• No results found

Hybrid Swarm Intelligence Method for Post Clustering Content Based Image Retrieval

N/A
N/A
Protected

Academic year: 2021

Share "Hybrid Swarm Intelligence Method for Post Clustering Content Based Image Retrieval"

Copied!
7
0
0

Loading.... (view fulltext now)

Full text

(1)

Procedia Computer Science 79 ( 2016 ) 509 – 515

Available online at www.sciencedirect.com

1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Organizing Committee of ICCCV 2016 doi: 10.1016/j.procs.2016.03.065

ScienceDirect

7th International Conference on Communication, Computing and Virtualization 2016

Hybrid Swarm Intelligence Method for Post Clustering

Content Based Image Retrieval

Shubhangi P. Meshram

a

, Dr. Anuradha D. Thakare

b

,Prof. Santwana Gudadhe

c

aStudent,Department of Computer Engineering, Pimpri chinchwad college of engineering, Pune, 411044, India b

Professor,Department of Computer Engineering, Pimpri chinchwad college of engineering, Pune, 411044, India

c

Professor,Department of Computer Engineering, Pimpri chinchwad college of engineering, Pune, 411044, India

Abstract

Content Based Image Retrieval is one of the most promising method for image retrieval where searching and retrieving images from large scale image database is a critical task. In Content Based Image Retrieval many visual feature like color, shape, and texture are extracted in order to match query image with stored database images. Matching the query image with each image of large scale database results in large number of disc scans which in turns slows down the systems performance.

The proposed work suggested an approach for post clustering Content Based Image Retrieval, in which the database images are clustered into optimized clusters for further retrieval process. Various clustering algorithms are implemented and results are compared. Among all, it is found that hybrid ACPSO algorithm performs better over basic algorithms like k-means, ACO, PSO etc. Hybrid ACPSO has the capability to produce good cluster initialization and form global clustering.

This paper discusses work-in-progress where we have implemented till clustering module and intermediate results are produced. These resulted clusters will further be used for effective Content Based Image Retrieval.

Keywords:Content Based Image Retrieval(CBIR);Image Clustering;Ant Colony Optimization(ACO);Particle Swarm Optimization(PSO);

Hybrid ACPSO.

1. Introduction

Content-Based Image Retrieval (CBIR) systems retrieve images similar to a user-defined specification or pattern (e.g., shape sketch, image example). Lots of attracting works has been done in database of art work which is grabbing the attention of users from the field of geography, medicine, architecture, advertising, design, fashion, and publishing [10]. The importance of Content Based Image Retrieval is act in a specific way by increasing the aim for retrieving images from abundentely large Database over the Internet.

Data clustering is often took as a step for speeding up image retrieval and improving accuracy especially in large database. Clustering is a process of separating a dataset into groups in such a way that the object in one group is

© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

(2)

more similar to those objects in the other group. A cluster is a collection of objects, which are similar in some pattern to each other and are dissimilar to the objects in other clusters.

In CBIR, feature vector are extracted from images. Query feature vector are matches with the stored feature vector on one to one basis. This results in slow down the processing time.

To improve the speed of executing and better result many researchers are paying attention to uses the clustering algorithm for CBIR.

Clustering is a process of separating a dataset into groups in such a way that the object in one group is more similar to those objects in the other group [4].

CBIR with the clustering algorithms gives the optimize result for image clustering. Images clusters are formed based on the features extracted from images such as color, shape and texture.

x CBIR using colour features: In this techniques colors are extracted from the image. And extracted colours

are considered as a feature vector. This color feature of images can be extracted in many ways like Histogram , Color moments , Color Correlogram , color coherence vector etc.[2]

x CBIR using texture feature: In this techniques image texture are used as a feature vector.

x CBIR using shape feature: In this techniques edges are used as a feature vector. Robert, Sobel, Prewitt,

Kirsch, Robinson, Marr-Hildreth, LoG and Canny,this are diferent kind of Edge Detection techniques which are use for feature vector [2].

CBIR with clustering algorithm is proven to be a better option for getting the optimum results. Number of clustering algorithm is available. K-means is the basic clustering algorithm but in case k-means is very sensitive for initial clustering and sensitive to outliers and noise [4, 7]. Also to form the clusters, K-mean depends on initial condition, which causes the algorithm to give suboptimal solution. As compare to K-means Ant colony algorithm are more prominent for initializing the cluster. Due to the global optima nature of Particle swarm optimization algorithm give the optimum solution for clustering. These most prominent features of Ant colony algorithm and particle swarm optimization are used for clustering.

This paper presents the Content based image retrieval using Ant colony and Particle swarm Optimization clustering algorithm. The proposed algorithms are hybridization of Ant Colony and Particle Swarm Algorithm for Optimization in Image Clustering.

The paper is organized as follows: Section 2 presents proposed system for optimal data clustering. Section 3 presents the experimental results and final conclusion is given in section 4.

