Performance Analysis on Texture Based Image Retrieval using Perceptual Model and MOGG

Performance Analysis on Texture Based Image Retrieval using Perceptual Model and MOGG

C. Baicy A. Anbarasa Pandian Dr. R. Balasubramanian

ME Scholar, Department of Computer Science and Engineering Manonmaniam Sundaranar

University, Tirunelveli-627012, India.

Research Scholar, Department of Computer Science and Engineering

Manonmaniam Sundaranar University, Tirunelveli-627012 India.

Professor, Department of Computer Science and Engineering Manonmaniam Sundaranar

University, Tirunelveli-627012 India

Abstract – The rise of interest in techniques for retrieving images on the basis of automatically-derived features such as color, texture and shape are increasing and this technology is named as Content Based Image Retrieval. In this paper, texture hasbeen used as feature and feature extraction was done by using Perceptual model and Mixtures of Generalized Gaussian Distribution. The perceptual features taken for analysis are Coarseness, Contrast, Directionality and Busyness. The Brodatz Database images are taken for analysis. The precision, recall and time are used as metrics to compare the performance of these methods.

Keywords: Content Based Image Retrieval, Mixtures of Generalized Gaussian Distribution, BrodatzDatabase

1. INTRODUCTION

Content based image retrieval, a techniquewhich uses visual contents to searchimages from large scale image databases according to users' interests, has been anactive and fast advancing research area since the 1990s. During the past decade,remarkable progress has been made in both theoretical research and systemdevelopment. However, many challenging research problems like speed and accuracy of the retrieval system thatcontinue to attract researchers from multiple disciplines.

Textureplays an important role in many machine vision tasks like surface inspection, scene classification, and surface orientation and human visual perception. Texture is characterized by the spatial distribution of gray levels in a neighborhood. Thus, Texture cannot be defined for a point. The ability to match and retrieve texture similar images is an important factor in differentiating the areas on images with same color (such as leaves, sky ,etc).

Tamura feature is a statistical method. Image texture such as the degree of contrast, coarseness, directionality and regularity [Tamura et al,1978], or periodicity, directionality and randomness [Liu and Picard, 1996]. In recent years attention to Gaussian is quite increasing[5,11,4].Mixtures of Generalized Gaussian is an extension of Generalized Gaussian and it has parameters like mean, variance, weight and shape parameter. The similarity measure used was Kullback-Leibler divergence.

1. RELATED WORK

   Texture features play an important role in computer vision and image processing. There are many available texture basedimage retrieval system in the academic arena [2,3,10].Abbadeni. N,D. Ziouand S.Wang[1] estimated perceptual textural features namely coarseness, contrast, directionality, and busyness based on auto covariance function. Liu. F and R.W Pichard[8] here an image model with a new set of features that address the challenge of perceptual similarity. It is useful for use in large collections. Manjuth. B. S and Ma. W. Y[6] used Gabor wavelet features for texture analysis and focuses on a multiresolution representation based on Gabor filters. Minh[4], here a new wavelet based texture retrieval method that is based on accurate modeling of the marginal distribution of wavelet coefficients using Generalized Gaussian Density(GGD).

2. FEATURE EXTRACTION

   Feature extraction is the basis for CBIR. Here, we have used texture feature extraction. Texture refers to visual patterns with properties of homogeneity that do not result from the presence of only single color or intensity. Textures are represented by texels, which are then placed into a no. of sets, depending on how many textures are detected in the image.Tree, clouds, water, bricks and fabrics are the examples of texture.Micro textures refer to textures with small primitives while macro textures refer to textures with large primitives [12].

   Texture analysis techniques have been used in several domains such as classification, segmentation, shape from

   1

   1 n 1 m 1 (i,j) m 1 n 1 (i,j )

   ()

   Cs=

   i =0 j =0 + j =0 i =0

   texture and image retrieval[10,8,6,3].Typically texture 2 n m

   features include contrast, uniformity, coarseness,

   roughness, frequency, density and directionality[1,8].If the features extracted from the images are presented as multidimensional points, thedistances between corresponding multi-dimensional points can be calculated. Cosine and KLD are the distance measure used in this paper.

   PRATICAL APPLICATIONS OF CBIR

   • Crime prevention

   • The military

   • Architectural and Engineering design

   • Fashion and Interior design

   • Medical diagnosis

   • Cultural heritage

   • Home entertainment

   • Web searching

   Let Maxx(i, j)= 1 if pixel (i,j) is a maximum on rows and Maxx(i, j)= 0 if pixel (i,j) is not a maximum on rows. Similarly, Let Maxy (i,j)= 1 if pixel (i,j) isa maximum on columns and Maxy (i,j)= 0 if pixel (i,j) is not a maximum on columns.

