83
CHAPTER 4
TEXTURE FEATURE EXTRACTION
This chapter deals with various feature extraction technique based
on spatial, transform, edge and boundary, color, shape and texture features. A
brief introduction to these texture features is given first before describing the
gray level co-occurrence matrix based feature extraction technique.
4.1
INTRODUCTION
Image analysis involves investigation of the image data for a
specific application. Normally, the raw data of a set of images is analyzed to
gain insight into what is happening with the images and how they can be used
to extract desired information. In image processing and pattern recognition,
feature extraction is an important step, which is a special form of
dimensionality reduction. When the input data is too large to be processed and
suspected to be redundant then the data is transformed into a reduced set of
feature representations. The process of transforming the input data into a set
of features is called feature extraction. Features often contain information
relative to colour, shape, texture or context.
4.2
TYPES OF FEATURE EXTRACTION
Many techniques have been used to extract features from images.
Some of the commonly used methods are as follows:
84
Spatial features
Transform features
Edge and boundary features
Colour features
Shape features
Texture features
4.2.1
Spatial Features
Spatial features of an object are characterized by its gray level,
amplitude and spatial distribution. Amplitude is one of the simplest and most
important features of the object. In X-ray images, the amplitude represents the
absorption characteristics of the body masses and enables discrimination of
bones from tissues.
4.2.1.1
Histogram features
The histogram of an image refers to intensity values of pixels. The
histogram shows the number of pixels in an image at each intensity value.
Figure 4.1 shows the histogram of an image and it shows the distribution of
pixels among those grayscale values. The 8-bit gray scale image is having 256
possible intensity values. A narrow histogram indicates the low contrast
region. Some of the common histogram features are mean, variance, energy,
skewness, median and kurtosis are discussed by Myint (2001).
85
Intensity
Figure 4.1 Histogram of an image
4.2.2
Transform Features
Generally the transformation of an image provides the frequency
domain information of the data. The transform features of an image are
extracted using zonal filtering. This is also called as feature mask, feature
mask being a slit or an aperture. The high frequency components are
commonly used for boundary and edge detection. The angular slits can be
used for orientation detection. Transform feature extraction is also important
when the input data originates in the transform coordinate.
4.2.3
Edge and Boundary Features
Asner and Heidebrecht (2002) discussed edge detection is one of the
most difficult tasks hence it is a fundamental problem in image processing.
Edges in images are areas with strong intensity contrast and a jump in
intensity from one pixel to the next can create major variation in the picture
quality. Edge detection of an image significantly reduces the amount of data
86
and filters out unimportant information, while preserving the important
properties of an image. Edges are scale-dependent and an edge may contain
other edges, but at a certain scale, an edge still has no width. If the edges in an
image are identified accurately, all the objects are located and their basic
properties such as area, perimeter and shape can be measured easily.
Therefore edges are used for boundary estimation and segmentation in the
scene.
4.2.3.1
Sobel technique
Sobel edge detection technique consists of a pair of 3 3 convolution
kernels. One kernel is simply the other rotated by 90° as shown in Figure 4.2.
These kernels are designed to respond maximally to edges running vertically
and horizontally relative to the pixel grid of the image, one kernel for each of
the two perpendicular orientations. The kernels can be applied separately to
the input image, to produce separate measurements of the gradient component
in each orientation. These can then be combined together to find the absolute
magnitude of the gradient at each point and the orientation of that gradient.
-1
0
+1
+1
+2
+1
-2
0
+2
0
0
0
-1
0
+1
-1
-2
-1
Gx
Gy
Figure 4.2 Masks used for Sobel operator
87
4.2.3.2
Robert technique
The Robert cross operator performs a simple, quick to compute, 2-D
spatial gradient measurement on an image. Pixel values at each point in the
output represent the estimated absolute magnitude of the spatial gradient of
the input image at that point. The operator consists of a pair of
2 2 convolution kernels as shown in Figure 4.5. One kernel is simply the
other rotated by 90°. This is very similar to the Sobel operator.
+1
0
0
+1
0
-1
-1
0
Gx
Gy
Figure 4.3 Masks used for Robert operator
4.2.3.3
Prewitt technique
Prewitt operator is similar to the Sobel operator and is used for
detecting vertical and horizontal edges in images.
