COMPOSITE TEXTURE SHAPE CLASSIFICATION BASED ON MORPHOLOGICAL SKELETON AND REGIONAL MOMENTS

Signal & Image Processing : An International Journal (SIPIJ) Vol.4, No.3, June 2013
DOI : 10.5121/sipij.2013.4303 31
COMPOSITE TEXTURE SHAPE CLASSIFICATION
BASED ON MORPHOLOGICAL SKELETON AND
REGIONAL MOMENTS
M. Rama Bai
Department of Computer Engineering, M.G.I.T, Hyderabad, Andhra Pradesh, India
vallapu.rama@gmail.com
ABSTRACT
After several decades of research, the development of an effective feature extraction method for texture
classification is still an ongoing effort. Therefore , several techniques have been proposed to resolve such
problems. In this paper a novel composite texture classification method based on innovative pre-processing
techniques, skeletonization and Regional moments (RM) is proposed. This proposed texture classification
approach, takes into account the ambiguity brought in by noise and the different caption and digitization
processes. To offer better classification rate, innovative pre-processing methods are applied on various
texture images first. Pre-processing mechanisms describe various methods of converting a grey level image
into binary image with minimal consideration of the noise model. Then shape features are evaluated using
RM on the proposed Morphological Skeleton (MS) method by suitable numerical characterization
measures for a precise classification. This texture classification study using MS and RM has given a good
performance. Good classification result is achieved from a single region moment RM10 while others failed
in classification.
KEYWORDS
Classification, Pre-processing, Morphological Skeleton, Hu Moments, Regional Moments
1. INTRODUCTION
Texture classification is an important step in many computer vision algorithms [1,2,3,4]. In
texture classification, images of same group should be homogeneous with respect to some
characteristics or features and different textures should have significant different features or
characteristics.
We know a good shape representation should provide an accurate and complete description of a
given object. One of the great advantages of shape representation is, only shape can be preserved
instead of the whole image. By this, the storage size will be reduced and the original image can be
reconstructed from the preserved shape. A simple and compact representation of a shape that
preserves many topological characteristics like size, length of a shape, separation of the shapes,
and other qualitative behavioral aspects of the shape can be provided by a skeleton representation.
Such kind of shape representation schemes is useful for fast image retrieval and in image
compression problems. By preserving intrinsic details of shape the original image can be
reconstructed in skeletonization approach. Thus it plays an essential function in human visual
perception in shape recognition and shape analysis problems.
It was Cayley and Sylvester who initially derived the theory of moment invariants based on
analytical geometry. Using their study it was Hu who first introduced the concept of algebraic

32
moment invariants. The set of Hu moments are invariant to any change in an object subjected to
rotation, scale and translation change [5-9]. Most of the shape representation algorithms using
moment invariants consider all pixels of a given image in classification taking a long time to
compute and thus are relatively less inefficient. To overcome this deficiency, a novel method is
proposed which extracts the skeleton of target image by applying preprocessing methods first,
then Regional Moments (RM) derived from Hu moments are applied on the extracted skeleton to
achieve classification in efficient manner.
The paper is organized as follows. Section 2 briefly described the methodology employed
introducing preprocessing methods, skeletonization, Hu moments, regional moments and the
classification algorithm, section 3 deals with results and discussions. Finally in section 4
conclusions are listed.
2. METHODOLOGY
The present section briefly outlines an effective method of shape classification by combining
innovative pre-processing techniques, morphological skeletonization method and regional
moments as shown in Figure 1. Pre-processing mechanism describes various methods of
converting a grey level image into binary image. In this paper, six different pre-processing
techniques are studied. They are mean, median, mode, maximum, minimum and max-min. The
binary images obtained from those pre-processing techniques are then subjected to
skeletonization using morphological methods. In other words, the most possible reduced image
(skeleton) is obtained. Further, to the skeleton extracted binary images the paper evaluated ten
regional moments on five groups of shape pattern images and derived an effective classifier.
Figure 1. Proposed block diagram for Texture classification
2.1 Skeleton and shape
One key technique for shape representation is the skeletonization approach. It has been studied
widely since skeletons have important properties which make them suitable for structural pattern
recognition [10, 11]. Skeletonization methods are of two types: pixel-based and non-pixel-based.
In a pixel-based method, all pixels in a shape are utilized in the skeletonization process. Pixel
based methods often use thinning techniques [11, 12] or distance transforms [13,12]. Contour
pixel of a shape is used for non-pixel-based skeletonization method. Hence, logically one can say
the skeleton shape is represented by its contour [14, 15].
For this, the present section proposes a novel approach for skeletonization based on MS approach.
One of the disadvantages of the existing skeletonization method is that it is not automatic and
needs human interaction. The novelty of the paper is that it advocated an innovative approach for
Input
Image
Convert to binary
using various
pre-processing
techniques
Apply
preprocessing
methods
Apply
skeletonization
method
Extract
skeleton for
the image
Determine the
best RM
classifier
Analyze
classification
results
Apply Regional
Moments RM1
to RM10

