IRJET- Shape based Image Classification using Geometric –Properties

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 09 | Sep 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 630
SHAPE BASED IMAGE CLASSIFICATION USING
GEOMETRIC –PROPERTIES
M. Ilamathi1, P. Sumathy2
1Research Scholar, School of Computer Science, Engineering and Applications, Bharathidasan University,
Tiruchirappalli, Tamilnadu, India
2Assistant Professor, School of Computer Science, Engineering and Applications, Bharathidasan University,
Tiruchirappalli, Tamilnadu, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - This paper has classified shapes based on their geometrical properties. The shape properties are used to analyze the
object dominant images such as area, perimeter, circularities, eccentricity, etc. Inthiswork, theDiscreteWaveletTransform(DWT)
method is used to remove the noise and compress the image. KNN (K Nearest Neighbor) classification technique is used to classify
the object shapes such as square, circle, ellipse and rectangle. MPEG_7 dataset is used to evaluate the accuracy of the shape. It
gives maximum accuracy for the given dataset.
Key Words: geometrical properties, Discrete Wavelet Transform, K Nearest Neighbor, MPEG_7 dataset
1. INTRODUCTION
Digital image processing is an important area for the current scenario due to rapid growth in information
technologies and World Wide Web because more images are stored in digital form. It is very important to define images in an
effective way to improve automatic searching. The digital image is considered as a numerical representation oftheimage.The
images can be collected in various ways, such as capture the image by the camera/variousdevicesordownloadfromwebsites.
Image processing consists the operations such as image acquisition, segmentation, edge detection, morphological operation,
feature extraction. These processes convert the original image into a digital image. This converted digital image is used to
extract some information from original images, its each image pixels have particular value. It can be stored in an array format.
Some of the applications of DIPs are pattern recognition, video processing microscopic imaging, image sharpening, image
restoration in the medical field, remote sensing, transmission, encoding, and machine/robot vision.
Content Based Image retrieval (CBIR) is a technique where images aredescribedbyfeaturessuchascolor,texture,
shape and combination of them. Shape is the fundamental visual features in CBIR. Various methods arediscussedinitsrelated
works and are categorized into two: they are region based and content based. Content based shape techniques are used only
boundary information, not considering the interior details. They include Fourier descriptor, Wavelet Transformation,
Curvature scale space descriptor. Region based shape descriptor is used both boundary and interior details. Many recent
research works are focused on defining new algorithms and feature extraction techniques are used to detect the shape of the
objects with their properties.
This paper proposes a shape based image classification forpreliminaryshapessuchas;square,rectangle,circleand
ellipse. Geometrical properties are extracted for the preliminary shape images. KNN classifier is used to classify the given
images. Classification accuracy is better than existing approaches. Section 2 elaborates on the recent research work in shape
based image classification and feature extraction techniques. Section 3 provides the three types of properties; they are
geometrical based shape properties, statistical based shape properties,margin based shapeproperties.Section4describesthe
implementation part and finally section 5 concludes the paper.
2. RELATED WORK
There are many approaches proposed based on shape based image retrieval usingtheFourierdescriptorandWavelet
descriptor. Xiaojun and Hongxing [1] have focused on the shape descriptor and theyhavespecifiedtwotypesofshapessuchas
global shape (Fourier Transformation) descriptor andlocal shape(WaveletTransform)descriptor.TheGaussianfuzzymethod
is used to calculate the similarity measure. This method is novel and efficienttoretrieveshapebasedobjects,which isinvariant
to RST (Rotation, Scaling, and Translation). Woo Chaw andSeyedHadi[2].HaveintroducedtheFruitsrecognitionsystemusing
color, shape and size feature analysis. Three combined feature analysis techniqueisusedamongvarioustypesoffruitsandthe
final result gives 90% accuracy. Ehsan Nadernejad et al [3] have described different types of edge detection methods used to
detect the objects and mainly focused to retrieve various matrices to the standard images.