-1
0
+1
+1
0
-1
-1
0
+1
+1
0
-1
-1
0
+1
+1
0
-1
Gx
Gy
Figure 4.4 Masks for the Prewitt gradient edge detector
88
The Prewitt operator measures two components. The vertical edge
component is calculated with kernel G x and the horizontal edge component is
calculated with kernel G y as shown in Figure 4.4. | G x | | G y | gives an
indication of the intensity of the gradient in the current pixel.
4.2.3.4
Canny technique
The Canny edge detection algorithm is known popularly as the
optimal edge detector. The Canny algorithm uses an optimal edge detector
based on a set of criteria which include finding the most edges by minimizing
the error rate, marking edges as closely as possible to the actual edges to
maximize localization, and marking edges only once when a single edge
exists for minimal response. According to Canny, the optimal filter that meets
all three criteria that can be efficiently approximated using the first derivative
of a Gaussian function. The first stage involves smoothing the image by
convolving with a Gaussian filter. This is followed by finding the gradient of
the image by feeding the smoothed image through a convolution operation
with the derivative of the Gaussian in both the vertical and horizontal
directions.
This
process
alleviates
problems
associated
with
edge
discontinuities by identifying strong edges, and preserving the relevant weak
edges, in addition to maintaining some level of noise suppression.
Figure 4.5 Input landsat image
89
Figure 4.6 Output of the edge detection techniques
Finally, hysteresis is used as a means of eliminating streaking.
Streaking is the breaking up of an edge contour caused by the operator output
fluctuating above and below the threshold. Figure 4.6 shows the output of the
different edge detection technique of given input image as shown in
Figure 4.5.
4.2.4
Colour Features
Colour is a visual attribute of object things that results from the light
emitted or transmitted or reflected. From a mathematical viewpoint, the
colour signal is an extension from scalar-signals to vector-signals. Colour
features can be derived from a histogram of the image. The weakness of
colour histogram is that the colour histogram of two different things with the
same colour can be equal. Platt and Goetz (2004) discussed colour features
are still useful for many biomedical image processing applications such as
90
cell classification, cancer cell detection and content-based image retrieval
(CBIR) systems.
In CBIR, every image added to the collection is analyzed to
compute a colour histogram. At search time, the user can either specify the
desired proportion of each colour or submit an example image from which a
colour histogram is calculated. Either way, the matching process then
retrieves those images whose colour histograms match those of the query
most closely.
4.2.5
Shape Features
The shape of an object refers to its physical structure and profile.
Shape features are mostly used for finding and matching shapes, recognizing
objects or making measurement of shapes. Moment, perimeter, area and
orientation are some of the characteristics used for shape feature extraction
technique. The shape of an object is determined by its external boundary
abstracting from other properties such as colour, content and material
composition, as well as from the object's other spatial properties.
4.2.6
Texture Features
Guiying Li (2012) defined texture is a repeated pattern of
information or arrangement of the structure with regular intervals. In a general
sense, texture refers to surface characteristics and appearance of an object
given by the size, shape, density, arrangement, proportion of its elementary
parts. A basic stage to collect such features through texture analysis process is
called as texture feature extraction. Due to the signification of texture
information, texture feature extraction is a key function in various image
processing applications like remote sensing, medical imaging and contentbased image retrieval.
91
There are four major application domains related to texture analysis
namely texture classification, segmentation, synthesis and shape from texture.
Texture classification produces a classified output of the input image where
each texture region is identified with the texture class it belongs.
Texture segmentation makes a partition of an image into a set of
disjoint regions based on texture properties, so that each region is
homogeneous with respect to certain texture characteristics.
Texture synthesis is a common technique to create large textures
from usually small texture samples, for the use of texture mapping in surface
or scene rendering applications.
The shape from texture reconstructs three dimensional surface
geometry from texture information. For all these techniques, texture
extraction is an inevitable stage. A typical process of texture analysis is
shown in Figure 4.7.
Input Image
Pre-processing
Feature extraction
Segmentation, Classification,
Synthesis, Shape from texture
Post-processing
Figure 4.7 Various image analysis steps
92
4.3
TEXTURE FEATURE EXTRACTION
Neville et al (2003) discussed texture features can be extracted using
several methods such as statistical, structural, model-based and transform
information.