33
the extraction of skeleton based on MS scheme, which is completely based on morphology, for
the classification purpose.
The MS scheme is a leading morphological shape representation algorithm [16, 26, 27]. In the
MS scheme, a given shape is represented as a union of all maximal disks contained in the shape.
The advantages of this basic algorithm include that they have simple and well-defined
mathematical characterizations and they are easy and efficient to implement.
In MS scheme the skeleton of an image(I) is derived in terms of simple morphological operations
erosions and openings. The morphological skeleton Sk of an image is defined by the Equation 1.
2.2 Hu Moment Invariants
An image classification problem involves sorting images based on their shape features. This is
achieved by suitable characterization of the object shape. One can easily say whether a given
image fall into the same set based on an important rule. That is ideally one can say dissimilar
objects should have dissimilar categorization and similar objects should have similar description.
A group of algebraic moments based on the combination of general moments were proposed for
the first time by Hu and were known as Hu moments. A greater part of image recognition
experiments achieved good outcomes using those Hu moments[17-21]. These moments have
achieved good results in the majority of 2D and 3D image recognition experiment. Let X and Y
represent the horizontal and vertical axis in 2-Dimension, a point (x,y) represents the gray level
value in the image and is given as f(x,y).

34
2.3 Regional Moments on MS
To illustrate shape of an object or to match it with patterns of similar shape and for classification
problems based on shapes, moment invariants have been in use for a long time. A variety of
texture classification methods are proposed in the literature for the past three decades. But so far
no one has attempted classification of textures based on MS schemes especially using RM. The
shape parameter derived from objects help in attaining it. Hence, the present study derived RM
based on HM.
The RMs is given by RM1 to RM10 in equations 14 to 23.

35
2.4 Classification of shape texture images based on RM
To demonstrate the classification process five sets of images with dissimilar shapes like brick,
circle, curves, lines and zigzag are considered. Each set consists of ten images of similar shape.
These texture images are displayed from Figure 2 to Figure 6. The classification process is given
in Algorithm 1 below.
Figure 2. Original images of brick textures.
Figure 3. Original images of Circles textures.
Figure 4. Original images of Curves textures.

36
Figure 5. Original images of Line textures.
Figure 6. Original images of Zigzag textures.
Algorithm 1: A Novel textures classification method using MS and RM
Classification is performed on five groups of texture images by the following steps.
Step 1: Read the gray scale texture images and change it into a binary one.
Step 2: A binary images is obtained by applying six different pre-processing techniques like
mean, median, mode, maximum, minimum and max-min.
Step 3: Each pre-processed binary image is then subjected to skeletonization using the MS
scheme.
Step 4: Evaluate RM on the skeleton texture generated.
Step 5: On each set consisting of ten textures images, evaluate average values of each RM’s and
store them in a database.
Step 6: Plot the classification graph for all ten RM on MS scheme and determine the significant
RM that classifies accurately and efficiently the given textures.
3. RESULTS AND DISCUSSIONS
3.1 Classification on shape texture by RM using local maximum pre-processing and
MS method
In order to illustrate the classification problem five sets of different shape textures are taken. Each
set comprises of ten images of similar shapes. For all set the average value of each regional

