PERFORMANCE EVALUATION OF DIFFERENT TECHNIQUES FOR TEXTURE CLASSIFICATION

Sundarapandian et al. (Eds): CoNeCo,WiMo, NLP, CRYPSIS, ICAIT, ICDIP, ITCSE, CS & IT 07,
pp. 353–361, 2012. © CS & IT-CSCP 2012 DOI : 10.5121/csit.2012.2434
PERFORMANCE EVALUATION OF
DIFFERENT TECHNIQUES FOR TEXTURE
CLASSIFICATION
Ashwini Dange1
, Mugdha Khade2
, Payal Kulkarni3
, Pooja Maknikar4
1
Department of Electronics Engineering, VIT, Pune
ashwini.dange8@gmail.com
mugdha_797@yahoo.co.in
payalkulkarni31@yahoo.co.in
pooja_maknikar@yahoo.in
ABSTRACT
Texture is the term used to characterize the surface of a given object or phenomenon and is an
important feature used in image processing and pattern recognition. Our aim is to compare
various Texture analyzing methods and compare the results based on time complexity and
accuracy of classification. The project describes texture classification using Wavelet Transform
and Co occurrence Matrix. Comparison of features of a sample texture with database of
different textures is performed. In wavelet transform we use the Haar, Symlets and Daubechies
wavelets. We find that, thee ‘Haar’ wavelet proves to be the most efficient method in terms of
performance assessment parameters mentioned above. Comparison of Haar wavelet and Co-
occurrence matrix method of classification also goes in the favor of Haar. Though the time
requirement is high in the later method, it gives excellent results for classification accuracy
except if the image is rotated.
KEYWORDS
Texture, wavelets, co-occurrence matrix, comparison.
1. INTRODUCTION
[4]Texture is that innate property of all surfaces that describes visual patterns, and that contain
important information about the structural arrangement of the surface and its relationship to the
surrounding environment. Texture consists of primitive or texture elements called as texels.
Texture analysis is important in many applications of computer image analysis for classification
or segmentation of images based on local spatial variations of intensity. The important task in
texture classification is to extract texture features which most completely embody the information
of texture in the original image. Here we study two texture classification methods: Wavelet
Transform and Co-occurrence matrix.
1.1 Wavelet Transform
Main feature of wavelet is Multi-resolution analysis and variable window size. Narrow window
gives good time and poor frequency resolution ,wide window gives good frequency resolution
and bad time resolution.

354 Computer Science & Information Technology ( CS & IT )
The WT has a good time and poor frequency resolution at high frequencies, and good frequency
and poor time resolution at low frequencies. [1]
1.2 Co-occurrence matrix
A co-occurrence matrix C is an n×n matrix, where n is the number of gray-levels within the
image. The matrix C (i , j) counts the number of pixel pairs having the intensities i and j. These
pixel pairs are defined by specified distances and directions, which can be represented by a
displacement vector d = (dx, dy), representing the number of pixels between the pair in the x and
y directions.
2. WAVELET
One method of representing image is in the form of 2 dimensional array, each value representing
brightness of the pixel. Smooth variations in these values are termed as Low freq components and
sharp variations as High freq. components. Low freq components constitute base of an image and
High frequency components refine them. In short smooth variations demand more importance
than the details.
A fundamental goal of data compression is to reduce the bit rate for transmission or storage while
maintaining an acceptable fidelity or image quality. Compression can be achieved by
transforming the data, projecting it on the basis of functions, and then encoding this transform.
The basic idea of wavelet transform is to represent any arbitrary function f as a superposition of
wavelets. Any such superposition decomposes f into different scale levels, where each level is
then further decomposed with a resolution adapted to the level. One way to achieve such a
decomposition writes f as an integral over a and b with appropriate weighting coefficients
where,
a=scale (1/frequency)
b=time shift.
Figure 1
By decomposing the image into a series of high-pass and low pass bands, the wavelet transform
extracts directional details that capture horizontal (cH), vertical (cV) and the diagonal (cD)
activity. Since lower spatial frequencies of an image are more signiﬁcant for the image’s
characteristics than higher spatial frequencies, further ﬁltering of the approximation is useful.

Computer Science & Information Technology ( CS & IT ) 355
Types of wavelet
2.1 Haar Wavelet Transform
[2]A Haar wavelet is the simplest type of wavelet. In discrete form, Haar wavelets are related to a
mathematical operation called the Haar transform. The Haar transform serves as a prototype for
all other wavelet transforms. Like all wavelet transforms, the Haar transform decomposes a
discrete signal into two sub signals of half its length. One sub signal is a running average or trend;
the other sub signal is a running difference or fluctuation. The Haar wavelet transform has a
number of advantages such as it is conceptually simple, fast, memory efficient.
[7] The Haar transform is based on the Haar functions, hk(z), which are defined over the
continuous, closed interval [0,1] for z, and for k=0,1,2,…,N-1, where N=2n
. The first step in
generating the Haar transform is to note that the integer k can be decomposed uniquely as
k=2p
+q-1
where 0≤p≤n-1, and
q=0 or 1 for p=0, and 1≤q≤2p
for p≠0.
With this background, the Haar functions are defined as
for
and
Haar wavelet is discontinuous, and resembles a step function.The Haar transform also has
limitations, which can pose a problem for some applications. In generating each of the averages
for the next level and each set of coefficients, the Haar transform performs an average and
difference on a pair of values. Then the algorithm shifts over by two values and calculates another
average and difference on the next pair. The high frequency coefficient spectrum should reflect all
high frequency changes.
2.2 Daubechies Wavelet Transform
[2]The Daubechies wavelet transforms are defined in the same way as the Haar wavelet transform
by computing the running averages and differences via scalar products with scaling signals and
wavelets, the only difference between them consists in how these scaling signals and wavelets are
defined.

