An Enhanced Glaucoma Identification using FDCT Classified by Multi SVM

DOI : 10.17577/IJERTV3IS090269

Download Full-Text PDF Cite this Publication

Text Only Version

An Enhanced Glaucoma Identification using FDCT Classified by Multi SVM

B. Sravani


Mylavaram, Vijayawada, Andhra Pradesh, India

Prof. B. Ramesh Reddy


Mylavaram, Vijayawada, Andhra Pradesh, India

ABSTRACT Identifying eye diseases (GLAUCOMA) was a complicated process, this was made easy by Multi resolution analysis with the feature extraction process. Texture features within fundus images are actively pursued for accurate and efficient glaucoma classification. In this paper a novel technique proposed energy texture features extraction using CURVELET transformations which is accessible under geometry conditions where wavelets were not defined to satisfy conditions and also compared with WAVELET transformation analysis. SVM classifier is used for the classification process and feature ranking procedure under the extension of multi SVM classifier for a nonlinear classification process we access RBF kernel. This is used for obtaining accurate results. Under the above mentioned conditions the resultant accuracy is about 98.18%.

Keywords Glaucoma, Multi resolution analysis, Texture features, Multi SVM classifier, Curvelet transformations.


    Glaucoma is caused due to various diseases of eyes. It causes mostly due to the pressure on the eye and the effect is on OPTIC NERVE. The optic nerve of the eye carries visual information to the brain. The optic nerve is made up of over one million nerve cells; it is extremely thin about one twenty-thousandth of an inch in diameter. When the pressure in the eye increased, the optic nerve got damaged.

    The disease is discovered in 17th century. It has been a major constraint in causing blindness from 19th century. In 1643 an entomologist named GK. Glaukoma used the entries of cataract, opacity of lens. But according to the civilization Hippocrates this is discovered in 400 B. C.

    Glaucoma is really about the problems that occur as a result of increased Intra Ocular Pressure (IOP). The average IOP in normal population is 14-16 mmHg. In normal population pressure is up to 20 mmHg may be within normal range.

    The sick ones of glaucoma are mainly in the age group of 38 to 45 years. According to the survey of NCBI website around 15 million people are suffering from this disease. This is due to strain and pressure which we forcing on the eye. The database we have used is from Fried rich Alexandria University database using the following web link

    This university helped by providing the secured database of fundus images.

    Automated clinical decision support systems (CDSSs) in ophthalmology, such as CASNET/ glaucoma, are designed to create effective decision support systems for the identification of disease in human eyes. These CDSSs have used glaucoma as a predominant case study for decades.

    Such CDSSs are based on retinal image analysis techniques that are used to extract structural, contextual, or textural features from retinal images to effectively distinguish between normal and diseased samples.

    The texture features are categorised into structural composition and statistical composition. Structural features are categorised into disk area, disk diameter, cup area, cup diameter, cup to disk ratio and topological features extracted from image. Pixel level information is obtained to complete any of the above mentioned schemes. So, here we are considering Statistical process for GLAUCOMA identification.

    These texture features are extracted from curvelet transformations. As advancement to DWT this transformation was used. These features were classified and ranked by using Multi SVM classifier. But the existences were used for linear or nonlinear individually here SVM classifier is used for both levels of transformations.

    This paper consists of seven segments Image acquisition; Image Pre-processing, Curvelet Transformation, Local Energy Theorem, Feature Extraction, Feature Selection and Classification.

    The main motive of this paper is to extract the features of the image effectively and utilise them in a perfect way i.e., to trace out GLAUCOMA is present for a person or not.


    This step is very important because the image has to be a noised freed one, which will help us to achieve better results. A good and clear image eliminates the process of noise removal and also helps in avoiding error calculation. In this case, computational errors are avoided due to

    absence of reflections, because the images have been taken from close proximity using fundus camera. By the help of Fried rich Alexandria University database on fundus this paper work has been completed.

    Fig 1. Left and right eye


    In this section the image is under going to pre- processing stage i.e., image registration this process includes resizing the image to fix to window and converting the image into gray as we are performing the analysis on 2D images. Here we resized our to 256X256 size and converted the image into gray color or 2D by selecting the green portion of RGB this done based on pixel level processing.