4.3.1
Structural based Feature Extraction
Structural approaches represent texture by well defined primitives
and a hierarchy of spatial arrangements of those primitives. The description of
the texture needs the primitive definition. The advantage of the structural
method based feature extraction is that it provides a good symbolic
description of the image; however, this feature is more useful for image
synthesis than analysis tasks. This method is not appropriate for natural
textures because of the variability of micro-texture and macro-texture.
4.3.2
Statistical based Feature Extraction
Statistical methods characterize the texture indirectly according to
the non-deterministic properties that manage the relationships between the
gray levels of an image. Statistical methods are used to analyze the spatial
distribution of gray values by computing local features at each point in the
image and deriving a set of statistics from the distributions of the local
features. The statistical methods can be classified into first order (one pixel),
second order (pair of pixels) and higher order (three or more pixels) statistics.
The first order statistics estimate properties (e.g. average and variance) of
individual pixel values by waiving the spatial interaction between image
pixels. The second order and higher order statistics estimate properties of two
or more pixel values occurring at specific locations relative to each other. The
most popular second order statistical features for texture analysis are derived
93
from the co-occurrence matrix. Statistical based texture features will be
discussed in section 4.4.
4.3.3
Model based Feature Extraction
Model based texture analysis such as fractal model and Markov
model are based on the structure of an image that can be used for describing
texture and synthesizing it. These methods describe an image as a probability
model or as a linear combination of a set of basic functions. The Fractal
model is useful for modeling certain natural textures that have a statistical
quality of roughness at different scales and self similarity, and also for texture
analysis and discrimination.
There are different types of models based feature extraction
technique depending on the neighbourhood system and noise sources. The
different types are one-dimensional time-series models, Auto Regressive
(AR), Moving Average (MA) and Auto Regressive Moving Average
(ARMA). Random field models analyze spatial variations in two dimensions.
Global random field models treat the entire image as a realization of a random
field, and local random field models assume relationships of intensities in
small neighbourhoods. Widely used class of local random field models are
Markov models, where the conditional probability of the intensity of a given
pixel depends only on the intensities of the pixels in its neighbourhood.
4.3.4
Transform based Feature Extraction
Transform methods, such as Fourier, Gabor and wavelet transforms
represent an image in space whose co-ordinate system has an interpretation
that is closely related to the characteristics of a texture. Methods based on
Fourier transforms have a weakness in a spatial localization so these do not
perform well. Gabor filters provide means for better spatial localization but
94
their usefulness is limited in practice because there is usually no single filter
resolution where one can localize a spatial structure in natural textures. These
methods involve transforming original images by using filters and calculating
the energy of the transformed images. These are based on the process of the
whole image that is not good for some applications which are based on one
part of the input image.
4.4
STATISTICAL BASED FEATURES
The three different types of statistical based features are first order
statistics, second order statistics and higher order statistics as shown in Figure 4.8.
Statistical based features
First order
Statistics
Second order
Statistics
Higher order
Statistics
Figure 4.8 Statistical based features
4.4.1
First Order Histogram based Features
First Order histogram provides different statistical properties such as
four statistical moments of the intensity histogram of an image. These depend
only on individual pixel values and not on the interaction or co-occurrence of
neighbouring pixel values. The four first order histogram statistics are mean,
variance, skewness and kurtosis.
95
A histogram h for a gray scale image I with intensity values in the
range I ( x, y )
0, K 1 would contain exactly K entries, where for a typical 8-bit
28
grayscale image, K
256 . Each individual histogram entry is defined as,
h(i ) = the number of pixels in I with the intensity value I for all
0
i
K . The Equation (4.1) defines the histogram as,
h( i )
cardinality ( x, y ) | I ( x, y )
i
(4.1)
where, cardinality denotes the number of elements in a set. The standard
deviation,
and skewness of the intensity histogram are defined in
Equation (4.2) and (4.3).
( I ( x , y ) m) 2
N
skewness
4.4.2
( I ( x , y ) m) 3
N 3
(4.2)
(4.3)
Second Order Gray Level Co-occurrence Matrix Features
Some previous research works compared texture analysis methods;
Dulyakarn et al. (2000) compared each texture image from GLCM and
Fourier spectra, in the classification. Maillard (2003) performed comparison
works bewteen GLCM, semi-variogram, and Fourier spectra at the same
purpose. Bharati et al. (2004) studied comparison work of GLCM, wavelet
texture analysis, and multivariate statistical analysis based on PCA (Principle
Component Analysis). In those works, GLCM is suggested as the effective
texture analysis schemes. Monika Sharma et al (2012) discussed GLCM is
applicable for different texture feature analysis.
