et al. [8] utilized the color
co-occurrence method to extract various textural features
from the color RGB images of citrus leaves. There are two
main analysis methods for calculation of texture viz.
1) Structural Approach 2) Statistical Approach
Statistical approach, which is used here, is a quantitative
measure of arrangement of intensities in a region. Statistical
methods use second order statistics to describe the
relationships between pixels within the region by
constructing Spatial Gray-level Dependency Matrices
(SGDM). A SGDM matrix is the joint probability
occurrence of gray levels i and j for two pixels with a
defined spatial relationship in an image. Distanced and
angle are used to define the spatial relationship. If the
texture is coarse and distanced is small compared to the
size of the texture elements, the pairs of points at distance d
should have similar gray levels. In turn, if the texture is fine
and distance d is comparable to the texture size, then the
gray levels of points separated by distance d should often be
quite distinct, so that the values in the SGDM matrix should
be disperse uniformly. Thus, texture directionality can be
analyzed by examining spread measures of SGDM matrices
created at various distanced. Extraction of a numerous
texture features are possible using the SGDM matrices
generated in the above manner.
The following steps were performed to generate SGDM
matrices:
The test image is then cropped such that around 2000
cropped image.
These RGB images are converted into HSI color space
representation.
Then each pixel map is used to generate a color cooccurrence matrix, resulting in three CCM matrices, one for
each of the H, S and I pixel maps.
These matrices measure the probability that a pixel at one
particular gray level will occur at a distinct distance and
orientation from any pixel, given that pixel has a second
particular gray level. For a position operator p, we can
define a matrix Pij that counts the number of times a pixel
1 0 3 0
I ( x, y ) =
2 2 0 3
3 2 0 2
1 3 2 3
P=
1
3
0
0
0
2
1
3
with grey-level i occurs at position p from a pixel with greylevel j. For example, if we have four distinct grey-levels 0,
1, 2 and 3, then one possible SGDM matrix P (i, j, 1, 0) is
given below as shown:
The SGDMs are represented by the function P (i, j, d, )
where i represents the gray level of the location (x, y) in
the image I(x, y), and j represents the gray level of the pixel
at a distance d from location (x, y) at an orientation angle of
. The nearest neighbour mask is exemplified in Fig.3,
where the reference pixel is shown as an asterisk.
V=
Ng 1
(i Mi) P (i)
(1)
i =0
Cluster Shade:
Cs =
Ng 1
(i P + j P ) P(i, j )
x
i , j =1
(2)
Cluster Prominence:
Cp =
Ng 1
(i P + j P ) P(i, j )
x
(3)
i , j =1
Mean Intensity:
Mi =
w.x b = 0
Ng 1
iPx (i)
(4)
i =0
Px =
Ng 1 Ng 1
i P(i, j)
i =0
Py =
j =0
Ng 1Ng 1
jP(i, j )
Mk =
V.
1 N N
x(i, j)
N 2 i =1 j =1
Which implies
ci ( w.xi b) 1
1 i n
Where, xi is input sample and ci output having either 1 or 1, a constant denoting the class. The vector w is
perpendicular to the separating hyperplane and the offset
parameter b allows in increasing the margin. For the linearly
separable training data, these hyperplanes can be considered
to maximize the distance between the extreme points of
each class. The distance between the hyperplanes can be
2/ w
i =0 j =0
IV.
(5)
(1 / 2) w
c ( w.x b) 1
i
1 i n
subject to i
The factor of 1/2 is used for mathematical convenience. The
dual form of this optimization problem leads to a
classification problem which is only a function of the
support vectors, i.e., the training data that lie on the margin.
Originally, the SVM algorithm was developed for as a linear
classifier. However, it was further extended Boser et al [13]
to create a nonlinear classifier by applying the kernel trick.
Some common kernels include, Polynomial (homogeneous):
k ( x, x) = ( x, x)d ;
Polynomial
(inhomogeneous):
k ( x, x) = ( x, x + 1)d
k ( x, x) = exp( x x )
2 2
VI.
, for
NUMERICAL RESULTS
a) Variance
b) Cluster Shade
a) Varience
c) Cluster Prominance
b) Cluster Prominence
d) Mean Intensity
e) Mean
Figure 4. 3D plots for feature analysis a) Variance b) Cluster Shade c)
Cluster Prominance d) Mean Intensity e) Mean
c) Mean Intensity
Figure 5. Contour plot of SVM classifier a) Variance b) Cluster
Prominence c) Mean Intensity
Nc
*100
N
(6)
SUCCESS RATE
Feature
Mean
Variance
Cluster Shade
Cluster Prominence
Mean Intensity
Success Rate
99%
99.80%
99.88%
99.8835%
99.81%
VII. CONCLUSION
In this paper, the application of nonlinear feature extraction
for Pomegranate is presented. The features are obtained
using HSI color model and wavelet transform, SVM is used
for classification of the data. The result of skin defect
detection of Pomegranate using SVM shows that all features
are showing success rate above 99%, this gives the
confidence that this algorithm can be used for automatic
detection of quality of fruit.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]