Classification of Gold Bangles Based on Tamura Texture Features Essay

Categorization of Gold Bangles based on Tamura Texture Features

Abstract— Mortgaging gold for money in the bank is common in India. Banks rely on assessor’s accomplishments to prove the pureness of gold, its weight and supply a description of the points. The absence of skilled assessor makes the loan allowing procedure boring and clip devouring when the quantum of gilded points is mortgaged. This paper provides an image processing solution to automatically supply a description of the mortgage of gold bracelet that would go a ready to hand note for borrowers every bit good. This work classifies the gold bracelets by disk shape and texture characteristics. The proposed work is oriented towards sorting the bracelet into different categories as field bracelet, Stone bracelet and kada bracelet utilizing SVM classifier. Its truth is obtained as 86.66 % and KNN classifier is used for comparing.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now

Index Footings —facet ratio, disk shape, texture characteristics, KNN classifier, SVM classifier.


I. Introduction

N the present Indian scenario, gold has become an of import portion of life. The jewellery industry maintains an of import place for economic growing in Indian economic system. It is a taking foreign exchange earner, every bit good as one of the fastest turning sectors in the state. Among the different types of cherished metals used, gold is one of the most expensive and tempting metal. There are many types of jewelleries which vary in a immense scope of forms and designs. Gold jewellery signifiers around 80 % of the Indian jewellery market, with the balance comprising studded jewellery that includes diamond and other cherished and semi-precious rocks. Among the different types of jewelleries, gold bracelets are the widely used 1s by adult females today.

Object acknowledgment and categorization has been an country of involvement in the field of Image Processing [ 1 ] . Though many research workers have put attempts on acknowledgment and categorization of solid articles, this work attempts to sort the bracelet into a Plain bracelet, Kada bracelet and Stone bracelet. In any object sensing attack, input images may incorporate objects of changing size and textures [ 2 ] . Hence, placing bracelets among other decorations and sorting them into several categories is a ambitious undertaking.

Haralick et Al. [ 3 ] have developed a set of characteristics for sorting or categorising pictural informations based on grey tone spacial dependences, and illustrated their application in category designation undertakings. Tamura et al [ 4 ] have proposed textural characteristics which correspond to human ocular perceptual experience. Six basic textural belongingss ( i.e. , saltiness, contrast, directivity, line-likeness, regularity, and raggedness ) have been measured by human topics. Ya-Li Qi [ 16 ] proposed a relevant feedback retrieval method based on texture characteristics. In their work, Tamura characteristic is used as a texture characteristic to work out the semantic spread. Ground-based cloud acknowledgment for automatic cloud observation utilizing Tamura characteristic is proposed by Chen Xiao-ying [ 17 ] . Bobo wang et Al [ 18 ] used the Tamura characteristic for gauging the crowd denseness.

The remainder of this paper is organized as follows: Section II explains about the proposed methodological analysis. The Experiments and consequences are reported in Section III and concluding subdivision IV concludes the proposed method.

II. Proposed methodological analysis

The proposed method helps to place bracelets among different gold decorations. A box covered with black coloring sheet is taken into consideration and the decoration which is to be classified is kept indoors. A camera is placed at an angle of 45° and it is kept still. Images are captured with unvarying background and formatted to jpeg for standardisation. The side position of the bracelet is acquired at changeless light and fixed distance. Geometric characteristics and disk shape are measured to place whether the given decoration is a bracelet or non. Aspect ratio is considered for sorting different assortment of bracelets.

The overall Block diagram of the proposed work is shown in Fig 1.

A. Image preprocessing

The primary measure is to change over the RGB colour image to binary image for the given ornament image. Subtract the dilated image from the original image, to obtain the boundary of the decoration. From the outermost boundary, perimeter is calculated. This boundary is localized and further processing is done on it.