    Fig 2. converting image to gray


    The following methodology consists of feature extraction, feature selection, feature ranking and classification schemes. These were followed as follows

    1. For feature extraction we use Curvelet


    2. For feature calculation we use Local Energy Theorem (LET).

    3. For feature ranking and classification schemes we use MULTI SVM technique.


        In this paper curvelet transformations are used for feature extraction and filtration process. Here the decomposition process is gone through under 3600 and this analysis is far known as angular decomposition on all frequency level with a scaling parameter . The process is completely depends on the angular frequency and the scaling parameters with wedges. The process is started from left top wedge values. This helps in feature calculation, classification. The decomposition is shown in below:

        Fig 3. Decomposition using FFT

        Using FFT analysis in the decomposition process to complete the initial phase later each segment is wrapped around the origin then IFFT is going to be applied. In this level it comprises of Low pass, band pass, and high pass filters, this is used for energy preservation.

        Then it will make its entrance into smoothing filter, here low pass filter is used to smooth the pixel values of the wedges. Then renormalisation will takes place, here every part has been moved to unit cell. And finally ridgelet transformation has to be done on the image. This transformation has two modes one is in square mode and second is in circle mode. It is used for tiling and by using Fourier Transformation for angular transformation.

        1. SUB-BAND decomposition is carried out by

          0 , 1, 2, 3 , 4 , . (1)

        2. SMOOTH partitioning is carried out by

          = . . (2)

        3. RENORMALIZATION is carried out by

          = 1. . (3)

        4. RIDGELET transformation is carried out by , = , . (4)

        SVM is a binary classifier which takes the values ±1. This consists of multiple classes based on the dataset and one class is chosen from the rest. Multi class svm is used for maximal output so it uses sgn function.

        , = , +


        p>The feature ranking process is carried out in the initial

        stage of training process. Under this decision function has been executed and carried to classification process.

        = 2 , 2 ,


        Fig 3. Curvelet transformation output

      2. LET (Local Energy Theorem)

        + , (6)

        , =1

        A nonlinear Multi svm classifier under RBF kernel has been used for implementation and feature selection is carried by equation (7)

        Before getting into LET there is a need to understand the concept of image texture. Image texture provides the information about spatial arrangement of intensities in an image or in a selected region. To analyze image texture segmentation has to be under go. By using structural approach or statistical approach these image texture are classified. Under texture we can compute (1) Angular


        , =

        Here K stands for kernel function.



    moment, (2) Contrast, (3) Correlation, (4) Entropy and (5) Energy. All the mentioned five calculating values are under Statistical approach. So we are moving with the statistical approach and calculating energy values after considering the real valued output of Curvelet transformation.


    The entire process is executed under MATLAB

    2012(b) GUI along with the help Curvelet transformation toolbox which has been extracted from

    STEP 1: Select input Fundus image. STEP 2: Image normalisation


    2 + 2

    , 2


    1. Resizing image into 256X256.

        1. SVM (Support Vector Machine)

      FEATURE ranking and FEATURE selection process are carried by MULTI SVM here C, GAMMA functions used to rank the features and select them under five fold cross validation technique. Classification process is also carried of SVM this made possible of obtaining 96.75% to wavelets and 98.18% to curvelets based on feature extraction, feature selection and feature ranking process.

      Fig 4. Nonlinear SVM classifier

    2. Converting image from RGB to GRAY or select GREEN value from RGB image.

    STEP 3: Applying Curvelet Transformation

    1. Sub band Decomposition using FFT.

    2. Smooth Partitioning using High, Low, Band pass filters.

    3. Reconstruction is performed.

    4. Ridgelet Transformations is applied to find and smooth all the edges.

    STEP 4: Selecting the real values from the output image of transformation.

    STEP 5: Texture feature Energy calculation. STEP 6: Feature Extraction for entire dataset.

    STEP 7: Training entire Feature extracted dataset using Multi SVM train.

    STEP 8: Calculating Accuracy depends on Multi SVM Classify using RBF for the entire dataset.

    Comparative analysis between Curvelet transformations and different wavelet transformations.

    Table1: Features and corresponding P-values.