This wavelet type has balanced frequency responses but non-linear phase responses. Daubechies
wavelets use overlapping windows, so the high frequency coefficient spectrum reflects all high
frequency changes. Therefore Daubechies wavelets are useful in compression and noise removal
of audio signal processing.
2.3 Sym8
Symlets are generally written as symN where N is the order. Some authors use 2N instead of N.
Symlets are only near symmetric, we used sym8 in this experiment.
General characteristics: Compactly supported wavelets with least asymmetry and highest number
of vanishing moments for a given support width. Associated scaling filters are near linear-phase
filters.
3. CO-OCCURRENCE MATRIX
Gray level co-occurrence matrix (GLCM) has been proven to be a very powerful tool for texture
image segmentation. The only shortcoming of the GLCM is its computational cost [8]. 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 [8].
The following figure shows how gray co-matrix calculates several values in the GLCM of the 4-
by-5 image I. Element (1,1) in the GLCM contains the value 1 because there is only one instance
in the image where two, horizontally adjacent pixels have the values 1 and 1. Element (1,2) in the
GLCM contains the value 2because there are two instances in the image where two, horizontally
adjacent pixels have the values 1 and 2. Gray co-matrix continues this processing to fill in all the
values in the GLCM[9].
Mathematically, a co-occurrence matrix C is defined over an n x m image I, parameterized by an
offset (∆x,∆y),[10]

4. METHODOLOGY
The texture classification algorithm consists of three main steps:
1. Segmentation of regions of interest,
2. Extraction of the most discriminative texture features, and
3. Creation of a classifier that automatically identifies the various textures.
4.1 Algorithm
4.1.1 For Wavelet
1. Standard images are collected and a database is formed for classification.
2. The number of decomposition levels was taken to be 3 and the steps 3, 4 and 5 were
carried out for level times i.e. 3 times.
3. Wavelet transform is applied to each of the image in the database. Each of the types such
as Haar, Sym8 and DB4 are applied and features of the image are extracted.
4. The energyis calculated by dividing addition of all elements of the array CH/CV by the
number of rows multiplied by number of columns in the CH/CV array.
5. If number of levels is reached then calculate energy for CA following the steps 3 and 4.
6. The calculated energy values are stored in an array called featureset. This array will have
seven columns as it is a three level decomposition and the number of rows is equivalent
to the number of images.
7. An image that is to be classified is considered; noise is added, contrast is changed by
histogram equalization. These three images are taken as inputs for the classification.
8. The features of these images are extracted and stored in an array called new feature set
following the steps 3, 4 and 5 above.
9. This new feature set array is compared with the database array, feature set using the
concept of Euclidean distance. This is done by using the function ‘dist’.
10. The minimum value of the distances is calculated and according to the threshold value
the classes are determined.
4.1.2 For Co-occurrence
1. Standard images are collected and a database is formed for classification.
2. We form the co-occurrence matrix of the images in the database along 0 degrees, 45
degrees, 90 degrees, 135 degrees along both the positive and negative axis.
3. The energy of these four matrices is calculated by addition of square of each element in
the four matrices.
4. The calculated energy values are stored in an array, say, feature set. This array will have
four columns as it is formed from four matrices.

5. An image that is to be classified is considered; noise is added, contrast is changed by
histogram equalization. The images are also rotated at various angles. These images are
taken as inputs for the classification.
6. The features of these images are extracted and stored in an array, say newfeatureset,
following the steps 2, 3, 4 and 5 above.
7. This new feature set array is compared with the database array, featureset using the
concept of Euclidean distance. This is done by using the function ‘dist’.
8. The minimum value of the distances is calculated and according to the threshold value
the classes are determined.
5. RESULT TABLES
5.1 Average time for Computations
5.1.1 Original Image:
Table 1
Transform Db4 Haar Sym8 Co-occurence
Time (Sec) 0.1986 0.221 0.1805 4.2799
5.1.2 Noisy Images (Salt & Pepper):
Table 2
Noise(Salt and paper) Haar Co-occurrence
0.02 0.1844 4.344
0.05 0.183 4.344
0.09 0.2097 4.3305
5.1.3 Image with changed contrast (Histogram Equalised):
Table 3
Transform Haar Co-occurrence
Time(sec) 0.2077 4.39
5.1.4 Rotated Image
Table 4
Rotation Haar Co-occurrence
2 degree 0.1914 0.713
4 degree 0.1915 0.73
30 degree 0.193 0.7346