   (2)Directionality

   Directionality is a global property in an image. It measures the degree of visible dominant orientation in an image.The degree of directionality is related to the dominant orientation(s) in an image, and refers to the number of pixels having the dominant orientation(s).

   The degree of directionality Ndof an image can be expressed by the following equation

   1 1 (, )

   Sample Images

   N = =0 =0

   d

   d represents dominant orientation. (3)Contrast

   (2)

   Contrast measures the degree of clarity with which one can distinguish between different primitives in a texture.

   n 1 m 1 M i,j t(i,j)

   a

   M = i=0 j =0

   Nt

   (3)

   Fig. 1. Images of Brodatz Database[11]

3. METHODOLOGY

   1. Perceptual Textural Features

      The perceptual textural features used are coarseness, contrast, directionality and busyness.

      (1)Coarseness

      Coarseness is a measure of granularity of the texture. A coarse texture is composed of large primitives and is

      Where Ma represents the average amplitude.M(i,j) is the amplitude of pixel (i,j), Ntthe number of pixels having an amplitude superior to threshold t.

      (4 )Busyness

      Busyness refers to the intensity changes from a pixel to its neighborhood. Busyness is a reverse relationship of coarseness.

      Bs = 1 1/ (4)

      Where Cs represents the computational measure coarseness (1/ is a quantity used to make Cs significant).

   2. Mixtures Of Generalized Gaussian Distribution

   The general Gaussian Distributions for a univariant random variable X R is defined in its form

   3

   characterized by a high degree of local uniformity of grey- levels. Coarseness is saved in this auto correlation function.

   p x

   , , =

   1

   exp(A()

   x

   ) (5)

   For coarse textures it presents few local variations

   2 1

   Coarseness Cs is expressed by equation:

   where = [

   ]/

   and represents mean and standard deviation. Parameter controls the pdf and determines whether the distribtion is peaked or flat. The larger the value of

   ,the flatter the pdf, and the smaller is, the more peaked the pdf is around its mean.

   The message length that encodes the wavelet coefficients in a given subband is given by,

   + + +

   1. Time, precision, and recall are taken for performance analysis.

4. EXPERIMENTAL RESULTS

1. Performance Metrics

   The metrics evaluatedfor performance evaluation are:

   1. Time.

      (6)

      Where, , , denote the prior distribution of the

      parameters .

      1. The Mean (k) is obtained by,

      2. Precision.

      3. Recall.

      Time

      Time is always important to do a process in minimum amount of time. Time also plays a vital role in the performance evaluation.

      Time, here calculated, includes training time, retrieval time, and the time to calculate the distance

      n k 2

      = i =1 p xi

      (7)

      measure.

      k n

      k 2

      i =1 p xi

      Precision

      Precision is defined as the ratio of the number of

      2. The Standard Deviation (k) is obtained by,

      retrieved relevant images to the total number of retrieved images. Precision P measures the accuracy of the retrieval.

      n p(k )

      = ( ) i=1 xi

      (8)

      xi

      n

      i =1

      p(k )

      Precision=

      ._ _ _ _

      _ _ _ _ _

      3. The shape parameter (k ) is obtained by,

      Recall

      Recall is the number of relevant and retrieved

      p x x

      k = k (2 log (log(p( )/k (9)

      k

      2

      IV.THE SEQUENCE OF IMPLEMENTATION:

      1. Brodatz database are of 11 classes of 8 images each. So totally 88 images.

      2. Each database has separate coding part for test and train images.

      3. First the images are undergone pre-processing stage where Perceptual model and Wavelet modelling are calculated.

      4. Features are extracted from Perceptualmodel and MOGG.

      5. They are converted into double format.

      6. The features are extracted from query image given by the user.

      7. The distance between query image and trained images are calculated using the distance measures.

      8. The distance are stored in a one-dimensional matrix.

      9. First three minimum values are extracted and regarding picture is obtained from trained database.

      images divided by the number of relevant images in the database for the considered query, measures the ability of a model to retrieve all relevant images.

      Recall= _ _ _ _

      _ _ _ _ _ _

2. Performance Evaluation

The level of retrieval accuracy achieved by a system is important to establish its performance. If the outcome is satisfactory and promising, it can be used as a standard in future research works.

