Content Based Image Retrieval Using Color Moments and Texture

Content Based Image Retrieval Using Color Moments and Texture

AbstractThis paper proposes an algorithm to retrieve the color images from the database using color moments and the texture of color image. The color moments integrates the mean, standard deviation and the skewness to estimate the brightness and the intensity of the image. The texture descriptors are calculated using the energy and entropy values from the curvelet coefficients. Finally, similarity measure is computed and compared with all the images present in the database to classify the feature vectors. Experimental results show that the proposed algorithm is more accurate and efficient in retrieving the images with the user interested objects. Evaluation metrics used are Precision and Time taken to retrieval.

KeywordsColor moments, Texture, Curvelet Transform, Similarity measure, Energy and Entropy values and Precision.

. INTRODUCTION

Digital images require a large amount of data to be stored and the users are able to find images quickly and accurately from the large number of images .In early years, image retrieval was done by the textual annotators using the keywords. Here, increasing the number of digital images stored is not feasible. So there are some limitations in using the textual annotators. To overcome this problem, the image retrieval process (1) can be taken by the Content Based Image Retrieval System (CBIR)(2).

The objective of the content based image retrieval system is to extract the features and can classify the images for retrieving the similar images related to the input query image. CBIR system is widely used in many modern technologies where the large amount of images is to be used.

This system utilizes the features like color, texture(3), shape and spatial relationships to extract the relevant images from the database(4). Color feature is not only closely related to objects in an image, but also dependent on size of an image, orientations and the angle (5). Color feature has

the robustness and the common representations of color features are color histograms, color moments and dominant color etc.,(6) these features are compared to similarity measure using the Euclidean distance measurement (7).This categorization of images can be processed by many techniques such as SVM (Support Vector Machine) kernel(8), Decision tree. There are some limitations in feature extraction (9), so the curvelet transform technique integrates the texture classification (10) effectively by evaluating the energy and entropy values as described later (11).Discrete curvelet transform (12) produces accurate results than the existing one.

. DIFFERENT CBIR METHODS

In a typical retrieval system, the query image and the images stored in the database are pre-processed automatically to extract the features using the curvelet. The query image is also processed same as the database images. The feature vectors are used to classify the images. And relevant images are retrieved based on the similarity measures from the database. In the image retrieval system, there are many approaches used to retrieve the feature based images.

There are different CBIR methods used for determining the features like color and texture. Color feature is evaluated using the primitives of the color moments using the known RGB color model.

Many methods have been proposed to extract the texture features either directly or from the image statistics, for example co-occurrence matrix or spatial frequency domain. The different methods used for feature extraction are: Markov Random Fields parameters, Gabor multichannel features, fractal-based features and co-occurrence features.

. PROPOSED METHOD

The proposed algorithm is modification of existing method Sasikumar et al method [1], which uses Discrete Sine Transform (DST). The developed algorithm is more efficient in comparison with [1]. Fig.1 shows the block diagram of an image retrieval system using the Curvelet

Transform. The color moments and Texture of an image are estimated as follows.

A function of the similarity between two image distributions is defined as the sum of the weighted

differences between the moments of the two distributions. Formally this is:

Feature similarity measure matching

Feature extraction using curvelet

, =

1 2 +

1 2

Query image

=1

1

2

Retrieved images

+ 3 1 2 (4)

Database feature extraction

Database

Figure. 1.Image retrieval mechanism using Curvelet Transform Technique.

Where,

(H,I) are the two image distributions being compared.

i is the current channel index

r is the number of channels

1 , 2 are the first moments (mean) of the two image

1. Color moments

   Color moments are calculated to estimate the brightness and the intensity of the images. The color moments used in this algorithm to extract the images on the

   distributions.

   1, 2 are the second moments(standard deviation) of the two image distributions

   1 , 2 are the second moments (skewness) of the two image

   color basis are mean, standard deviation and Skewness.