    Db p avg




    Db v1 energy




    SYM2 p avg




    SYM3 v1





    Bior3.3 p avg




    Bior3.5 v1













    Here the proposed work is mainly on multi resolution analysis with texture feature extraction for a reason of future work these can be mostly explained by using neighbouring pixel calculation with component analysis or by using enhancing pixel under genetic calculations.


In this paper a filtration process with local energy theorem of rare combination is provided on all edges of 00

3600. This is done possible by using CURVELET transformations. Because curvelet transformations are used for multi resolution analysis so it needs to integrate all 3600. With the combination of SVM curvelet provides an accuracy of 98.18% this make the system is all set ready for every kind of fundus image data base for detecting glaucoma.


  1. G. R. Cross and A. K. Jain, \Markov Random Field Texture Models," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. PAMI-5, 1983, pp. 25-39.

  2. R. M. Haralick, K. Shanmugam, and I. Dijnstein, \Textural features for image classi_cation," IEEE Transactions on Systems, Man, and Cybernetics, Vol. 3, 1973, pp.610-621.

  3. K. Laws, \Rapid Texture Identi_cation," in SPIE Vol. 238: Image Processing for Missile Guidance, 1980, pp. 376-380.

  4. H. Tamura, S. Mori and T. Yamawaki, \Textural Features Corresponding to Visual Perception," IEEE Transactions on Systems, Man, and Cybernetics, Vol. 8, NO. 6,1978, pp. 460- 473.


  6. American Academy of Ophthalmology (2010). Primary Angle Closure (Preferred Practice Pattern). San Francisco: American Academy of Ophthalmology. Also available online:

  7. American Academy of Ophthalmology (2010). Comprehensive Adult Medical Eye Evaluation (Preferred Practice Pattern). San Francisco: American Academy of Ophthalmology. Available online:

  8. Shah R, Wormald RPL (2011). Glaucoma, search date May 2010. Online version of BMJ Clinical Evidence:

  9. American Academy of Ophthalmology (2007). Vision Rehabilitation for Adults (Preferred Practice Pattern). San Francisco: American Academy of Ophthalmology. Available online:

  10. Salmon JF (2011). Glaucoma. In P Riordan-Eva, ET Cunningham, eds., Vaughan and Asburys General Ophthalmology, 18th ed., pp. 222237. New York: McGraw- Hill.

  11. See JLS, Chew PTK (2009). Angle-closure glaucoma. In M Yanoff, JS Duker, eds., Ophthalmology, 3rd ed., pp. 11621171. Edinburgh: Mosby Elsevier.

  12. Tan JC, Kaufman PL (2009). Primary open-angle glaucoma. In M Yanoff, JS Duker, eds., Ophthalmology, 3rd ed., pp. 1154 1158. Edinburgh: Mosby Elsevier.

  13. Trobe JD (2006). The red eye. Physician's Guide to Eye Care, 3rd ed., chap. 4, pp. 4751. San Francisco: American Academy of Ophthalmology.

  14. Vass C, et al. (2007). Medical interventions for primary open angle glaucoma and ocular hypertension. Cochrane Database of Systematic Reviews (4).

  15. Walker RS, Piltz-Seymour JR (2009). When to treat glaucoma. In M Yanoff, JS Duker, eds., Ophthalmology, 3rd ed., pp. 1211 1215. Edinburgh: Mosby Elsevier.

  16. Yanoff M, Cameron D (2012). Diseases of the visual system. In L Goldman, A Shafer, eds., Goldmans Cecil Medicine, 24th ed., pp. 24262442. Philadelphia: Saunders.


  18. Tony F. Chan and Jackie (Jianhong) Shen, Image Processing and Analysis Variational, PDE, Wavelet, and Stochastic Methods, Society of Applied Mathematics, ISBN 0-89871-589-X (2005)

  19. Ingrid Daubechies, Ten Lectures on Wavelets, Society for Industrial and Applied Mathematics, 1992, ISBN 0-89871-274-2

  20. Gerald Kaiser, A Friendly Guide to Wavelets, Birkhauser, 1994,

    ISBN 0-8176-3711-7

  21. Stéphane Mallat, &qut;A wavelet tour of signal processing" 2nd Edition, Academic Press, 1999, ISBN 0-12-466606-X

  22. Donald B. Percival and Andrew T. Walden, Wavelet Methods for Time Series Analysis, Cambridge University Press, 2000, ISBN 0-521-68508-7.