96
The GLCM is a well-established statistical device for extracting
second order texture information from images. A GLCM is a matrix where
the number of rows and columns is equal to the number of distinct gray levels
or pixel values in the image of that surface. GLCM is a matrix that describes
the frequency of one gray level appearing in a specified spatial linear
relationship with another gray level within the area of investigation. Given an
image, each with an intensity, the GLCM is a tabulation of how often
different combinations of gray levels co-occur in an image or image section.
Texture feature calculations use the contents of the GLCM to give a
measure of the variation in intensity at the pixel of interest. Typically, the cooccurrence matrix is computed based on two parameters, which are the
relative distance between the pixel pair d measured in pixel number and their
relative orientation
. Normally,
is quantized in four directions (e.g., 0º,
45 º, 90 º and 135 º), even though various other combinations could be
possible.
GLCM has fourteen features but between them most useful features
are: angular second moment (ASM), contrast, correlation, inverse difference
moment, sum entropy and information measures of correlation. These features
are thoroughly promising.
4.4.3
Gray Level Run Length Matrix Features
Petrou et al (2006) defined gray level run length matrix (GLRLM) is
the number of runs with pixels of gray level i and run length j for a given
direction. GLRLM generate for each sample of image fragment. A set of
consecutive pixels with the same gray level is called a gray level run. The
number of pixels in a run is the run length. In order to extract texture features
gray level run length matrix are computed. For each element, (i, j ) the run
length, r of the GLRLM represents the number of runs of gray level i having
97
length j . GLRLM can be computed for any direction. Mostly five features are
derived from the GLRLM. These features are: Short Runs Emphasis (SRE),
Long Runs Emphasis (LRE), Gray Level Non-Uniformity (GLNU), Run
Length Non-Uniformity (RLNU), and Run Percentage (RPERC). These are
quite improved in representing binary textures.
4.4.4
Local Binary Pattern Features
Local binary pattern (LBP) operator is introduced as a
complementary measure for local image contrast. Lahdenoja (2005) discussed
the LBP operator associate statistical and structural texture analysis. The LBP
describes texture with smallest primitives called textons (or, histograms of
texture elements). For each pixel in an image, a binary code is produced by
thresholding, its neighbourhood with the value of the center pixel. A
histogram is then assembled to collect the occurrences of different binary
codes representing different types of curved edges, spots, flat areas, etc.
This histogram is an arrangement as the feature vector result of
applying the LBP operator. The LBP operator considers only the eight nearest
neighbours of each pixel and it is rotation variant, but invariant to monotonic
changes in gray-scale can be applied. The dimensionality of the LBP feature
distribution can be calculated according to the number of neighbours used.
LBP is one of the most used approaches in practical applications, as it has the
advantage of simple implementation and fast performance.
Some related features are Scale-Invariant Feature Transform (SIFT)
descriptor (SIFT is a distinctive invariant feature set that is suitable for
describing local textures), LPQ (Local Phase Quantization) operator, CenterSymmetric LBP (CS-LBP) and Volume-LBP.
98
4.4.5
Auto Correlation Features
An important characteristic of texture is the repetitive nature of the
position of texture elements in the image. An autocorrelation function can be
evaluated that measures this coarseness. Based on the observation of
autocorrelation feature is computed that some textures are repetitive in nature,
such as textiles. The autocorrelation feature of an image is used to evaluate
the fineness or roughness of the texture present in the image. This function is
related to the size of the texture primitive for example the fitness of the
texture. If the texture is rough or unsmooth, then the autocorrelation function
will go down slowly, if not it will go down very quickly. For normal textures,
the autocorrelation function will show peaks and valleys. It has relationship
with power spectrum of the fourier transform. It is also responsive to noise
interference. The autocorrelation function of an image I ( x, y ) is defined in
Equation (4.4) as follows
N
N
I (u , v) I (u
P ( x, y )
x, v
y)
u 0 v 0
N
(4.4)
N
2
I (u , v)
u 0v 0
4.4.6
Co-occurrence Matrix – SGLD
Statistical methods use second order statistics to model the
relationships between pixels within the region by constructing Spatial Gray
Level Dependency (SGLD) matrices. A SGLD matrix is the joint probability
occurrence of gray levels i and j for two pixels with a defined spatial
relationship in an image. The spatial relationship is defined in terms of
distance, d and angle,
. If the texture is coarse and distance d is small
compared to the size of the texture elements, the pairs of points at distance d
should have similar gray levels. Conversely, for a fine texture, if distance d is
99
comparable to the texture size, then the gray levels of points separated by
distance d should often be quite different, so that the values in the SGLD
matrix should be spread out relatively uniformly.