   Mean can be understood as the average color value in the image. Square root of the variance can be defined as the standard deviation. Skewness can be understood as a measure of the degree of asymmetry in the distribution.

   An image is characterized by 9moments- 3moments for each 3 color channels.

   It defines the ith color channel at the jth Image pixel as pij. The three color moments can then be defined as:

   MOMENT 1 – Mean:

   =1

   1

   distributions

   wi are the weights for each moment.

   Pairs of images can be ranked based on their dm values.

   Those with greater values are ranked lower

   and considered less similar than those with a higher rank and lower dm values.

2. Texture feature extraction

   Texture features are determined using the curvelet coefficients. The features are the key points for comparing the query image with the images loaded in the database.

   From the co-occurrence matrix is obtained from curvelet coefficients, which is used to determine the texture

   =

   MOMENT 2 -Standard Deviation:

   =1

   (1)

   features such as energy and entropy are determined.

   Entropy measures the randomness of a gray-level distribution. The entropy is considered to be high if the gray levels are distributed randomly throughout the image. Energy measures the number of repeated pairs. The energy is considered to be high if the occurrence of repeated pixel pairs is high.

   =

   ( 1

   ( )2) (2)

   Texture features are calculated from each Curvelet Sub-band. In order to improve the classification gain, co- occurrence matrix is formed for each sub-band of discrete

   The standard deviation is the square root of the variance of the distribution.

   MOMENT 3 Skewness The degree of asymmetry in the distribution can be defined as the kewness.

   curvelet transform. From the co-occurrence matrix, the texture features such as energy and entropy are calculated.

   1. Entropy

      =1

      3 1

      = (

      ( )3) (3)

      = [, ] [, ] (5)

      =1 =1

      Here p[i,j] are the pixel values at the (i,j) coordinates of the image. The size of the image is MXN.

      4. Euclidean distance is measured for similarity measurement of query image with database images.

   2. Energy

   V.ANALYSIS OF RESULTS

   A.Retrieval of Images

   The developed algorithm is tested for various test

   = 2 , (6)

   =1 =1

   Here P[i,j]are the pixel values at the (i,j)Coordinates of the image. The size of the image is MXN.

3. Similarity Measure

   Similarity measure can be computed using Euclidean distance which is estimated using color moments and Texture features of an image. The measured Euclidean distance of a query image is compared with the database images results in bringing out the similar images.

   The retrieval efficiency can be rated by using the precision and the elapsed time. Precision is defined as the ratio of number of relevant images retrieved to the total number of images retrieved. By using these color moments, texture features and the distance measure the retrieval performance can be improved.

   Similarity metrics can be computed from the relevant curvelet coefficients described as

   2 1

   images.The experimental results are tabulated as shown in the Table 1.Database consists of different colour and texture based images.From the Table 1 it is observed that the developed algorithm is more efficient and accurate over [1]

   .Computational complexity is reduced which can be observed from Table1.

   TABLE 1: PERFORMANCE RESULTS OF PROPOSED ALGORITHM

   2

   1

   , = ( ( ) )2

   =0

   (7)

   Where Q={Q0,Q2,.Q2n-1}is the feature vector of the query image and T={T0,T2,.T2n-1} is the feature vector of target image in the database. The images with the

   distance d which is the closest similarity are arranged in ascending order for retrieving the image. The efficiency of the retrieval system can be measured by precision and elapsed time as follows.

   Test images	Evaluation metrics	Method
   		Proposed algarithm	Sasikumar et.al
   Image1	Precision	98.4%	96.7%
   	Elapsed time	0.080s	0.283s
   Image2	Precision	98.4%	96.7%
   	Elapsed time	0.074s	0.253s
   Image3	precision	98.4%	96.7%
   	Elapsed time	0.068s	0.284s
   Image4	precision	98.4%	96.7%
   	Elapsed time	0.074s	0.297s
   Image5	precision	98.4%	96.7%
   	Elapsed time	0.078s	0.258s

   Retrieval of images by the given test images are as shown below in fig 2&3.