37
moment from 1 to 10 is computed and given in the Table 1. A graph is also displayed based on
these values as shown in Figure 7.
Table 1. Average RM on MS schemes obtained after pre-processing the image using local
maximum.
Image RM1 RM2 RM3 RM4 RM5 RM6 RM7 RM8 RM9 RM10
Brick 0.0898 1.1107 2.9135 2.3506 0.8937 0.1342 0.1860 0.2199 0.7803 59.2652
Circle 0.0308 1.0313 2.1898 3.3016 1.2229 0.1342 0.1739 0.1374 1.4000 133.2895
Curve 0.1234 1.1435 1.6063 1.7279 1.2570 0.3019 0.3401 0.2549 0.6254 83.0574
Line 0.1639 1.4251 1.3655 1.5130 1.0728 0.2716 0.3398 0.3208 0.9978 115.2402
Zigzag 0.0994 1.1127 2.3796 2.1447 1.0944 0.1921 0.2337 0.2120 0.8345 38.7425
Figure 7. Classification graph on average RM’s obtained after pre-processing texture images
using local maximum
3.2 Classification on shape texture by RM using local minimum pre-processing and
MS method
The Algorithm 1 is applied on the same texture images after applying local minimum pre-
processing method. The computed values are given in Table 2. A graph is also displayed based on
these values. Figure 8 gives a representation of it.
Table 2. Average RM on MS schemes obtained after preprocessing the image using local
minimum.
Brick 0.0756 1.0831 2.1575 1.9674 1.0595 0.1143 0.1863 0.1780 0.7368 56.1803
Circle 0.0436 1.0449 3.1076 2.2917 1.0493 0.1069 0.1400 0.1419 0.8891 47.2131
Curve 0.1746 1.2078 1.3197 1.4866 1.2078 0.2758 0.3816 0.2927 0.8514 43.0058
Line 0.1720 1.4639 1.2169 1.2872 1.0428 0.2842 0.3268 0.3241 0.9273 66.6750
Zigzag 0.1115 1.1221 2.3051 2.0425 1.0191 0.1370 0.2164 0.2183 0.6487 32.5595

38
using local minimum
3.3 Classification on shape texture by RM using local max-min pre-processing and
MS method
The Algorithm 1 is applied on the same texture images after applying local Max-Min pre-
Table 3. Average RM on MS schemes obtained after pre-processing the image using local
maxmin.
Brick 0.0333 1.0340 2.0578 1.7756 1.3190 0.3052 0.1750 0.1297 0.8208 96.4573
Circle 0.0194 1.0198 1.4130 1.5529 1.2337 0.1710 0.1210 0.1023 0.9466 269.5946
Curve 0.1376 1.1544 4.8250 4.1798 1.0732 0.3138 0.3206 0.2890 0.9263 69.0026
Line 0.0689 1.0740 1.1491 1.1258 1.1157 0.3371 0.2675 0.2274 1.0667 231.0970
Zigzag 0.0792 1.0876 1.7605 1.7629 1.1150 0.1125 0.1780 0.1591 0.5890 89.6390
using local max-min

39
3.4 Classification on shape texture by RM using local mode pre-processing and MS
method
The Algorithm 1 is applied on the same texture images after applying local mode pre-processing
method. The computed values are given in Table 4. A graph is also displayed based on these
values. Figure 10 gives a representation of it.
Table 4. Average RM on MS schemes obtained after pre-processing the image using local mode
Brick 0.0754 1.0871 2.1417 1.7884 1.0626 0.1429 0.1811 0.1787 0.9167 86.4546
Circle 0.0276 1.0283 2.1994 2.0984 1.0566 0.0756 0.1259 0.1176 0.9296 1299.9533
Curve 0.1640 1.1880 1.2157 1.3007 1.1373 0.3213 0.3586 0.3181 0.9417 72.5362
Line 0.1443 1.4063 1.5700 1.5445 1.2058 0.3532 0.3068 0.2668 1.0231 74.9628
Zigzag 0.0849 1.0917 1.7182 1.8181 1.2964 0.1815 0.2437 0.1944 0.9972 43.5181
using local mode
3.5 Classification on shape texture by RM using local mean pre-processing and MS
method
The Algorithm 1 is applied on the same texture images after applying local mean pre-processing
method. The computed values are given in Table 5. A graph is also displayed based on these
values. Figure 11 gives a representation of it.
Table 5. Average RM on MS schemes obtained after preprocessing the image using local mean
Brick 0.0815 1.0960 2.4534 2.4520 1.1317 0.1576 0.2333 0.2083 1.1167 102.0747
Circle 0.0297 1.0303 3.8515 2.6614 0.8926 0.0761 0.1101 0.1294 0.6858 249.6272
Curve 0.1705 1.2024 2.8460 2.2374 1.1046 0.2557 0.3830 0.3262 0.8872 40.6571
Line 0.1637 1.4257 1.2451 1.2122 1.1184 0.3930 0.3555 0.3151 1.0295 118.1763
Zigzag 0.1087 1.1224 1.6402 1.5388 1.0120 0.1886 0.2285 0.2254 0.6376 40.9032

40
Figure 11. Classification graph on average RM’s obtained after preprocessing texture images
using local mean
3.6 Classification on shape texture by RM using local median pre-processing and
MS method
The Algorithm 1 is applied on the same texture images after applying local median pre-
TABLE 6. AVERAGE RM ON MS SCHEMES OBTAINED AFTER PRE-PROCESSING THE IMAGE USING
LOCAL MEDIAN
Brick 0.0830 1.0983 3.4013 2.8371 1.1507 0.1387 0.2040 0.1977 0.9983 90.6251
Circle 0.0257 1.0261 3.4835 2.6516 0.8904 0.0722 0.1082 0.1179 0.8420 105.3247
Curve 0.1805 1.2119 1.2130 1.2407 1.0669 0.3180 0.3543 0.3416 0.8720 66.5121
Line 0.1505 1.4473 6.3547 3.1408 1.0918 0.2799 0.2888 0.2452 1.0024 103.9337
Zigzag 0.0997 1.1100 2.3648 2.0695 1.2327 0.2775 0.2727 0.2137 0.8874 42.9339
Figure 12. Classification graph on average RM’s obtained after preprocessing texture images
using local median.