  23. Press, WH; Teukolsky, SA; Vetterling, WT; Flannery, BP (2007), "Section 13.10. Wavelet Transforms", Numerical Recipes: The Art of Scientific Computing (3rd ed.), New York: Cambridge University Press, ISBN 978-0-521-88068-8

  24. Hsu, Chih-Wei; Chang, Chih-Chung; and Lin, Chih-Jen (2003). A Practical Guide to Support Vector Classification (Technical report). Department of Computer Science and Information Engineering, National Taiwan University.

  25. Duan, K. B.; Keerthi, S. S. (2005). "Which Is the Best Multiclass SVM Method? An Empirical Study". Multiple Classifier Systems. Lecture Notes in Computer Science 3541. p. 278. doi:10.1007/11494683_28. ISBN 978-3-540-26306-7

  26. Hsu, Chih-Wei; and Lin, Chih-Jen (2002). "A Comparison of Methods for Multiclass Support Vector Machines". IEEE Transactions on Neural Networks.

  27. Platt, John; Cristianini, N.; and Shawe-Taylor, J. (2000). "Large margin DAGs for multiclass classification". In Solla, Sara A.; Leen, Todd K.; and Müller, Klaus-Robert; eds. Advances in Neural Information Processing Systems. MIT Press. pp. 547553.

  28. Bilwaj Gaonkar, Christos Davatzikos Analytic estimation of statistical significance maps for support vector machine based multi-variate image analysis and classification

  29. R. Cuingnet, C. Rosso, M. Chupin, S. Lehéricy, D. Dormont, H. Benali, Y. Samson and O. Colliot, Spatial regularization of SVM for the detection of diffusion alterations associated with stroke outcome, Medical Image Analysis, 2011, 15 (5): 729-737

  30. Statnikov, A., Hardin, D., & Aliferis, C. (2006). Using SVM weight-based methods to identify causally relevant and non- causally relevant variables. sign, 1, 4.

  31. John C. Platt (1999). "Using Analytic QP and Sparseness to Speed Training of Support Vector Machines". NIPS.

  32. Ferris, M. C.; Munson, T. S. (2002). "Interior-Point Methods for Massive Support Vector Machines". SIAM Journal on Optimization 13 (3): 783.doi:10.1137/S1052623400374379.

  33. Shai Shalev-Shwartz; Yoram Singer; Nathan Srebro (2007). "Pegasos: Primal Estimated sub-GrAdient SOlver for SVM". ICML.

  34. R.-E. Fan; K.-W. Chang; C.-J. Hsieh; X.-R. Wang; C.-J. Lin (2008). "LIBLINEAR: A library for large linear classification". Journal of Machine Learning Research 9: 8711874.

  35. Zeyuan Allen Zhu et al. (2009). "P-packSVM: Parallel Primal grAdient desCentKernel SVM". ICDM.

B. SRAVANI received her bachelors degree in ECE in the year 2012 from Sreekavitha Engineering College, JNTUH University, Karepalli, Andhra Pradesh, India. She is pursuing Masters in Lakireddy Bali Reddy

College of Engineering, Mylavaram. She was selected for National level workshop conducted by ISRO scientists approved by AICTE on Atmosphere Weather RADAR and GPS.


received his B.Tech., degree in Electronics and Communication Engineering in the year 1995 from NBKRIST, SV University, Tirupati, Andhra Pradesh, India. He obtained M.E., degree in Electronics in the year 2002 from

UVCE, Bangalore, Karnataka, India. Now he is pursuing PhD in the area Improvement in SNR of MST RADAR Signals at JNTU, Hyderabad, India. He has been working as Professor in the department of Electronics and Communication Engineering in Lakireddy Bali Reddy College of Engineering, Mylavaram, Krishna Dist., Andhra Pradesh, India. He has 18 years of teaching experience and published Eight international publications. He is a life member of ISTE and Fellow of IETE.

Leave a Reply