Fig 1. Block diagram for proposed methodological analysis

B. Feature extraction

Categorization of the bracelet is implemented in two phases. First, form designation is done to happen if the given decoration is a bracelet or other decoration by agencies of disk shape. Then, texture analysis is done in order to sort the bracelet based on texture characteristics like saltiness, contrast, directivity, line-likeness, regularity, raggedness.

Circularity describes the form of an object as in ( 1 ) . The value is 1 for perfect circle and between 1 and 1.5 for oval. A bangle image is acquired in side position to acquire the inside informations about the etched designs. The form of the acquired bracelet image is ellipse as shown in Fig 2. , whereas for earring, the form differs and therefore the disk shape will be greater than 1.5 as shown in Fig 3.

iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ iˆ( 1 )

TABLE I. Geometric Feature Extraction For Bangles

Types of bracelet





Plain Bangle





Kada Bangle





Stone Bangle





Fig 2. Shape description for bracelet

Fig 3 Shape description for Drop earring

TABLE II. Geometric Feature Extraction For Drop Earring

Types of other decorations





Drop earring 1





Drop earring 2





The scopes of the extracted characteristics ( country, margin, solidness and disk shape ) for different bracelets and bead earrings are shown in Table I and Table II, severally.

A part of a bracelet is cropped in such a manner that the length is unbroken consistent for all trial instances whereas width depends on bangle assortment. A Cropped part is marked as green rectangular box in Fig 2 for mention. The aspect ratio is calculated as follows

R=Length / Width( 2 )

It is found to be larger for field bracelet than kada and rock bracelet as tabulated in III. Scaling job is besides alleviated by facet ratio attack.

TABLE III. Aspect Ratio For Various Bangles

Type of Bangle


Aspect Ratio

Plain Bangle



Stone Bangle



Kada Bangle



C. Tamura texture characteristic

Coarseness relates to spacial assortment of grey degree that is implicitly to the size of crude elements. The norm is taken over the vicinity of sizeat every point ( x, y ) of grey image which can be computed as ( 3 ) :

( 3 )

Whereis the grey obtained as the value which maximizes the differences of the traveling norms along the horizontal and perpendicular waies. The saltiness can be computed as ( 4 )

( 4 )

Wheredenotes the image size and the amount is carried out for every pel. A harsh texture will incorporate less figure of big primitives, while a smooth texture contains a big figure of little primitives.

Contrast refers the difference in strength between neighbouring pels and grey degree distribution. A high contrast, texture will hold big differences in strength between neighbouring pels, where a texture of low contrast has little difference. For ciphering the contrast, Kurtosiscan be calculated from ( 5 ) .

( 5 )

Whereis the 4th minute about the mean and discrepancy is. Contrast can be computed as follows

( 6 )

Nis the average Grey degree, i.e. the first order minute of the gray degree chance distribution. The valueN=0. 25 is recommended as the best for know aparting the textures.

Directionality is a planetary belongings over the given image part. This characteristic involves both the form of texture primitives and their arrangement regulation. Histogram of local border chances is calculated against their way, for this magnitudeare approximated as ( 7 ) follows as:

=0.5 ( |?ten(ten, Y) | + |?Y(ten, Y) | )

= sunburn-1(?Y(ten, Y) /?ten(ten, Y) ) ( 7 )

( 7 )

Where?ten(ten, Y) and?Y(ten, Y) are the horizontal and perpendicular Greies flat differences between the adjacent pels, severally.

The coveted histogramcan be obtained fromand points with the magnitudeas follows ( 8 ) :

( 8 )

Whereis the count of points that can be calculated by& lt ;andT, where values n=16 and t=12 are considered.

A directional texture has one or more recognizable orientation of primitives. Directionality () can be computed as in ( 9 ) .

( 9 )

whereNPis the figure of extremums,is the place of thePThursdayextremum,tungstenPis the scope of PThursdayextremum,Rdenotes a normalizing factor,way of happening of grey degree strength.

