Image Retrieval Based on Image Mean Mapping Method R.Venkata Ramana Chary, Associate Professor Padmasri Dr.B.V Raju Institute of Technology, Hyderabad,India
[email protected]
Dr. D. Rajya Lakshmi , Professor and HOD Department of CSE, JNTUK, University Collegeof Engineering, Vizianagaram
[email protected]
Dr. K.V.N Sunitha, Professor & Principal BVRIT HYDERABAD College of Engineering for Women ,Hyderabad,
[email protected]
ABSTRACT Image searching is often done in the world of computer visualization. Time and again users get to see many new digital images uploaded on web. They are fascinated towards automatic image retrieval. In the past few years, many methods were introduced to retrieve images which were based on similarity of size and shape of the image. Content Based Image Retrieval (CBIR) method created the need for proficient and intelligent schemes for classifying and retrieval of Images. One of the main advantages of CBIR is automatic retrieval process in place of the keyword-based approach. The CBIR technology has been used in several applications such as biodiversity information systems, medicine, digital libraries, crime investigation and Historical research. In this method, mean-mapping techniques have been used for the retrieval of images. Each image is converted into the gray form. The database contains threshold values and cluster mean values of more then 10000 images which are extracted using k-mean retrieval method. The query image is compared with the database image using mean mapping methods. The main aim of this work is to extract images with high-class similarity. Keywords CBIR, Gray image, feature Values, K-means clustering, mean values, Euclidian distance, threshold values, query, performance evaluation. 1 INTRODUCTION 1.1. Clustering Clustering or Cluster analysis is the task of grouping various objects. These groups are called clusters. The objects in a cluster or a group have many similarities among them. Clustering is a technique commonly used for the analysis of statistical data, mainly used in image analysis and pattern recognition. 1.2 K-Means Clustering K-means is the clustering algorithm used to determine the natural spectral grouping present in a data set. This accepts from analyst the number of clusters to be located in the data. The algorithm then arbitrarily seeds or locates, that number of cluster centers in multidimensional measurement space. Each pixel in the image is then assigned to the cluster whose arbitrary mean vector is closest. we can calculate the mean of points in each segments of an image. The K-means algorithm is an iterative technique that is used to partition an image into K clusters. The basic algorithm is: 1.
Select K cluster centers, either randomly or based on some heuristic
c 978-1-4799-6629-5/14/$31.00 2014 IEEE
377
2.
3. 4.
Assign each pixel in the image to the cluster that minimizes the distance between the pixel and the cluster center Re-compute the cluster centers by averaging all of the pixels in the cluster Repeat above steps 2 and 3 until convergence is attained.
2 Proposed Method ¾