   Test Images:

   Precision = No .of relevant images retrieved

   Total no .of images retrieved

   (8)

   Elapsed time is the time taken to process the image retrieval of the images.

4. Algorithm of proposed method

   The proposed algorithm is explained in 4 steps as follows.

   1. The query image of size MXN is chosen.

   2. Color moments are calculated using mean, standard deviation and the skewness of an image.

   3. Texture feature is estimated using Entropy and Energy values from the curvelet coefficients.

Figure2:Output images after retrieval of similar images.

Figure.3.Output images after retrieval of similar images.

Here, this represents the retrieval of images using the discrete curvelet transform with the similarity measurement. The images are retrieved by evaluating the color moments and texture results. The similarity metrics of the proposed algorithm is compared with the Sasi kumar et al method. Hence the result shows that the image retrieval system using curvelet transform is efficient than the existing one.

V. CONCLUSIONS

The proposed curvelet algorithm is used for feature extraction in the image retrieval system. The objective of this approach is to improve the precision values for the feature representation.The performance is measured interms of precision which is defined as the ratio of number of relevant images retrieved to the total number of images retrieved.

By observing the characteristics of the image retrieval methods, the efficiency of the image retrieval system by using the curvelet is better than Discrete Sine Transform and the elapsed time taken to retrieve the images is less when compared with the existing algorithm.

REFERENCES

1. Sasi kumar. M, Dr. Y. S.Kumaraswamy, An improved support vector machine kernel for an image retrieval system, Proceedings of the International Conference on Pattern Recognition, Informatics and Medical Engineering. March, 2012, pp. 257-260.

2. Rajakumar and Muttan, A Framework for MRI image retrieval using curvelet transform and euclidean distance, Journal of Computer Science, Vol 9(3),2013, pp: 285-290.

3. Haralick R.M, Statistical and structural approaches to texture, Proceedings of the IEEE, Vol. 67, 1979, pp. 786-804.

4. J.L. Starck, E. Candes, and D.L. Donoho, The Curvelet Transform for Image Denoising , IEEE Transactions on Image Processing , Vol. 11, No.6, 2002, pp.670 -684.

5. Lucia Dettori and Lindsay Semler, A comparison of wavelet, ridgelet and curvelet-based texture classification algorithm in computed tomography, Computers in biology and medicine, Vol.37, 2007, pp.486-498.

6. Unser M., and Eden M, Multi-resolution feature extraction and selection for texture segmentation, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 11, No.7,1989, pp. 717-728.

7. Wang, J. and X. Zhao, A new approach for image retrieval with integrated Euclidean distance and rotational correlation. Proceedings of the IEEE International Conference on Wireless Communications, Networking and Information Security, Jun. 25-27, 2010, pp: 391-395.

8. Sasi Kumar. M, Kumaraswamy. Y. S, Medical Image Matching and Retrieval using Discrete Sine Transform, International Journal on Computer Science and Engineering, Vol. 2, No. 9, 2010,pp.2880- 2882.

9. Schumacher, P. Jun Zhang, "Texture classification using neural networks and discrete wavelet transform," Image Processing Proceeding, IEEE International Conference, vol.3, 13-16 Nov 1994, pp. 903-907

10. Soo Beom Park, Jae Won Lee, Sang Kyoon Kim, Content-based image classification using a neural network, Pattern Recognition Letters, vol. 25, 2004, pp. 287300

11. Tuceryan M. and Jain A. K, Texture Segmentation Using Voronoi Polygons, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 12, 1990, pp. 211-216.

12. S.Prabha, Dr. M.Sasikala, Texture classification using curvelet transform technique, International Journal of Advancements in Research & Technology, Volume 2, Issue4,April 2013,pp.249-253.