41
From this proposed stratagem one can conclude that RM10 alone displays all five groups of
textures in a unique manner while other RM’s have failed. That is RM1 to RM9 show nearly
same values and failed in classification of the textures. The classification obtained after applying
local mode preprocessing method is poor when compared to other proposed methods. In all the
Circle images are having maximum value than other shape textures after application of these
proposed techniques.
4. CONCLUSIONS
The present paper proposes a novel Morphological Skeleton(MS) representation method on
Regional Moments(RM) for classification of textures with similar shape components. A variety of
texture classification methods are proposed in the literature for the past three decades. But so far
no one has attempted classification of textures based on MS schemes especially using RM. The
present paper evaluated RM on MS schemes to classify texture images of different shapes.
Classification has been carried out by applying pre-processing methods first followed by
extracting the skeleton of the target image on which regional moments are computed to achieve
precise classification of textures. The present paper taken into consideration the following
preprocessing methods applied on local neighborhoods, which are listed below. a) maximum,
minimum, mode, median , mean and maxmin i.e((max-min)/2). Results generated prove that
RM10 is sufficient to classify the given group of textures. One need not apply RM1 to RM9 for
classification of any texture images as they have totally failed in classification even in the case of
pre-processed texture images.
ACKNOWLEDGEMENTS
The author would like to thank who directly or indirectly helped in carrying out the work. I also
thank reviewers for their valuable comments, management of MGIT for their encouragement and
family members for their support which led to improvise the presentation quality of this paper.
REFERENCES
[1] A. Laine & J. Fan, (1993) “Texture classification by wavelet packet signatures”, IEEE Trans. on
PAMI, Vol.15, No.11, pp. 1186-1190.
[2] A. Bovik , M. Clark & W. S. Geisler, (1990)“Multichannel Texture Analysis Using Localized Spatial
Filters”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.12, No.1, pp. 55-73.
[3] A. K. Jain & F. Farrokhnia, (1991) “Unsupervised texture segmentation using Gabor filters”, Pattern
Recognition, Vol.24, No.12, pp. 1167-1186.
[4] M. Unser & M. Eden, (1989) “Multiresolution feature extraction and selection for texture
segmentation”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.11, pp. 717-
728.
[5] Ming-Kuei Hu, (1962) “Visual pattern recognition by moment invariants”, Information Theory, IRE
Transactions, Vol. 8, pp. 179-187.
[6] Boyce J. F & Hossack W. J, (1983)"Moment Invariants for Pattern recognition", Pattern Recognition
Letters, Vol. 1, pp. 451-456.
[7] Chen C.C. & Tung-I T Sai, (1993) “Improved moment invariants for shape discrimination”, Pattern
Recognition, Vol.26, No.5, pp. 683-686.
[8] Sluzek A, (1995) “Identification and inspection of 2-D objects using new moment-based shape
descriptors”, Pattern Recog., Vol. 16, pp.687-697.
[9] Teaque M. R, (1980)"Image Analysis via the General Theory of Moments", Journal of the Optical
Society of America, Vol. 70, pp. 920-930.