Hence, one of the ways to analyze texture coarseness would be, for
various values of distance d , some measure of scatter of the SGLD matrix
around the main diagonal. Similarly, if the texture has some direction, i.e., is
coarser in one direction than another, then the degree of spread of the values
about the main diagonal in the SGLD matrix should vary with the direction
.
Thus texture directionality can be analyzed by comparing spread measures of
SGLD matrices constructed at various distances of d . From SGLD matrices, a
variety of features may be extracted.
From each matrix, 14 statistical measures are extracted including:
angular second moment, contrast, correlation, variance, inverse difference
moment, sum average, sum variance, sum entropy, difference variance,
difference entropy, information measure of correlation, information measure
of correlation II and maximal correlation coefficient. The measurements
average the feature values in all four directions.
4.4.7
Edge Frequency based Texture Features
A number of edge detectors can be used to yield an edge image from
an original image. An edge dependent texture description function E can be
computed using Equation (4.5) as follows
E | f (i, j ) f (i d , j ) | | f (i, j )
| f (i, j ) f (i, j d ) |
f (i d , j ) |
| f (i, j )
f (i, j
d) |
(4.5)
This function is inversely related to the autocorrelation function.
Texture features can be evaluated by choosing specified distances d. It varies
the distance, d , parameter from 1 to 70 giving a total of 70 features.
100
4.4.8
Primitive Length Texture Features
Coarse textures are represented by a large number of neighbouring
pixels with the same gray level, whereas a small number represents fine
texture. A primitive is a continuous set of maximum number of pixels in the
same direction that have the same gray level. Each primitive is defined by its
gray level, length and direction. Let B(a, r ) represents the number of
primitives of all directions having length r and gray level a . Assume M , N be
image dimensions, L is the number of gray levels, N r is the maximum
primitive length in the images and K is the total number of runs. It is given
by the Equation (4.6) as
L
Nr
(4.6)
B ( a, r )
a 1 r 1
Then, the Equations (4.6) – (4.10) define the five features of image
texture.
Short primitive emphasis =
Long primitive emphasis =
1
Gray level uniformity =
K
L
1
K
a 1 r 1
1
K
Nr
L
B ( a, r )
r2
(4.7)
B ( a, r ) 2
(4.8)
a 1 r 1
2
Nr
L
B ( a, r ) r
a 1
L
Nr
a 1
r 1
K
L
Nr
rB (a, r )
a 1 r 1
2
(4.9)
r 1
1
Primitive length uniformity =
K
Primitive percentage =
Nr
2
B ( a, r )
(4.10)
K
MN
(4.11)
101
4.4.9
Law’s Texture Features
Law’s of texture observed that certain gradient operators such as
Laplacian and Sobel operators accentuated the underlying microstructure of
texture within an image. This was the basis for a feature extraction scheme
based a series of pixel impulse response arrays obtained from combinations of
1-D vectors shown in Figure 4.9. Each 1-D array is associated with an
underlying microstructure and labeled using an acronym accordingly. The
arrays are convolved with other arrays in a combinatorial manner to generate
a total of 25 masks, typically labeled as L5, E5, S5, W5 and R5 for the mask
resulting from the convolution of the two arrays.
Level L5
Edge E 5
Spot
S5
Wave W 5
Ripple R 5
[
[
[
[
[
1
1
1
1
1
4
2
0
2
4
6
0
2
0
6
4
2
0
2
4
1
1
1
1
1
]
]
]
]
]
Figure 4.9 Five 1D arrays identified by laws
These masks are subsequently convolved with a texture field to
accentuate its microstructure giving an image from which the energy of the
microstructure arrays is measured together with other statistics. The
commonly used features are mean, standard deviation, skewness, kurtosis and
energy measurements. Since there are 25 different convolutions, altogether it
obtains a total of 125 features.
For all feature extraction methods, the most appropriate features are
selected for classification using a linear stepwise discriminant analysis.
