I am working on image processing. I have to apply k-means upon them. But I am confused with the initialization of k-means that either I should use just first frame or all the frames to initialize it. And each frame is row vector with distinct number of rows. Kindly clear my concept

  • $\begingroup$ It is fast, so I have always used random initialization. One might do well to use an adaptive variation on Poisson disc initialization. It likely depends on how naively the data is stored. $\endgroup$ Dec 18, 2020 at 5:24
  • $\begingroup$ Okay. But I am concerned about the data to be used for initialization of kmeans. As I mentioned, I am working on image processing and I have to pass frames value (comprised of three attributes) to my kmeans. And I am asking that should I use first frames values for kmean initialization or all the frames values. $\endgroup$
    – TariqS
    Dec 19, 2020 at 7:45
  • $\begingroup$ If you are using K means in the images, how are you contriving it? Are you contriving the locations of the pixels as XY? What are the names of your three columns? Is your data discreet as in eight distinct colors, or is it continuous on the domain between zero and one? There are several useful axes to consider here. Also, what are you gonna do when you’re done? Are you using an information criteria to determine the correct number of clusters? $\endgroup$ Dec 19, 2020 at 17:24
  • $\begingroup$ These three attribute give me background information, optical flow and histogram of optical flow. The data is continuous. Yes I am using information criteria. $\endgroup$
    – TariqS
    Dec 20, 2020 at 11:00


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.