Rajbhadur Yadav et al [4] have applied Fourier Descriptor (FD) and the WaveletDescriptor(WD)technique.Tofind
the shape feature vector value. Euclidean distance is used to measure the shape similarity. Wavelet Descriptor gives better
performance compared to the Fourier Descriptor in this shape based classification they haveidentifiedinvehicular objects. In
the year 2015 Cahya and Kohei [5] compared various contour extraction methods and Fourier Descriptor.EDsused tofindthe
similarity matching between the query image and the database image. If the measured valuesare0(matchtothequeryimage)
and 1(it means does not match to the query image). Shalu Gupta et al [6] have discussed various image features and they are
global and local features. Feature extraction can be specified into four types of extraction such as Chord, diagonal, sub widow
based, connected and nonconnected contour segment pixels. These techniques are used to extract image features. There are
two classifiers used to classify the images. One is KNN and another is SVM. Finally KNN gives better accuracy than SVM.
In order to identify the object similarity matching, shape based image retrieval method is used to Rehmanetal [7]
in the year 2016. It gives better results in CBIR Application. Wavelet Transformation is a signal analysis method and has the
advantage of multiresolution analysis. Pallavi and Megha [8] is applied this technique to face recognition and found that WT
gives better accuracy. ED is used to calculate the similarity betweenthe queryimageandthedatabaseimage.Prochazka [9]has
focused on image restoration using Wavelet Transformation. Signal decomposition includesDiscrete WaveletTransformation
and Discrete Fourier Transform. Two dimensional object recognitionsystemisdeterminedintheyear2018byKamleshKumar
et al [10].WT is applied and orthogonal functions are identified. Finally low pass filter and high pass filter are applied. It is
proved that WT gives better accuracy to identify objects.
3. SHAPE BASED IMAGE RETRIEVAL
Content Based Image Retrieval (CBIR) is the application of computer vision technique to the imageretrieval problem,
the problem of searching digital images in a large database. The research analyses the content of the image rather than
metadata such as keywords, tags or descriptions associated with the image. The term content refers to color,shape,textureor
any other information that is derived from the image itself.
Some researcher has proven that the objects are noticeable based on their visual appearance/features. Among these
features shape is the most important property for recognizing objects. In this section,geometrical basedpropertiesareusedto
extract shape representation that is invariant to translation, rotation, and scaling.
The images are collected for our work is only object dominant. Shape basedpropertiesareextractedfromtheoriginal
image base on geometrical properties. Region props functions are used to measure the image region properties and it
recognizes the image shape. The basic image identification properties are extracted, such as area, bounding box, centroid,
convex area, convex hull, convex image, circularity,eccentricity,equivalentdiameter,eulernumber, extent,extrema,filledarea,
filled image, image, maxferet properties, minferet properties, major axislength,minoraxislength,orientation,perimeter, pixel
index list, pixel list, solidity, sub-array index. It can be considered as the feature vectorvalues.Thegeometrical based featureis
the powerful representation of the image shape recognition.
The black diagram of Figure 2 presented the overall shape detection such as image acquisition,preprocessing,feature
extraction, feature vector generation, classification and shape detection which are given below
Figure1. Black diagram of shape detection

