Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

19
Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram

Transcript of Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Page 1: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Content Based Image Retrieval using Normalized Similarity Measure in

Fuzzy Color Histogram

Page 2: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Authors

Alina Banerjee

Department of Computer Science and Engineering

Birla Institute of Technology, Mesra, Kolkata Campus

Kolkata, India

e-mail: [email protected]

Ambar Dutta

Department of Computer Science and Engineering

Birla Institute of Technology, Mesra, Kolkata Campus

Kolkata, India

e-mail: [email protected]

Page 3: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Research Problem

• Fuzzy linked histogram - a very popular approach of linking three separate color channel histograms according to some fuzzy rules into one histogram

• Fuzzy linked histogram and its bucket applied histogram is quite similar

• From this distribution a normalized similarity measure is calculated according to which the top k similar images are displayed

Page 4: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Fuzzy Linked Histogram

• The linking of 3 histograms to 1 dimension is known as linking.

• Linking method of the 3-dimensional histogram to 1-dimension based on fuzzy approach was proposed [8] in 2004 .

• Color histogram creation proposed based on the L*a*b* color space which provided a histogram with only 10 bins.

• Linking RGB color space to L*a*b* color space and then converting into a single histogram using a fuzzy expert system [8]

Page 5: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Fuzzy Linked Histogram

• a* is subdivided into five regions representing green, greenish, the middle of the component, reddish and red .

• b* is subdivided into five regions representing blue, bluish, the middle of the component, yellowish and yellow .

• L* is subdivided into only three regions: black, grey and white areas .

Page 6: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Membership functions of L*, a* and b*.

Page 7: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Fuzzy Linked Histogram

• The Mamdani type of fuzzy inference is used.• The output of the system has only 10 equally

divided trapezoidal membership functions.• The defuzzification phase is performed using the

lom (largest of maximum) .• Fuzzy linking of the three components is made

according to 27 fuzzy rules which leads to the output of the system [8].

Page 8: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

27 Fuzzy Rules1. If (L is Black) and (a is amiddle) and (b is bmiddle) then (fuzzy_histogram is black) (1) 2. If (L is white) and (a is amiddle) and (b is bmiddle) then (fuzzy_histogram is white) (1) 3. If (L is grey) and (a is red) and (b is yellow) then (fuzzy_histogram is red) (1) 4. If (a is reddish) and (b is yellow) then (fuzzy_histogram is brown) (1) 5. If (L is white) and (a is green) and (b is yellow) then (fuzzy_histogram is green) (1) 6. If (L is white) and (a is green) and (b is yellowish) then (fuzzy_histogram is green) (1) 7. If (L is Black) and (b is Blue) then (fuzzy_histogram is blue) (1) 8. If (L is white) and (a is green) and (b is Bluish) then (fuzzy_histogram is cyan) (1) 9. If (L is grey) and (a is amiddle) and (b is bmiddle) then (fuzzy_histogram is darkgrey) (1) 10. If (a is greenish) and (b is Bluish) then (fuzzy_histogram is blue) (1) 11. If (a is red) and (b is Bluish) then (fuzzy_histogram is blue) (1) 12. If (L is white) and (b is yellow) then (fuzzy_histogram is yellow) (1) 13. If (L is Black) and (a is reddish) and (b is Bluish) then (fuzzy_histogram is blue) (1) 14. If (a is red) and (b is Blue) then (fuzzy_histogram is blue) (1) 15. If (L is grey) and (a is red) and (b is yellowish) then (fuzzy_histogram is red) (1) 16. If (L is white) and (a is reddish) and (b is yellowish) then (fuzzy_histogram is yellow) (1)17. If (L is Black) and (a is reddish) and (b is yellowish) then (fuzzy_histogram is red) (1) 18. If (a is reddish) and (b is yellow) then (fuzzy_histogram is yellow) (1) 19. If (L is Black) and (b is Bluish) then (fuzzy_histogram is blue) (1) 20. If (L is grey) and (b is Blue) then (fuzzy_histogram is blue) (1) 21. If (L is grey) and (a is reddish) and (b is Bluish) then (fuzzy_histogram is magenta) (1) 22. If (L is grey) and (a is amiddle) and (b is Bluish) then (fuzzy_histogram is cyan) (1) 23. If (L is grey) and (a is amiddle) and (b is yellowish) then (fuzzy_histogram is brown) (1) 24. If (L is white) and (a is amiddle) and (b is yellowish) then (fuzzy_histogram is yellow) (1)25. If (L is grey) and (a is red) and (b is bmiddle) then (fuzzy_histogram is red) (1) 26. If (L is grey) and (a is reddish) and (b is bmiddle) then (fuzzy_histogram is red) (1) 27. If (L is white) and (a is reddish) and (b is Bluish) then (fuzzy_histogram is magenta) (1)

Page 9: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Fuzzy Linked Histogram(Membership function of the output of the fuzzy system)

Page 10: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Creation Of Split Histogram From Fuzzy Linked Histogram

Fig. 1. Sample image

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10

Bins

No

. o

f p

ixel

s

Fig. 2. Fuzzy Linked Histogram

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10

Bins

No

. o

f p

ixel

s

Fig 3. Fuzzy linked histogram after bucket application Fig 4. Split Histogram

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10Bins

No

. o

f p

ixel

s

Coherent Pixels Incoherent Pixels

Page 11: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Normalized Similarity Measure• When comparing two images with color coherence vectors

(j, j) and (j, j), for j = 1... J, the following feature can be used:

• For this feature (above equation.) the differences for coherence pairs (0, 1), (0,100) and (4500, 4501), (4500, 4600) are equivalent. So to remove this equality a normalized similarity measure (NSM) was used as follows

J

jjjjjH

1

J

j jj

jj

jj

jjNSM

1 11

Page 12: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Algorithm1. A query image is selected.2. Its fuzzy linked image and hence its histogram is obtained.3. The CCV of the obtained fuzzy linked image is calculated

using n discrete color buckets.4. For all images stored in the database, repeat steps 5 to 8.5. An image from database is opened.6. Its fuzzy linked histogram is obtained.7. The CCV with the same n buckets of the obtained fuzzy linked

image is calculated.8. The similarity measure (NSM) is calculated between the query

image and the image in the database under consideration.9. Based on the values of the similarity measure, the top k similar

images are displayed [3]. • In the proposed method the value of n is taken as 10

since the fuzzy linked histogram has 10 bins.

Page 13: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Experimental Results• Test Images

The experimentation of the proposed approach is done on an image database comprising of more than 1100 color images which are taken from [15, 16]. The database is classified into 21 classes with more than 50 images in each class. The representative images of different classes of the image database used for experimentation purpose are given

Page 14: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Experimental Results

• Observations:

Fig. 5 Average precision (%) versus number of retrieved images Fig. 6 Average recall (%) versus number of retrieved images

Page 15: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Experimental Results

• Observations:

Fig. 7 Crossover Point of Precision and Recall [18]

Method Crossover Point

Proposed Method

0.582

Konstantinidis et. al.

0.570

Suhasini et. al.

0.513

Page 16: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Conclusion & Future Scope

• A novel approach to CBIR proposed where a normalized similarity measure is derived from split histogram created from the fuzzy linked histogram

• Technique tested upon a heterogeneous image database comprising of more than 1100 color images

• Outputs (average precision. average recall, precision-recall crossover point) proves that it gives better performance than some well established methods of CBIR

• It could further be improved so as to achieve higher precision and recall values.

• It can be thought to incorporate it in content based video retrieval with some modification

Page 17: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

References  

1. Goodrum.A., Rorvig.M, Jeong.K, Suresh, C. "An Open Source Agenda for Research Linking Text and Image Content Features," Journal of the American Society for Information Science, 52(11), 948-953, 2001

2. Shandilya. S.K., Singhai.N, “A Survey On: Content Based mage Retrieval Systems”. International Journal of Computer Applications, 4 (2), 22-26, July 2010

3. Smith. J.R, “Integrated Spatial and Feature Image Systems: Retrieval, Analysis and Compression”, PhD Thesis, Graduate School of Arts and Sciences, Columbia University, 1997

4. Zhang. R, Zhang. Z,” Addressing CBIR efficiency, effectiveness, and retrieval subjectivity simultaneously” , MIR '03 Proceedings of the 5th ACM SIGMM international workshop on Multimedia information retrieval , Berkeley, CA, USA, 71-78, 2003

5. Kulkarni. S, “Interpretation of Fuzzy Logic For Texture Queries in CBIR”, Proceedings Vision, Video, and Graphics, VVG 2003, University of Bath, UK, 175-179, 2003

6. Lin. H, Chiu. C, Yang. S, “Finding textures by textual descriptions, visual examples and relevance feedbacks”. Pattern Recognition Letters 24(2003), 2255-2267, 2003

7. Younes. A, Truck. I, Akdag. H, “Color Image Profiling using Fuzzy Sets”, Turkish Journal Of Electrical Engineering & Computer Sciences, 13(3), 343-360, 2005

8. Konstantinidis. K., Gasteratos. A., Andreadis. I., “Image retrieval based on fuzzy color histogram processing”, Optics Communications 248, 375– 386, 2005

9. Pass.G, Zabih.R, and Miller.J, “Comparing Images Using Color Coherence Vectors”, Proceedings of the fourth ACM international conference on Multimedia, New York, USA, 65-73, 1996

Page 18: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

References  

10. Suhasini. P.S., Sri Rama Krishna. K., Murali Krishna. I.V., “CBIR Using Color Histogram Processing”, Journal of Theoretical and Applied Information Technology, 6(1), 116-122, 2009

11. Swain. M and Ballard. D, “Color indexing”, International Journal of Computer Vision, 7(1), 11-32, 1991.

12. Shamir. L, “Human Perception-based Color Segmentation Using Fuzzy Logic”, Proceedings of the 2006 International Conference on Image Processing, Computer Vision and Pattern Recognition, Las Vegas Nevada, USA, 2, 496-505, 2006

13. Chatzichristofis. S and Boutalis. Y, “FCTH: Fuzzy Color and Texture Histogram A Low Level Feature for Accurate Image Retrieval ”, Ninth International Workshop on Image Analysis for Multimedia Interactive Services, 191-196, 2008

14. Bourjandi. M, “Image Retrieval Based on Eten Fuzzy Color Histogram and place location”, 5th Symposium on Advances in Science and Technology, Mashhad, Iran, 2011

15. http://www.cs.washington.edu/research/imagedatabase/groundtruth

16. http://-utopia.duth.gr/~konkonst

17. www.intelligence.tuc.gr/~petrakis/courses/multimedia/retrieval.pdf

18. Kekre. H.B, Thepade. S. D, Banura. V.K, “Amelioration of Colour Averaging Based Image Retrieval Techniques using Even and Odd parts of Images”, International Journal of Engineering Science and Technology (IJEST), 2(9), 4238-4246, 2010

Page 19: Content Based Image Retrieval using Normalized Similarity Measure in Fuzzy Color Histogram.

Thank You!