2. Proposed System

Proposed work is based on combination of CBIR and Data Clustering techniques. This uses the CBIR technique and hybrid ACPSO algorithm for searching and fast retrieval of Images.

In Proposed system, images are stored in database. Feature vector are generated from images based on shape and these feature vector are stored in the database. Using clustering algorithm images are grouped and clusters are formed. Input query image are passed as an input and depending on similarity it will assign to the cluster.

Proposed system shown in Fig. 1 consists of the following step.

Image Database- Image database collected from CORAL database [9]. Images are colour images and already pre-processed.

(3)

Feature Extraction- In this system, shape features are considered. To extract the shape feature sobel edge detection method is used. Initially image gradient is calculated in X direction and Y direction and finally gradient magnitude is calculated using the formula shown below [2]. After obtaining the image gradient, gradient of an image are represented in the energy coefficient format by using transformation. And finally mean of an energy coefficient in obtained to form the feature vector. The reduced set of feature vector represents the image matrix.

(1)

Where, Gx is the image gradient in x direction whereas Gy is the gradient in y direction. G is the gradient magnitude

of an image.

Image Clustering- After obtaining the feature vectors, all feature vectors are stored in database. For grouping the image based on similarity ACPSO clustering algorithm is used. Ant colony algorithm is used to initialize the centroid of cluster. And PSO is used for optimum clustering.

2.1 ACO Algorithm for initialization of centroid of cluster[1]

The Features of Images are extracted through CBIR and dataset is given as input to the ACO algorithm the data items are randomly scattered into a two-dimensional grid.

Initially, all the Data items which are extracted through the edge detection technique are distributed over the 2D space in a grid, Each ant move randomly around the grid with picking and dropping the Data items. The decision is random to pick up or drop an item but is influenced by the data items in the ant‟s immediate neighborhood. The probability of dropping an item is increased if ants are surrounded with similar data in the neighborhood. In contrast, the probability of picking an item is increased if a data item is surrounded by dissimilar data, or when there is no data in its neighborhood [3]. to obtain the centroid of the cluster, ant are randomly placed. Multiobjective function is used to obtain the cost. The Multiobjective function is given below [1]. In next iteration ant position get reform.

­

D

E

°

T ij.Kij if c  N (s

p

) p k ®¦ c N (sp)TD.KE ij (2) ij

°

il ij il °0 otherwise ,

¯

Based on the calculation, every ants position are simplified for every repetition which will give a new result until it reaches n repetition. The final iteration gives the centroid set which is given for the PSO algorithm.

2.2. PSO Algorithm for Optimal Clustering

The initial set of centroids obtained from ACO algorithm is then passes to the particle solution in PSO algorithm [5, 6, 8]. The „p‟ solution which is good in terms of cost estimation is given as a first population of particles for PSO clustering. Here, each solution (particles) is a centroid set of the input data. So, the population scope of the PSO clustering is p X (k*d) matrix and the velocity cost of each particle is initialized to zero. Then, for each outcome, fitness is computed based on multi objective functions i.e. the objective of minimization or maximization for correct clustering [1].

(4)

2.3. Performance Measures:

The accuracy of the various clustering algorithms for proposed work is obtained by using F-Measure.

F-Measure: The precision and recall this combination is use to measures the accuracy of similar data objects in a given cluster and all objects of that class. The F-measure [1] of cluster i with respect to class j is

F(i,j) = (2 × precision(i, j) × recall(i, j))/(precision(i, j) + recall(i, j)) (3)

The overall F-measure is calculated as:

(4) Where the maximum is taken over all clusters i at all levels, mj is the number of objects in class j, and m is the total number of objects. measure (FM) [1] is a measure of the quality of a solution given the true clustering. For F-measure, the optimum score is 1.

(5)

3. Experimental Results and Discussion

Platform - The proposed CBIR with ACPSO clustering is implemented with a system of having i3 processor and main memory of 2 GB RAM using MATLAB (R2012b).

Test Bed- The Image dataset collected from CORAL database [9]. The test results are generated on 100 images. In test bed images are considered. Sample test images are shown in fig.2. The test database consists of variable images of flower, butterfly and Nature.

3.1 Results

The CBIR with ACPSO clustering algorithm is implemented. The objective functions are applied for cluster optimization. The combination of CBIR and hybrid ACPSO gives the best result for the image Retrieval from large database.

For proposed system’s performance is evaluated using k-mean, ACO, PSO and ACPSO algorithm. The numbers of iteration considered are 100. Results are obtained for cluster size 4. K-means algorithm results in accuracy of 0.91. ACO and PSO algorithms results in accuracy of o.88 and 0.96 respectively. The hybrid ACPSO algorithm, results in improved accuracy of 0.98.

For proposed CBIR method using ACPSO algorithm, the performance is checked by using a single objective function and multiple objective functions. The F-measure values are as shown in Table I. shows the accuracy of algorithm

(6)

Table 1. Result of Data Clustering Algorithm.