In this proposed work focused on geometrical basedfeaturevalues.Imagefeaturesarebasedonthreetypessuchas
geometrical, statistical, margin based features. Which are as follows:
3.1. Geometrical Based Features:
Shanmugavadivu et al [11] in the year 2012, have proposed 7 geometric based properties, 2 margin basedproperties
for shape based image retrieval applications.
(a) Equivalent Diameter:
Equivalent Diameter (ED) is used to compute thesilhouetteofimagediameterthattheregionhasthesamearea.Itis
given in Eq. (1)
…. (1)
(b) Area:
The area can be determined by the actual number of pixels in the image region. Each pixel has different weights. This
value differs from the binary image area and it is given below
… (2)
(c) Major Axis length:
Major Axis Length can be determined to the Image pixel distance betweenthe majoraxisendpointandgivenbythe
relation. This property only supported for 2-D input label matrices. The result is to measure the object length and it is
… (3)
(d) Roundness:
Roundness can be determined by how to close of the object to the circle shape. The perimeter value determines
the length of the boundary of the object. Which is given below
… (4)
(e) Radius:
Radius can be specified in the half-length of its diameter. It is given below
… (5)
(f) Compactness:
The ratio of the area of an object to the area of a circle with the same perimeter. The mean squared distance of
the object's pixels from the centroid divided by the area. A filled circle will have compactness (CN) of 1, with irregular objects
or objects with holes having a value greater than 1. It is given in eq (6)
… (6)
(g) Elongatedness:
Elongatedness (EN) is the Ratio between the length and width of the object bounding box. The ratiovalueis
between o and 1. It is given below

… (7)
(h) Eccentricity:
Eccentricity (ECT) of the image has the same normalized second central moments as a region. The range of
the value between 0 and 1. Eccentricity value can be determined to the Ratio of the major axis length and minor axis
length. Which is presented below
… (8)
3.2. Statistical Based Shape Feature:
Statistical shape property values can be classified into three categories for further processing such as first order,
second order and higher order. The first order describes one individual pixel (Average and variance). Second and third order
shape properties are determined to the two or more pixel values are relative to each other. Statistical based shape property
values are such as mean,standarddeviation,RMS,homogeneity,correlation,contrast, variance,smoothness,kurtosis,skewness
(a) Mean:
The mean value can be determined to the sum of all image pixel values and dividebythenumberofanimagepixel.Itis
given below
… (9)
(b) Variance:
Variance can be measured to the sum of all squared image pixelsanddividethenumberofthepixel.Whicharepresented
below
… (10)
(c) Standard Deviation:
The square root of the variance value can be determined by the standard deviation. It is given below
… (11)
3.3. Margin based shape feature:
Margin based shape features are used to identify the boundary details. The margin based shape properties are such as
convexity, the standard deviation of edge, shape index.
(a) Convexity:
The convexity value is 1 is the convex object. If the convex has lessthan1valuehavetotheobjectit’sconsideredto
the irregular boundary. Convexity and shape index formula is given below
… (12)
(b) Shape Index:
… (13)

4. IMPLEMENTATION
The MPEG-7 dataset and Matlab tool can be used for further processing. This dataset is used to analyze the
images based on their shapes. The following diagram has shown some of sample images from dataset. Fig 3 (a) represents
original image, (b) represents contour image and (c) defined silhouette image
(a) (b) (c)
Figure3. Sample images from dataset (a) Original image (b) Contour image (c) Silhouette image
In preprocessing includes such as image acquisition, segmentation, edge detection, image enhancement,
noise removal, morphological operation. Gaussian filter method is used in DWTtechnique.Itcanfilterthecoefficient valueand
identify the exact image shape. The active contour method is appliedtoanalyzetheregionoftheimageanditisprocessedmore
than 100 iterations to extract the boundary of the image. Sobel edge detection technique used to calculate the gradient of the
image intensity within each image pixel. This method used to detect the edgesoftheinputimageandincreasethebrightness of
their boundaries. The dilation operation is applied to further processing. Strel function used to find the nearest neighborhood
pixels and dimensionality. This morphological operation reconstructs theimage withoutanychangesinitspixels andstoredin
a special format of decomposition image. The decomposition of an image is known as a silhouette image. DWT applied to
remove the noise and filter the image. The resulting of the boundary classification are twoclassessuchasinternal boundaryor
global. In this proposed work DWT technique mainly used to compress the image.
Twelve geometrical shape features are extracted to evaluate such as area, area percentage, perimeter,
minimum radius, maximum radius, equivalent diameter, compactness, elongatedness, circularity, modified circularity,
eccentricity, thinners ratio. In table1, have presented each image is listed in the column and 12 geometricpropertiesarelisted
in the row. Which is presented below
Table-1: Geometrical shape feature extraction

