0

I am trying to rewrite the mutual information (MI) similarity measure between two images could support cluster environment (i.e. Hadoop) based on MapReduce mechanism.

MI similarity measure contains these processes:

  1. Histogram for each image.
  2. Joint histogram for both images
  3. Entropy or probability for each histogram bin.
  4. MI equation

How to distribute the previous MI processes on Map and Reduce?

Mohammad nagdawi
  • 553
  • 4
  • 18
  • Compute MI on an image pair on a single node, there is no sense in parallelizing such a trivial operation. Spread your mages over the nodes. – Cris Luengo Feb 11 '18 at 16:45
  • 1
    I mean to say: the cost of communications is going to be worse that the cost of actual computations. Even for large images. If your images are truly humongous (whole slide images, satellite imagery), try computing the joint histogram in parallel, the rest is irrespective of image size. – Cris Luengo Feb 11 '18 at 16:46

0 Answers0