Among the above mentioned techniques, researchers suggested the
GLCM is one of the very best feature extraction techniques. From GLCM,
102
many useful textural properties can be calculated to expose details about the
image. However, the calculation of GLCM is very computationally intensive
and time consuming.
4.5
GRAY LEVEL CO-OCCURRENCE MATRIX
In 1973, Haralick introduced the co-occurrence matrix and texture
features which are the most popular second order statistical features today.
Haralick proposed two steps for texture feature extraction. First step is
computing the co-occurrence matrix and the second step is calculating texture
feature based on the co-occurrence matrix. This technique is useful in wide
range of image analysis applications from biomedical to remote sensing
techniques.
4.5.1
Working of GLCM
Basic of GLCM texture considers the relation between two
neighbouring pixels in one offset, as the second order texture. The gray value
relationships in a target are transformed into the co-occurrence matrix space
by a given kernel mask such as 3 3 , 5 5 , 7 7 and so forth. In the
transformation from the image space into the co-occurrence matrix space, the
neighbouring pixels in one or some of the eight defined directions can be
used; normally, four direction such as 0°, 45°, 90°, and 135° is initially
regarded, and its reverse direction (negative direction) can be also counted
into account. It contains information about the positions of the pixels having
similar gray level values.
Each element (i, j ) in GLCM specifies the number of times that the
pixel with value i occurred horizontally adjacent to a pixel with value j . In
Figure 4.8, computation has been made in the manner where, element (1, 1) in
the GLCM contains the value 1 because there is only one instance in the
103
image where two, horizontally adjacent pixels have the values 1 and 1.
Element (1, 2) in the GLCM contains the value 2 because there are two
instances in the image where two, horizontally adjacent pixels have the values
1 and 2.
Figure 4.10 Creation of GLCM from image matrix
Element (1, 2) in the GLCM contains the value 2 because there are
two instances in the image where two, horizontally adjacent pixels have the
values 1 and 2. The GLCM matrix has been extracted for input dataset
imagery. Once after the GLCM is computed, texture features of the image are
being extracted successively.
4.6
HARALICK TEXTURE FEATURES
Haralick extracted thirteen texture features from GLCM for an
image. The important texture features for classifying the image into water
body and non-water body are Energy (E), Entropy (Ent), Contrast (Con),
Inverse Difference Moment (IDM) and Directional Moment (DM).
104
Andrea Baraldi and Flavio Parmiggiani (1995) discussed the five
statistical parameter energy, entropy, contrast, IDM and DM, which are
considered the most relevant among the 14 originally texture features
proposed by Haralick et al. (1973). The complexity of the algorithm also
reduced by using these texture features.
Let i and j are the coefficients of co-occurrence matrix, M i, j is
the element in the co-occurrence matrix at the coordinates i and j and N is
the dimension of the co-occurrence matrix.
4.6.1
Energy
Energy (E) can be defined as the measure of the extent of pixel pair
repetitions. It measures the uniformity of an image. When pixels are very
similar, the energy value will be large. It is defined in Equation (4.12) as
N 1
N 1
M 2 i, j
E
i 0
4.6.2
(4.12)
j o
Entropy
This concept comes from thermodynamics. Entropy (Ent) is the
measure of randomness that is used to characterize the texture of the input
image. Its value will be maximum when all the elements of the co-occurrence
matrix are the same. It is also defined as in Equation (4.13) as
N 1
N 1
i 0
j o
Ent
4.6.3
M i, j ( ln(M (i, j )))
(4.13)
Contrast
The contrast (Con) is defined in Equation (4.14), is a measure of
intensity of a pixel and its neighbour over the image. In the visual perception
105
of the real world, contrast is determined by the difference in the colour and
brightness of the object and other objects within the same field of view.
N 1
N 1
Con
i
i 0
4.6.4
2
j M i, j
(4.14)
j o
Inverse Difference Moment
Inverse Difference Moment (IDM) is a measure of image texture as
defined in Equation (4.15). IDM is usually called homogeneity that measures the
local homogeneity of an image. IDM feature obtains the measures of the closeness
of the distribution of the GLCM elements to the GLCM diagonal. IDM has a range
of values so as to determine whether the image is textured or non-textured.