Clustering algorithm analysis

Sr. No. Clustering Algorithm Number of Iterations Accuracy of

No. of Clusters Algorithm

1 K-means 100 4 0.91

2 ACO 100 4 0.88

3 PSO 100 4 0.96

4 ACPSO 100 4 0.98

The graphical representation of accuracy obtained by various algorithms is as shown in Fig 3.

Fig.3. Analysis of clustering algorithm in terms of Accuracy

4. Conclusion

This paper presents the hybrid swam intelligence method for post clustering content based image retrieval system. To obtain the optimized clusters, the hybrid Ant Colony Optimization and Particle Swarm Optimization algorithm is applied.

In proposed system, shape feature are extracted from images and feature vector are formed using the energy coefficient of edge images. Hybrid Ant Colony and Particle Swarm Optimization algorithm is used to cluster the similar images.

For feature extraction, the minimum distance between features of same image class is obtained. However the distance is maximum for images of different classes. This approach of feature extraction has improved the performance. Also, all irrelevant and redundant features are dropped by ant colony optimization and Particle swarm optimization which selects the most relevant features among complete feature set

(7)

The accuracy of proposed system is obtained for k-means, Ant colony optimization (ACO), Particle swarm optimization (PSO) and finally hybrid form of ant colony optimization and particle swam optimization algorithm (ACPSO).

From the experimental results, it is observed that ACPSO works better as compare to the other algorithm. ACPSO gives maximum score (accuracy) as 0.98 as compared to K-means and ACO.

References

1. Deepali Kharche and Anuradha thakare ,”An Improved ACPSO method for Data Clustering With Multiple Oblective Function” ,IEEE 2015.

2. Kalyan Roy, Joydeep Mukherjee, "Image Similarity Measure using Color Histogram,Color Coherence Vector, and Sobel Method," Acm Computing Surveys, International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064.

3. O.A. Mohamed Jafar and R. Sivakumar,” Ant-based Clustering Algorithms: A Brief Survey”, International Journal of Computer Theory and Engineering, Vol. 2, No. 5, October, 2010 1793-8201.

4. Juli Rejito, Retantyo Wardoyo, Sri Hartati, Agus Harjoko,” Optimization CBIR using K-Means Clustering for Image Database “,International Journal of Computer Science and Information Technologies,Vol. 3 (4) , 2012,4789 4793

5. Li-Yeh Chuang, Yu-Da Lin, and Cheng-Hong Yang, Member, IAENG,” An Improved Particle Swarm Optimization for Data Clustering”,proceeding of the International muticonference of Enginneres and scientist 2012 VOl I,IMECS 2012,March 14-16,2012. 6. Cheng-Lung Huang, Wen-Chen Huang, Hung-Yi Chang, Yi-Chun Yeh, Cheng-Yi Tsai,” Hybridization strategies for continuous ant

colony optimization and particle swarm optimization applied to data clustering”, 2013.

7. Taher Niknam a, Babak Amiri,” An efficient hybrid approach based on PSO, ACO and k-means for cluster analysis”, 2009 Elsevier. 8. Mr. Pankaj K. Bharne, Mr. V. S. Gulhane, Miss. Shweta K. Yewale,” Data Clustering Algorithms B ased On Swarm Intelligence”,

2011 IEEE.

9. https://sites.google.com/site/dctresearch/Home/content-based-image-retrieval Last referred on 10 september 2015.

10. Mr. Kondekar V. H., Mr. Kolkure V. S., Prof.Kore S.N.,” Image Retrieval Techniques based on Image Features: A State of Art approach for CBIR”, (IJCSIS) International Journal of Computer Science and Information Security,Vol. 7, No. 1, 2010.

Figure

Fig. 2. Block diagram of the proposed ACPSO algorithm for post clustering of CBIR.
Fig. 2. Sample Dataset
Table 1. Result of Data Clustering Algorithm.

References

Related documents

Fluopyram decreased nematode eggs per gram of root when applied to soil containing plants that were inoculated with either Meloidogyne incognita or Heterodera glycines.. The type

we use and analyze the MQTT protocol and take four scenarios and evaluate the proposed design by using OMNET++ simulation environment and its INET frame work, from the

The garbage monitor system proposed by ShashankShukla[2Jose M.Gutierrez et al [3] in their proposed work understood the smart waste collection monitor system using the

Objective To determine whether ascertainment of childhood obesity by surveillance followed by structured secondary prevention in primary care improved outcomes in overweight or

Background: As a public safety regulator, the Technical Standards and Safety Authority (TSSA) of Ontario, Canada predicts and measures the burden of injuries and fatalities as

An anomalous diffusion model based on a new general fractional operator with the Mittag Leffler function of Wiman type Liang et al Advances in Difference Equations (2018) 2018 25

Adaptive tracking control for a class of uncertain nonlinear systems with infinite number of actuator failures using neural networks Lv and Wang Advances in Difference Equations (2017)