The proposed system mainly concentrates on similarities in shape and size of the image.
¾
It implements various feature extraction methods which are very helpful in retrieval of images.
¾
Using user query method, similar images are retrieved. In this process, the cluster mean values of the image are extracted and further analysis is done.
Table 1 : Processing of the Method 3. Proposed Method (PM) and Implementation The proposed algorithm is summarized as following : Step1: The query image is taken. Step 2: It is converted from color to gray form. Step 3: The images in the database are grouped according to the threshold values and their cluster mean values are calculated using vector form. Step 4: The cluster wise average is calculated and the difference between that average point and image is evaluated. Step 5: Then the root of all obtained averages is taken. .Step 6: The similarity measures between the query image and retrieved images are assessed. Step 7: Results and performance is evaluated using precision and recall 4. Results Analysis: Images with similarities are retrieved using threshold values and analyzed. Table 2 and table 3 represent 60 such images. C1-C6 actual cluster mean values are tabulated in table 2 whereas table 3 gives us the d1-d6 average point values to database image distances. It can be observed that images 9823.jpg, 9818.jpg and 9822.jpg which have 4.15, 4.58 and 4.63 distances from the average point respectively have many similarities. This can be visualized in table 4. Image 9769.jpg 9823.jpg
378
C1 12 17
C2 52 59
C3 86 95
C4 120 127
C5 157 164
C6 208 217
9818.jpg 9822.jpg 9892.jpg 9895.jpg 9766.jpg 9765.jpg 9819.jpg 9820.jpg 9891.jpg 9888.jpg 9772.jpg 9821.jpg 9883.jpg 9770.jpg 9894.jpg 9771.jpg 9773.jpg 9825.jpg 9826.jpg 9893.jpg 9767.jpg 9792.jpg 9808.jpg 9824.jpg 9785.jpg 9768.jpg 9890.jpg 9896.jpg 9793.jpg 9885.jpg 9889.jpg 9781.jpg 9787.jpg 9786.jpg 9782.jpg 9805.jpg 9790.jpg 9886.jpg 9887.jpg 9804.jpg 9783.jpg 9775.jpg 9788.jpg
25 16 13 13 13 13 27 30 17 17 8 30 14 9 13 10 8 9 9 14 16 14 24 33 17 16 12 14 35 12 13 10 17 18 17 25 15 13 12 24 22 14 16
65 58 53 50 55 56 64 68 63 63 47 69 67 56 58 60 41 41 42 48 61 49 56 71 62 62 46 58 72 53 51 40 59 60 52 67 55 55 54 71 64 70 53
97 95 87 85 91 95 94 99 101 101 85 100 102 93 94 100 78 75 75 84 98 83 89 99 95 100 79 102 104 86 83 80 96 99 92 99 96 89 87 108 98 108 92
130 128 119 120 127 131 122 131 134 134 123 130 131 130 132 139 117 107 107 124 133 113 116 129 122 136 111 140 132 115 114 122 129 130 136 131 126 117 115 142 129 150 129
170 165 151 163 167 169 156 169 165 165 159 166 157 160 170 171 155 143 141 168 172 145 158 171 175 175 141 174 159 145 143 176 162 166 176 185 164 147 145 191 183 180 169
220 217 201 222 221 220 209 225 211 211 205 222 193 194 222 205 195 198 195 226 224 189 230 229 221 227 185 211 202 181 181 234 238 240 227 221 240 177 175 231 235 210 245
9802.jpg
15
51
77
113
135
226
9794.jpg 9884.jpg 9774.jpg
22 15 7
50 64 37
83 94 67
105 119 101
155 142 129
249 171 158
2014 International Conference on Contemporary Computing and Informatics (IC3I)