KNN classification technique is used to classify this work. Four types of shapes are classified in this proposed work
such as circle, rectangle, square, and ellipse. This classification is based on the geometrical shapes of feature values. KNN
classification technique give the 96% accuracy. Accuracy formula is given below
Classification Accuracy = (Images classified correctly/ All images classified)*100 … (14)
Table-2: Degree of primitive shape of the object
5. CONCLUSION
In this research work, the proposed method recognizes the primitive shapes using geometrical based properties.
DWT technique has been applied to compress the given input image and geometrical features are extracted using DWT. For
experimental purpose MPEG_7 dataset is used and KNN classifier is applied for image classification performance of the
proposed method gives 96% accuracy
References:
[1] Xiaojun Qi and Hongxing Zheng," a new integrated method for shape based imageretrieval"ComputerScienceDepartment,
Utah State University, Logan, UT, 84322-4205
[2] Woo Chaw Seng, Seyed Hadi Mirisaee, "a new method for fruits recognitionsystem",UniversityofMalaya,September2009.
[3] Ehsan Nadernejad, Sara Sharifzadeh, Hamid Hassanpour, "edge detection techniques: evaluation and comparisons”
Department of Computer Engineering, Faculty of Engineering Mazandaran Institute of Technology P.O. Box: 744, Babol, Iran
ehsan_nader@yahoo.com, Vol. 2, 2008, no. 31, 1507 – 1520.
[4] Raj Bahadur Yadav, Naveen K Nishchal, Arun K Gupta and Vinod K Rastogi, "vehicular shape-based objects classification
using Fourier descriptor technique", 'vol.68, June 2009, pp.484-495.
[5] Cahaya Rahmad, Kohei aria, "comparison contour extraction based on layered structure and Fourier descriptor on image
retrieval", international journal of advanced computer science applications, vol.6, No.12, 2015.
[6] Shalu Gupta, Yumnam Jayanta, Munish Kumar,”object detection using multiple shape based features”, fourth international
conference on parallel, distributed and grid computing, 2016.
[7] M.Rehman, M.sharif, M.Raza, "shape feature extraction method for content based image retrieval", Sindh Univ. Res. Jour.
(Sci. Ser.) Vol. 48 (1) 45-48 (2016).
[8] Pallavi D.Wadkar, Megha Wankhade, "Face recognition using Discrete Wavelet Transforms ", International Journal of
Advanced Engineering Technology, E-ISSN 0976-3945, IJAET/Vol.III/ Issue I/January-March, 2012/239-242

[9] A. Prochazka, J. Ptacek, and I. Sindelarova, "Wavelet Transform In Signal and Image Restoration", Prague Institute of
Chemical Technology, Department of Computing and Control Engineering Technica 1905, 166 28 Prague 6, Czech Republic.
[10] Kamlesh Kumar, Riaz Ahmed Shaikh, Rafaqat Hussain Arain, SafdarAliShah,HidayatullahShaikh,“2-DObject Recognition
Approach using Wavelet Transform”,(IJACSA)International Journal ofAdvancedComputerScienceandApplications,Vol.9,No.
3, 2018.
[11] P.Shanmugavadivu, P.Sumathy, A.vadivel, “image retrieval fromdistributedenvironmentusinggeometricfeatures”,IEEE-
(ICAESM) international conference on advances in engineering, science and management 2012
[12] P.Shanmugavadivu, P.Sumathy, A.vadivel, “Fuzzy Object Shape for Image Retrieval Applications”, ELSEVIER journal,
volume 171, 2016

IRJET- Shape based Image Classification using Geometric –Properties

More Related Content

What's hot (18)

Similar to IRJET- Shape based Image Classification using Geometric –Properties (20)

More from IRJET Journal (20)

Recently uploaded (20)