Shape Based Image Retrieval

Hello, I'm working on CBIR, I've implemented color and texture based search. Both are not efficient, and thus I implemented Entropy search too for better result. Even though system is not efficient.

Now I'm trying Shape, where in algorithm it is mentioned that Based on the extracted shape features, image classification process has been performed using Support Vector Machine (SVM) tool. there is an inbuilt function for canny edge detection, what features does this outcome have? and how classification is done using SVM? please help me, please suggest me the relevant codes.

Answer 1

[Anand]

Are you talking about this paper?

"Content Based Image Retrieval Using Color, Texture and Shape Features", Hiremath, Pujari

If you are, the formulas for calculating the shape features are in there on page 32.

If you want to use Support Vector Machines, you can look at this page:

Support Vector Machines in MATLAB

[Chethan]

Well, thank you. In this paper at last we get a total of 15 features (shape descriptors). After that distance between 2 images is computed using D=D1+D2+D3. I'm not getting where to use those 15 features further. Also, it is mentioned that D1 & D2 are distance computed by integrated matching scheme at 2 resolutions and D3 is distance resulting from shape comparison. What are D1,D2,D3 ? how to compute those?

[Anand]

Part 2 titled system overview and proposed methods describes the integrated matching scheme as well as how to get D1 and D2. You use the Canberra distance defined at the end of the shape feature description to compute the similarities.

Answer 2

[Image Analyst]

See this video: http://www.youtube.com/watch?v=g7mOpkWqQUQ The whole thing is interesting, but for you, especially the part starting at 2:55. Though there are probably lots of other methods and I have no idea if that one is the best for you. I just thought it looked pretty interesting.

[Chethan]

Well thank you, may i have your e-mail id please.