The line similitude characteristicFLinis defined as an mean happenstance of the border waies that co-occurred in the braces of pels separated by a distancevitamin Dalong the border way in every pel. The border strength is expected to be greater than a given threshold extinguishing fiddling “ weak ” borders. The happenstance is measured by the cosine of the difference between the angles, so that the accompaniments in the same way are weighted as +1 and those in the perpendicular waies by -1 can be computed as follows ( 10 )

( 10 )

Whereis the N?nlocal way accompaniment matrix with distance vitamin D.

Regularity characteristic refers to fluctuations of the texture crude arrangement regulation. A regular texture is composed of indistinguishable to similar primitives while an irregular texture is composed of assorted primitives which are indiscriminately arranged the regularity (Freg) is defined as ( 11 )

Freg=1-R(schromium+scon+sdir+sLin) ( 11 )

WhereRis a normalizing factor and eachsmeans the standard divergence of the corresponding characteristicF.

Roughness describes the fluctuation of physical surface as given by ( 12 ) merely summing the saltiness and contrast steps

Frgh=Fchromium+Fcon( 12 )

D. Classification utilizing SVM

For categorization, the proposed method uses the Support vector machine Classifier. SVM is the most used method in pattern acknowledgment and object categorization. The cardinal thought of SVM is to set a discriminating map so that it makes optimum usage of the dissociable information about boundary instances. Give a set of instances, developing an SVM consists of seeking for the hyper plane that leaves the largest figure of instances of the same category on the same side, while maximising the distance of categories from the hyper plane. The hyper plane in higher dimensional infinite constitutes the set of points whose interior merchandise with the vectors in that infinite is changeless. If the preparation set is linearly dissociable, so a dividing hyper plane, defined by a normal tungsten and a prejudice B, will fulfill the inequalities ( 13 ) :

( 13 )

Where eleven ?? is a instance of the preparation set ( one = 1, … , N ) .

Bing the dimension of the input infinite, and YI? { -1, +1 } is the corresponding category. Among the dividing hyper planes, the

SVM attack selects the 1 for which the distance to the closest instance is maximum. Since such a distance is 1/ tungsten, happening the optimum hyper plane is tantamount to minimising tungsten under restraints.

III. Results And Discussion

The scopes of the characteristic extracted are shown in Table IV. It is found that for kada bracelet, the form of the design will be repeated and therefore there will be difference in grey strength. This fluctuation in the grey strength is measured through the Tamura texture characteristic. From the extracted characteristics, it shows that saltiness is the step of strength fluctuation due to repeated fluctuation in primitives, the value will be high for kada bracelet. While for rock bracelet, the saltiness will be similar to kada bracelet but contrast value will be more than a field bracelet. Directionality and roughness value will be relatively higher for kada bracelet than other bracelets. For apparent bracelet at unvarying strength, the regularity texture characteristic will be regular. SVM classifier is used to sort the bracelet into different assortment. Accuracy of SVM is achieved about 86.66 % . Experiment is performed in MATLAB.

TABLE IV. Texture Feature Extraction





















































































Fig 4. Classified Bangle Output

Fig 4 shows the images of trial images of each assortment. It is seen that every sample is classified to allow categories.

TABLE V. Confusion matrix for the proposed method

Overall Accuracy ( 86.66 % )

Target Classs

Plain bracelet

Kada bracelet

Stone Bangle

Plain bracelet

85 %

10 %

5 %

Kada bracelet

4 %

89 %

7 %

Stone bracelet

0 %

14 %

86 %

The confusion matrix for the proposed method is shown in the Table V. The most evident behaviour of the classifier is quantified by confusion matrix which shows a strong correlativity between true category and categorization consequence. This method achieves the overall truth of 86.66 % .

TABLE VI. Accuracy for different classifier

Type of classifier



86.66 %


82 %

With a proper preparation procedure, it shows that SVM outperforms KNN classifier by 5 % .