9806.jpg 21 66 105 141 192 9784.jpg 17 62 97 125 186 9789.jpg 16 63 122 146 167 9801.jpg 19 62 84 115 133 9807.jpg 20 62 102 136 200 9803.jpg 13 59 81 110 124 9795.jpg 14 49 70 99 116 9791.jpg 29 71 120 181 216 9796.jpg 12 55 75 104 116 9797.jpg 13 47 68 97 114 Average Point 16.67 57.28 91.95 125 160 Table 2: Cluster Mean Values And Average Abs (avgc1 -d1) 9769 .jpg 9823 .jpg 9818 .jpg 9822 .jpg 9892 .jpg 9895 .jpg 9766 .jpg 9765 .jpg 9819 .jpg 9820 .jpg 9891 .jpg 9888 .jpg 9772 .jpg 9821 .jpg 9883 .jpg 9770 .jpg 9894 .jpg 9771 .jpg 9773 .jpg
4.59 0.41 8.41 0.59 3.59 3.59 3.59 3.59 10.41 13.41 0.41 0.41 8.59 13.41 2.59 7.59 3.59 6.59 8.59
d2
d3
d4
d5
d6
av g
5.1 8 1.8 2 7.8 2 0.8 2 4.1 8 7.1 8 2.1 8 1.1 8 6.8 2 10. 82 5.8 2 5.8 2 10. 18 11. 82 9.8 2 1.1 8 0.8 2 2.8 2 16. 18
5.8 4 3.1 6 5.1 6 3.1 6 4.8 4 6.8 4 0.8 4 3.1 6 2.1 6 7.1 6 9.1 6 9.1 6 6.8 4 8.1 6 10. 16 1.1 6 2.1 6 8.1 6 13. 84
4.9 5 2.0 5 5.0 5 3.0 5 5.9 5 4.9 5 2.0 5 6.0 5 2.9 5 6.0 5 9.0 5 9.0 5 1.9 5 5.0 5 6.0 5 5.0 5 7.0 5 14. 05 7.9 5
3.6 5 3.3 5 9.3 5 4.3 5 9.6 5 2.3 5 6.3 5 8.3 5 4.6 5 8.3 5 4.3 5 4.3 5 1.6 5 5.3 5 3.6 5 0.6 5 9.3 5 10. 35 5.6 5
0.0 8 8.9 2 11. 92 8.9 2 7.0 8 13. 92 12. 92 11. 92 0.9 2 16. 92 2.9 2 2.9 2 3.0 8 13. 92 15. 08 14. 08 13. 92 3.0 8 13. 08
4.0 5 3.2 8 7.9 5 3.4 8 5.8 8 6.4 7 4.6 5 5.7 1 4.6 5 10. 45 5.2 8 5.2 8 5.3 8 9.6 2 7.8 9 4.9 5 6.1 5 7.5 1 10. 88
238 230 237 155 240 140 133 251 130 129 208
srt avg 2.89 4.15 4.58 4.63 4.87 6.12 6.54 6.71 7.02 7.10 7.42 7.42 7.88 7.97 8.18 8.25 8.48 8.80 8.88
9825 .jpg 9826 .jpg 9893 .jpg 9767 .jpg 9792 .jpg 9808 .jpg 9824 .jpg 9785 .jpg 9768 .jpg 9890 .jpg 9896 .jpg 9793 .jpg 9885 .jpg 9889 .jpg 9781 .jpg 9787 .jpg 9786 .jpg 9782 .jpg 9805 .jpg 9790 .jpg 9886 .jpg 9887 .jpg 9804 .jpg 9783 .jpg 9775 .jpg 9788 .jpg 9802 .jpg 9794 .jpg 9884
7.59 7.59 2.59 0.59 2.59 7.41 16.41 0.41 0.59 4.59 2.59 18.41 4.59 3.59 6.59 0.41 1.41 0.41 8.41 1.59 3.59 4.59 7.41 5.41 2.59 0.59 1.59 5.41 1.59
16. 18 15. 18 9.1 8 3.8 2 8.1 8 1.1 8 13. 82 4.8 2 4.8 2 11. 18 0.8 2 14. 82 4.1 8 6.1 8 17. 18 1.8 2 2.8 2 5.1 8 9.8 2 2.1 8 2.1 8 3.1 8 13. 82 6.8 2 12. 82 4.1 8 6.1 8 7.1 8 6.8
16. 84 16. 84 7.8 4 6.1 6 8.8 4 2.8 4 7.1 6 3.1 6 8.1 6 12. 84 10. 16 12. 16 5.8 4 8.8 4 11. 84 4.1 6 7.1 6 0.1 6 7.1 6 4.1 6 2.8 4 4.8 4 16. 16 6.1 6 16. 16 0.1 6 14. 84 8.8 4 2.1
17. 95 17. 95 0.9 5 8.0 5 11. 95 8.9 5 4.0 5 2.9 5 11. 05 13. 95 15. 05 7.0 5 9.9 5 10. 95 2.9 5 4.0 5 5.0 5 11. 05 6.0 5 1.0 5 7.9 5 9.9 5 17. 05 4.0 5 25. 05 4.0 5 11. 95 19. 95 5.9
17. 65 19. 65 7.3 5 11. 35 15. 65 2.6 5 10. 35 14. 35 14. 35 19. 65 13. 35 1.6 5 15. 65 17. 65 15. 35 1.3 5 5.3 5 15. 35 24. 35 3.3 5 13. 65 15. 65 30. 35 22. 35 19. 35 8.3 5 25. 65 5.6 5 18.
2014 International Conference on Contemporary Computing and Informatics (IC3I)
10. 08 13. 08 17. 92 15. 92 19. 08 21. 92 20. 92 12. 92 18. 92 23. 08 2.9 2 6.0 8 27. 08 27. 08 25. 92 29. 92 31. 92 18. 92 12. 92 31. 92 31. 08 33. 08 22. 92 26. 92 1.9 2 36. 92 17. 92 40. 92 37.
14. 38 15. 05 7.6 4 7.6 5 11. 05 7.4 9 12. 12 6.4 3 9.6 5 14. 22 7.4 8 10. 03 11. 22 12. 38 13. 30 6.9 5 8.9 5 8.5 1 11. 45 7.3 7 10. 22 11. 88 17. 95 11. 95 12. 98 9.0 4 13. 02 14. 66 12.