5.2 Accuracy
5.2.1 Original Image
Table 5
Db4 Sym8 Haar Co-occurrence
100% 100% 100% 100%
5.2.2 Noisy Images
Table 6
Noise (Salt and pepper) Haar Co-occurrence
0.02 78% 70%
0.05 70% 22.50%
0.09 15% 15%
5.2.3 Image with changed contrast (Histogram Equalised):
Table 7
Transform Haar Co-occurrence
Accuracy 15% 40%
5.2.4 Rotated Images :
Table 8
Rotation Haar Co-occurrence
2 degree 73.30% 2.50%
4 degree 73.33% 2.50%
30 degree 53.33% 2.50%
As seen from the table of rotation for the Haar wavelet and Co-occurrence Matrix it can be
concluded that the accuracy of co-occurrence matrix method is greatly reduced. It is as less as
2.5% whereas for Haar wavelet it is around 66.66 % (average).
The main reason for this is because Co occurrence matrix applied captures the spatial dependence
of wavelet and detail coefficients depending on different directions and distance specified which
means that the matrix is totally dependent on the values of the neighboring pixels. When the
image is rotated the pixel values are also changed which results in the inaccurate classification
results.
6. CONCLUSION:
As initiated in the abstract and consolidated progressively in subsequent points, the strength and
capabilities of some of the important mathematical descriptions of texture property of an image to
produce perceptual quality texture classification are reviewed and compared.

ACKNOWLEDGEMENTS
We would like to extend our heartfelt gratitude to our Supervisor Mr M. M. Kulkarni for being so
helpful and for providing us with his invaluable time and guidance. We would also like to thank
Prof. Pooja Kulkarni and Prof. Tornekar without who we would not have been able to
successfully complete our project.
REFERENCES
[1] Wavelet tutorial by Robi Polikar .
[2] Mohamed I. Mahmoud, Moawad I. M. Dessouky, Salah Deyab, and Fatma H. Elfouly ‘’Comparison
between Haar and Daubechies Wavelet Transformions on FPGA Technology’’ World Academy of
Science, Engineering and Technology 26 2007
[3] Marc Antoni, Michel Barlaud, Ingrid Daubechies ‘’Image coding using Wavelet transform ‘’ IEEE
transaction on Image Processing vol1 No 2 April 1992
[4] T Ojalaand M Pietikäinen’’ Texture classification’’Machine Vision and Media Processing
Unit,University of Oulu,Finland.
[5] Digital image processing using MATLAB by Gonzales.
[6] Milan Sonka Vaclav hlavac, rojer boyle ’’ Image processing,analysis and machine visio’’,2nd edition,
Thomas learning.
[7] Digital image processing by A. K. Jain

[8] Gray Level Co-Occurrence Matrix Computation Based On Haar Wavelet, M. M. Mokji1, S.A.R. Abu
Bakar2 Faculty of Electrical Engineering University of Technology Malaysia, Malaysia
1musa@fke.utm.my, 2syed@fke.utm.my
[9] http://www.mathworks.in/help/toolbox/images/ref/graycomatrix.html
[10] http://en.wikipedia.org/wiki/Co-occurrence_matrix
AUTHORS
Ashwini Dange
Author is now working as Security Product Developer at Cyberonyx Technologies,
Pune. She has completed her B.Tech Electronics from Vishwakarma Institute of
Technology, Pune. Besides profession she likes interacting with people and reading
books.
Mugdha Khade
Author has completed her graduation in Electronics and Telecommunication from
VIT, Pune. She will be working in Deloitte Consultancy as a Business Technical
Analyst. She is currently interning with ‘Teach For India’ which is a non-profit
organisation. Her hobbies include Basketball, reading and travelling
Payal Kulkarni
Author has completed B. Tech Electronics from Vishwakarma Instit ute of
Technology, Pune.. She is currently working as a Secur ity Project Developer with
Cyberonyx Technologies, Pune. Beside profession her hobbies includes Badminton
and Travelling.
Pooja Maknikar
Author graduated as B. Tech Electronics with Digital Signal processing as majors
from Vishwakarma Institutes of Technology, Pune.Her professional career outside of
academics, is working as ProjectEngineer with WIPRO technologies, Bangalore.
Outside of Professional interests, she reads widely and enjoys Trekking.

PERFORMANCE EVALUATION OF DIFFERENT TECHNIQUES FOR TEXTURE CLASSIFICATION

More Related Content

What's hot (18)

Similar to PERFORMANCE EVALUATION OF DIFFERENT TECHNIQUES FOR TEXTURE CLASSIFICATION (20)

More from cscpconf (20)

Recently uploaded (20)

PERFORMANCE EVALUATION OF DIFFERENT TECHNIQUES FOR TEXTURE CLASSIFICATION