What similarity measure should we use?

[pinakinathc]

I was confused regarding what similarity measure should we use?
Few of the options are:

• Hausdroff distance
• Hu-moments as given in this Link
• Any other methods you can propose

[kousikdasgupta]

Dear Pinaki, Choosing similarity measure depends on the type of work you are doing. In other words the similarity measure depends on the objects(domains) whose similarity you are finding. Also the more measures you use more good is your quality of work. Use both now and evaluate your performance, if required we would find some more.

…

On Wed, Jan 10, 2018 at 11:07 PM, Pinaki Nath Chowdhury < ***@***.***> wrote: I was confused regarding what similarity measure should we use? Few of the options are: -[ ] Hausdroff distance -[ ] Hu-moments as given in this Link <http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_contours/py_contours_more_functions/py_contours_more_functions.html#match-shapes> -[ ] Any other methods you can propose — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF5E7x83rZrsMLB5kftR22H6f7Je5A5gks5tJPVFgaJpZM4RZsDG> .

-- Dr. Kousik Dasgupta, Assistant Professor, Department of Computer Science and Engineering, Kalyani Government Engineering College, Kalyani, Nadia, West Bengal, India-741235.

[pinakinathc]

@kousikdasgupta thanks sir. yes, it would be better to use both and compare results.
Both the above methods are invariant under transformations like:

• translation
• scaling
• rotation
  Whatever method we choose (like if the members come up with new dissimilarity measures, they must be invariant to the above-mentioned transformations).

As a start I would want everyone to please go through the following links and try to understand as much as possible:

• Shape Matching (or rather similarity b/w shapes) using Hu-Moments
• using t-SNE for dimensionality reduction
• Some preprocessing of our input data(input to t-SNE and other algorithms) i.e. making mean 0 and scaling the data

Please note learning these methods are important and also google how to create a pull request and submit your changes, as I shall use Git as the main collaboration method.