9.05 9.44 9.50 9.59 10.8 4 11.0 3 11.2 7 11.4 6 12.0 8 12.1 0 13.0 6 13.3 9 13.6 2 13.9 3 14.1 5 14.3 0 14.6 2 14.7 6 15.0 4 15.0 8 15.8 9 16.4 7 17.0 3 17.6 2 18.0 2 18.1 3 18.5 9 19.9 8 20.4
379
.jpg 9774 .jpg 9806 .jpg 9784 .jpg 9789 .jpg 9801 .jpg 9807 .jpg 9803
9.59 4.41 0.41 0.59 2.41 3.41 3.59
2 20. 18 8.8 2 4.8 2 5.8 2 4.8 2 4.8 2 1.8
6 24. 84 13. 16 5.1 6 30. 16 7.8 4 10. 16 10.
5 23. 95 16. 05 0.0 5 21. 05 9.9 5 11. 05 14.
65 31. 65 31. 35 25. 35 6.3 5 27. 65 39. 35 36.
08 50. 08 29. 92 21. 92 28. 92 53. 08 31. 92 68.
04 26. 72 17. 28 9.6 2 15. 48 17. 63 16. 78 22.
8 21.5 1 21.9 2 22.1 8 25.9 8 29.3 3 30.7 6 39.2
.jpg 2 84 95 65 08 66 0 9795 8.1 21. 25. 44. 75. 29. 42.9 2.59 .jpg 8 84 95 65 08 72 2 9791 13. 28. 56. 55. 42. 34. 43.2 12.41 .jpg 82 16 05 35 92 78 5 9796 2.1 16. 20. 44. 78. 27. 45.5 4.59 .jpg 8 84 95 65 08 88 4 9797 10. 23. 27. 46. 79. 31. 45.5 0.60 .jpg 18 84 95 65 08 38 9 Table 3:Deference From Average Image Mean To Other Images
9806.jpg
9784.jpg
9823.jpg
9818.jpg
9822.jpg
9892.jpg
9895.jpg
9766.jpg
9765.jpg
9819.jpg
9820.jpg
9891.jpg
9888.jpg
9772.jpg
9821.jpg
9883.jpg
9770.jpg
9894.jpg
9771.jpg
9773.jpg
9825.jpg
9826.jpg
9893.jpg
9767.jpg
9792.jpg
9808.jpg
9824.jpg
9785.jpg
9768.jpg
9890.jpg
9896.jpg
9793.jpg
9885.jpg
9889.jpg
9781.jpg
9787.jpg
9786.jpg
9782.jpg
9805.jpg
9790.jpg
9886.jpg
9887.jpg
9804.jpg
9783.jpg
9775.jpg
9788.jpg
9802.jpg
9794.jpg
9884.jpg
9774.jpg
9789.jpg
9801.jpg
9807.jpg
9803.jpg
9795.jpg
9791.jpg
9796.jpg
Table 4: Query Results Based On Proposed Method
380
2014 International Conference on Contemporary Computing and Informatics (IC3I)
9797.jpg
Table 4 represents the results obtained based on meanmapping method. Through this table it can be observed that many similar images are retrieved together. One of the observations is that 9822.jpg, 9823.jpg and 9818.jpg are 3 out of 9 similar images retrieved in top 5. Here the percentage of retrieval is 34. 6 similar images are retrieved in top 15 and percentage of retrieval is 67. It can also be observed that all 9 images are retrieved in top 30. This shows 100% retrieval. The same can be represented graphically in table 5.
image 9782.jpg are retrieved in top5. Similarly 3 images similar to 9288.jpg were retrieved in top 10. In the same way 3 images relevant to 9787.jpg were retrieved in top 5 out of 6. This table is represented in graphical form below.