IV. Decision

This work has proposed a method to sort gold bracelets based on its texture characteristic. Shape description utilizing geometric characteristics are besides taken into consideration for know aparting bracelet from other decorations. The proposed algorithm uses the efficient characteristics like object disk shape and texture analysis. Based on these characteristics, bracelets are classified into field, Kada and Stone bracelet utilizing SVM. This mechanization is applied in the field of banking for gold ornament mortgage and jewellery store for depicting the cosmetic assortment.


  1. Kumar D, Vinu Prasad G, ” Image Processing Techniques to Acknowledge and Classify Bottle Articles” , International Journal of Computer Applications. Volume 2, 2012, 101-104.
  2. S. F. Lilhare ; DrN. G. Bawane, ” Classification of paddy assortments utilizing image processing” , International Journal of Computer Applications, 2012, Volume 1, 32-45.
  3. R.M. Haralick, K. Shanmugam, and I. Einstein, “Textural Features for Image Classification” , IEEE Transactions on Systems, Man and Cybernetics, 1973. 610-621.
  4. Tamura Hideyuki, Shunji Mori, and Takashi yamawaki, “Textural Features Matching to Visual Perception” , IEEE Transaction of systems Man and Cybernetics.1978, 460-473.
  5. V. Vapnik. The Nature of Statistical Learning Theory. New york: Springer-Verlag. 1995.
  6. Lei Zhang, Tsinghua Univ. , ” support vector machine for image retrieval” , IEEE Transaction on Image processing. 721 – 724 Vol. 2, 2001.
  7. Bottema, ” Circularity of objects in images” , IEEE International Conference on continuing Acousticss, Speech, and Signal Processing, Volume:6.2000, pg. 2247 – 2250 Vol. 4.
  8. M. J. Monika Sahu, M.A. Rizvi, ” optimized Image Retrieval System: Texture and Shape Based Approach” , International Journal of Computer Application, Volume 4, 2013, 201-213.
  9. Sklansky, J, “Image Segmentation and Feature Extraction, ” IEEE Transactions on Systems, Man, and Cybernetics, SMC-8, pp. 237-247, 1978.
  10. R. C. Gonzalez, R. E. Woods, “Digital Image Processing” . Prentice Hall, 2004
  11. Juan Martinez-Alajarin, Jose D. Luis-Delgado, and Luis M. Tomas Balibrea.”Automatic System for Quality-Based Classification of Marble Textures” , IEEE minutess on systems, adult male, and cybernetics—part degree Celsius: applications and reappraisals, vol. 35, no. 4, November 2005.
  12. Somkait Udomhunsakul, Pichet Wongsita, ” Feature Extraction in Medical MRI Images ” Proceedings of IEEE Conference on Cybernetics and Intelligent Systems Singapore, 1-3 December, 2004.
  13. Durgesh K. , Srivastava, and Lekha bhambhu, ” Data Classification utilizing Support Vector Machine” , Journal of Theoretical and Applied Information Technology, volume 23,2005 – 2009.
  14. H.C Lin, L.L. Wang, and S.N Yang, ” Extracting cyclicity of a regular texture based on car right map, ” Pattern Recognition Letters, Vol 18, no. 5, pp. 433-443,1997.
  15. Ya-Li Qi, ” A Relevance Feedback Retrieval Method Based on Tamura Texture” , IEEE, published on Knowledge Acquisition and Modeling, Nov. 30 2009-Dec. 1 2009.
  16. Chen Xiao-ying, Ai-guo Song, Yuan Wen, Zhen Jun-jie, ” On the application of BEMD and Tamura textural characteristic for acknowledging ground-based cloud” , International Conference on Computer Application and System Modeling,22-24 Oct. 2010.
  17. Boba Wang, Hong Bao, Shan Yang, Haitao Lou.”Crowd Density Estimation Based on Texture Feature Extraction” , Journal of Multimedia, Vol, no 4, pg no.331-337, Aug 2013.

I'm Ruth!

Would you like to get a custom essay? How about receiving a customized one?

Check it out