The Performance analysis shown in the tables and graphs, gives the result thatMOGG is the best method by comparing both methods with the time value, precision and recall.

Time (secs)

Table 1: Computing time (secs) with cosine distance measure in perceptual method and KLD in MOGG model.

30

25

20

15

10

5

0

Perceptual

Method

MOGG

Image Image Image

3 5 6

Brodatz database image

Fig 2: Retrieval time values from the image 3, image 5, image 6 in Perceptual method and MOGG model with cosine and KLD distance measure.

Table 2: Computing precision (%) with cosine distance measure in perceptual method and KLD in MOGG model.

Precision

Percentage

150

100

Table 3: Computing recall (%) with cosine distance measure in perceptual method and KLD in MOGG model.

40

30

20

10

0

Recall

Perception

Model

MOGG

Img 3Img 5Img 7

Brodatz Images

Percentage

Fig 4:Recall values for the image 3, image 5, image 6 in Perceptual method and MOGG model with cosine and KLD distance measure.

VI. CONCLUSION

In this paper, Texture based image retrieval is done and the feature extraction was done by using Perceptual model and MOGG. The distance measures used are cosine and KLD. The perceptual features extracted are Coarseness, Contrast, Directionality and Busyness.we have maintained two stuff that is Training and Testing phase.In Training phase, We trained all the images in the Database by the process of Feature Extraction.In Testing phase, the Feature of the selected current image was extracted and compared with feature database.Finally, we got therelevant images from the database by the process of feature matching.The images

are grouped into 11 classes of 8 images each class for a

50

0

Img Img Img 3 5 7

Perception model

MOGG

total of 88 images.

The analysis result shows thattime,precision and recall values of MOGG is faster than Perceptual Model.

Brodatz images

Fig 3: Precision values for the image 3, image 5, image 6 in Perceptual method and MOGG model with cosine and KLD distance measure.

REFERENCES

1. Abbadeni. N,D. Ziou and S.Wang, (2000) Auto covariance-based perceptual textural features corresponding to human visual perception, in proc.15th IAPR/IEEE Int. Conf. Pattern Recognit., Barcelona, Spain, vol.pp.901-904t.

2. Lin, C.H., Chen, R.T., Chan, Y.K.: A smart content-based image retrieval system based on color and texture feature, Image Vis.Comput., 2009, 27, (6), pp. 658665.

3. Liu, G.H., Yang, J.Y.: Image retrieval based on the texton co- occurrence matrix, Pattern Recognit., 2008, 41, (12), pp. 3521 3527.

4. Do.M.NandM.Vetterli,(2002) Wavelet-based texture retrieval using generalized Gaussian density and Kullback-Leibler distance, IEEE Trans.Image Process., vol.11, no.2, pp.146-158.

5. Allili. M. S, (2010) Wavelet- based texture retrieval using a Mixture of Generalised Gaussian Distributions. In Proc. IEEE Inf. Conf. pattern Recog., pp. 3143-3146.

6. Manjuth. B. S and Ma. W. Y, Texture features for browsing and retrieval of image data,(1996), IEEE Trans. Pattern Anal. Mach. Intell., vol. 18, special Issue on digital libraries, no. 8, pp. 837-842.

7. Huang, P.W., Dai, S.K.: Image retrieval by texture similarity, Pattern Recognit., 2003, 36, (3), pp. 665679.

8. Liu.F and R.W. Picard (1996)Periodicity, Directionality and Randomness: Wold Features for Image representation and Retrieval IEEE Transactions on Pami, Vol 18,Nov 7.

9. Tamura,Hideyuki ; Mori,shunji ; Yamawaki, Takashi(1978) Textural Features Corresponding toVisual PerceptionSystems, Man and Cybernetics, IEEE Transactions on Volume:8 ,Issue: 6 .

10. Liu. X and Wang.D (2003),Texture classification using spectral histograms, IEEE Trans. Image Process, vol.12, no.6, pp.661-670.

11. G. Verdoolaege and P. Scheunders, Geodesics on themanifold ofmultivariategeneralized Gaussian distributions with an application to multicomponenttexture discrimination, Int. J. Comput. Vis., vol. 95, no.3, pp. 265286, Dec. 2011.

12. N. Abbadeni, Computational perceptual features for texture representation and retrieval, IEEE Trans. Image Process., vol. 20, no. 1, pp.236246, Jan. 2011.