Table 7: Image Retrievals percentages Observations of Image 9784.jpg, 9822.jpg, 9787.jpg and 9801.jpg
Table 5: Image Retrievals Observations of Image 9822.jpg
9822. jpg 9784. jpg 9787. jpg 9801. jpg
to p5
top 10
top 15
top 20
top 25
top 30
top 35
top 40
tot al
3
4
6
6
8
9
9
9
9
1
3
4
5
6
9
9
11
11
3
3
4
5
6
6
6
6
6
3
5
5
5
5
5
5
5
5
Table 7 represents the percentages of retrieval of various image groups. From the table it can be observed that 50% retrieval is seen in the case of big horses in top 5. This percentage is 60 in the case of small horses in top 5 and 100 in top 10. 5. Performance Measurements Precision : It is the ratio of the number of relevant images retrieved to the total number of images retrieved. recall. It is the ratio of the number of relevant images retrieved to the total number of relevant images in the database. Relevant Not Relevant Retrieved
Table 6: Number of relevant images retrieved
Not Retrieved
Correct
Unexpected
Result(Tp)
Result (Fp)
Missed
Correct Absence
Result(Fn)
Of Result
Table 6 represents the number of relevant images retrieved for each image. It can be observed that 3 images relevant to
2014 International Conference on Contemporary Computing and Informatics (IC3I)
381
retrieving relevant images. It is one of the most efficient methods when compared to the existing ones. References 1.
P-10
R-10
P-20
R-20
P-30
R-30
9822
0.40
0.44
0.30
0.67
0.30
1.00
9784
0.30
0.27
0.25
0.45
0.30
1.00
9787 9801
0.30 0.50
0.50 1.00
0.25 0.25
0.83 1.00
0.20 0.17
1.00 1.00
2.
3.
4.
5.
Table 8: Precision and Recall Results 6.
In table 8 according to the results obtained precision and recall values are evaluated and compared for four images . 6. Conclusion In this paper we presented an approach for Image Retrieval using mean mapping method. K-Mean clustering techniques have also been used, for clustering the images into groups having similar threshold values. Image feature values are obtained from the images and stored into the database. Using mean mapping method, the image feature values are compared. This method can be used for efficient and fast image retrieval. From the above results we can conclude that mean-mapping method is very useful in
382
Rui Xu, Donald Wunsch II, “Survey of Clustering Algorithms”, IEEE Transactions On Neural Networks, Vol. 16, No. 3, May 2005. Grundhofer, A.; Bimber, O.; , "Real-Time Adaptive Radiometric Compensation," Visualization and Computer Graphics, IEEE Transactions on , vol.14, no.1, pp.97-108, Jan.-Feb. 2008 doi: 10.1109/TVCG.2007.1052
7.
ia Wan; Kuo, C.-C.J.; , "A new approach to image retrieval with hierarchical color clustering," Circuits and Systems for Video Technology, IEEE Transactions on , vol.8, no.5, pp.628-643, Sep 1998 doi: 10.1109/76. “Similar Image Searching From Image Database Using Cluster Mean Sorting and Performance Estimation”. International conference on Machine Vision and Image Processing (MVIP12) is organized by Department of Instrumentation and Control Systems Engineering, PSG College of Technology on December 14 & 15, 2012. Proceedings of Conference published in to IEEE Explor International Journal. 978-1-4673-2320-8 “Edge connecting points based image searching and Measurements”, IEEE International Conference on Computational Intelligence and Computing Research (ICCIC-2012).18th, 19th and 20th of December, 2012. Tamilnadu College of Engineering, Coimbatore 641659, India. Proceedings of Conference published in to IEEE Explor International Journal. CFP1220JPRT, ISBN: 978-1-4673-2481-6/12IEEE., “Image Searching Based on Image Mean Distance Method” International Conference on RADAR, Communication and Computing in Computer Science and Engineering (ICRCC-12). December 21-22, 2012. S K P Engineering College, Thiruvannamalai, Tamilnadu, India. Proceedings of Conference published in to IEEE Explor International Journal:ISBN:978-81-925286-5-6 IEEE , “Unsupervised methods on image database using cluster mean average Methods for image searching” Fourth Workshop on Applications of Graph Theory in Wireless Ad hoc Networks and Sensor Networks (GRAPH-HOC – 2012, NETCOM 2012.) .(21-23 DEC 2012). DOI :10.1007/978 -1-4614-6154-8_75 .(NetCom lecture notes in Electrical Engineering 131 ,Springer science +Business Media New York 2013)
2014 International Conference on Contemporary Computing and Informatics (IC3I)