42
[10] Loncaric S., (1998)“A Survey of Shape Analysis Techniques”, Pattern Recognition, Vol. 31, pp. 983–
1001.
[11] Lam L., Lee S. & Suen C., (1992)“Thinning methodologies –A comprehensive survey”, IEEE
Transactions on Pattern Analysis and Machine Intelligence, Vol. 14, pp. 869-885.
[12] Smith R.W.,(1987) “Computer Processing of Line Images: A Survey,” Pattern Recognition, Vol. 20,
pp. 7–15.
[13] Chang H.S & Yan H, (1999)“Analysis of Stroke Structures of Handwritten Chinese Characters”,
IEEE Trans. Sys. , Vol. 29, pp. 47–61.
[14] Man Cybern B. et al., (1993) “B -spline contour representation and symmetry detection”, IEEE Trans.
Pattern Anal. Mach. Intell., Vol. 15, pp. 1191–1197.
[15] Zou J.J and Yan H, (2001) “Skeletonization of Ribbon-Like Shapes Based on Regularity and
Singularity Analyses”, IEEE Trans. Sys. Man Cybern. (B), Vol. 31, pp. 391–395.
[16] Hu M. K., (1962) “Visual pattern recognition by moment invariants”, IRE 2lan.s. Inform. Theory, IT-
8, pp. 179-187.
[17] Jia-Guu Leu, (1991) “Computing a shape's moments from its boundary”, Pattern recognition, Vol 24,
No. 10, pp. 949-957.
[18] Chaur-Chin Chen, (1993) “Improved Moment Invariants for Shape Discrimanation”, Pattern
Recognition, Vol. 26, No.5, pp.683-686.
[19] H. Sunder & D. Silver et al, (2003) “Skeleton Based Shape Matching and Retrieval”, Proceedings of
the shape modeling International, pp. 130-290.
[20] Gang Xu &Yuqing Lei, (2008) “A New Image Recognition Algorithm Based On Skeleton”, IEEE
International Joint Conference on Neural Network, pp. 777-782.
[21] Zhihu Huang & Jinsong Leng, (2010) “Analysis of Hu’s Moments Invariants on Image Scaling and
Rotation”, IEEE 2nd
International Conference on Computer Engineering and Technology, Vol.7,
pp.476-480.
[22] V. Vijaya Kumar, et al , (2007) “A measure of patterns trends on various types of preprocessed
images”, IJCSNS, Vol.7, No.8, p.p. 253-257.
[23] B.V. Ramana Reddy & A. Suresh et al, (2009)“Classification of Textures based on Features Extracted
from Preprocessing Images on Random Windows”, International Journal of Advanced Science
and Technology, Vol. 9, pp. 9-18.
[24] Andrzej Sluzek, (1995)“Identification and inspection of 2-D objects using new moment-based shape
descriptors”, Pattern Recognition Letters, Vol. 16, pp. 687-697.
[25] Dr V Venkata Krishna & M Rama Bai, (2012)” Morphological Boundary Based Shape
Representation Schemes on Moment Invariants for Classification of Textures", International Journal
of Computer Science and Communication , Vol.-III, No.-I, pp.125-130
[26] Dr M Ashok , J SreeDevi & Dr M Rama Bai, (2012)”An Approach to Noisy Image Skeletonization
using Morphological Methods”, International Journal of Scientific & Engineering Research Vol. 3,
Issue 9, pp.1-8.
[27] Dr. V.Vijaya Kumar, A. Srikrishna & Dr. D.V.L.N Somayajulu, (2008)” An Improved Iterative
Morphological Decomposition Approach for Image Skeletonization”, International Journal of
Graphics, Vision and Image Processing, Vol.8, Issue 1, pp. 47-54.
[28] B. Eswara Reddy & A. Nagaraja Rao et al, (2007)” Texture Classification by Simple Patterns on Edge
Direction Movements”, IJCSNS, Vol.7, No.11, pp. 221-225.
[29] Rafael C. Gonzalez & Richard E. Woods, (2008) Digital Image Processing (Third Edition): Prentice
Hall.
[30] J. Serra, Image Analysis and Mathematical Morphology,(1982) Academic Press, London.

43
Author
Dr. M. Rama Bai received, her B.E degree from Bharathiar University,
Coimbatore(T.N) and her M.Tech (CSE) from College of Engineering, Osmania
University, Hyderabad. She received her Ph.D. degree in Computer Science from
Jawaharlal Nehru Technological University, Kakinada (JNTUK) in 2012. She served
Amrita University, Coimbatore and Sri Hindu College of Engineering, Machilipatnam
before joining in MGIT for some period. She then joined as Assistant Professor in the
Dept of Computer Science & Engineering, Mahatma Gandhi Institute of Technology
(MGIT) in 1999. At present she is working as Professor in Dept of CSE, MGIT. Her research interests
include Image Processing, Pattern Recognition, Digital Water Marking and Image Retrieval Systems. She
has published 18 research publications in various National and International Journals and conferences.

COMPOSITE TEXTURE SHAPE CLASSIFICATION BASED ON MORPHOLOGICAL SKELETON AND REGIONAL MOMENTS

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to COMPOSITE TEXTURE SHAPE CLASSIFICATION BASED ON MORPHOLOGICAL SKELETON AND REGIONAL MOMENTS (20)

Recently uploaded (20)

COMPOSITE TEXTURE SHAPE CLASSIFICATION BASED ON MORPHOLOGICAL SKELETON AND REGIONAL MOMENTS