N 1
N 1
i 0
j o
IDM
4.6.5
1
1
i j
2
M i, j
(4.15)
Directional Moment
Directional moment (DM), as the name signifies, this is a textural
property of the image computed by considering the alignment of the image as
a measure in terms of the angle and it is defined as in Equation (4.16)
N 1
N 1
i 0
j o
DM
M i, j i
j
(4.16)
The Table 4.1 shows some of the texture features extracted using
GLCM, to classify an image into water body and non-water body region.
106
Table 4.1 Texture features extracted using GLCM
Energy
0.2398
0.1949
0.3168
0.1524
0.7568
0.1655
0.313
0.2236
0.5483
0.5583
0.5143
0.2486
0.1608
0.4855
0.1613
0.2853
0.1477
0.316
0.3046
0.2796
0.573
0.1729
0.3145
0.7637
0.6113
0.7586
0.3124
0.5817
0.1226
0.1993
0.7293
0.5257
0.3006
0.1576
0.1929
0.1727
0.8759
0.285
0.1382
0.3316
Entropy
6.8042
7.0086
6.4868
6.7707
5.5702
7.3033
6.6852
6.9529
5.8905
5.9409
6.1439
6.6115
6.88
5.9474
6.9496
6.4627
7.0368
5.9372
6.4706
6.4406
6.0185
7.2134
6.804
5.3457
5.8042
5.3523
6.2919
6.0175
7.1201
7.2553
5.5209
6.4206
6.4985
6.8883
7.1205
7.1763
5.0943
6.7064
7.4154
6.7746
Contrast
0.124
0.1904
0.2488
0.2025
0.12
0.1999
0.1464
0.1739
0.1019
0.1524
0.0794
0.1654
0.1993
0.0953
0.1639
0.2106
0.3293
0.1803
0.1998
0.2019
0.1416
0.1497
0.1592
0.0753
0.138
0.0594
0.1397
0.1585
0.2642
0.2249
0.1787
0.091
0.1749
0.1738
0.2127
0.2284
0.0371
0.1587
0.19
0.1325
IDM
6.15E+04
6.01E+04
5.99E+04
5.95E+04
6.28E+04
5.93E+04
6.12E+04
6.10E+04
6.26E+04
6.26E+04
6.31E+04
6.06E+04
6.05E+04
6.28E+04
6.07E+04
6.01E+04
5.73E+04
6.03E+04
6.03E+04
6.02E+04
6.21E+04
6.05E+04
6.09E+04
6.35E+04
6.19E+04
6.37E+04
6.15E+04
6.14E+04
5.73E+04
5.89E+04
6.16E+04
6.30E+04
6.10E+04
6.04E+04
5.86E+04
5.80E+04
6.46E+04
6.09E+04
5.94E+04
6.16E+04
DM
237.2482
324.5662
389.9733
334.9738
272.9614
286.0215
270.9697
289.1159
229.464
253.8671
197.9385
261.9178
302.5752
211.5594
283.433
339.5505
444.9242
281.7755
339.1179
322.4984
288.6223
234.1717
274.1024
184.061
264.5534
164.2614
281.9666
290.4872
337.0597
365.6291
279.578
211.6174
318.4937
291.3021
294.5526
277.6838
137.2638
300.2948
287.7632
268.2331
107
4.7
APPLICATION OF TEXTURE
Texture analysis methods have been utilized in a variety of
application domains such as automated inspection, medical image processing,
document processing, remote sensing and content-based image retrieval.
4.7.1
Remote Sensing
Texture analysis has been extensively used to classify remotely
sensed images. Land use classification where homogeneous regions with
different types of terrains (such as wheat, bodies of water, urban regions, etc.)
need to be identified is an important application.
4.7.2
Medical Image Analysis
Image analysis techniques have played an important role in several
medical applications. In general, the applications involve the automatic
extraction of features from the image which is then used for a variety of
classification tasks, such as distinguishing normal tissue from abnormal tissue.
Depending upon the particular classification task, the extracted features
capture morphological properties, colour properties, or certain textural
properties of the image.
4.8
SUMMARY
This chapter detailed the gray level co-occurrence matrix based
feature extraction to obtain energy, entropy, contrast, inverse difference
moment and directional moment. These texture features are served as the
input to classify the image accurately. Effective use of multiple features of the
image and the selection of a suitable classification method are especially
significant for improving classification accuracy. The chapter 5 discusses
classification techniques for improving accuracy along with their applications.