Ask Your Question

Is there any other (faster) technique to retrieve pairwise images in a huge dataset?

asked 2017-06-14 11:39:49 -0500

swiss_knight gravatar image

I was wondering if it exists any other technique as the one which consist of pairwise check on 2 images descriptors proximity to find, in a really huge set of images, which are the pairs which share a portion of the same area or objects?

Because finding matches need to compute every possible combinations, 2 by 2, which dramatically increase the computation time when datasets are really big.

I'd like to find a faster way (at this point, I don't really need to check for precise matching, I just have to be sure that 2 images may share some same information to make a fast clustering).

Any idea (even outside openCV) ? Thanks.

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted

answered 2017-06-14 19:49:56 -0500

Tetragramm gravatar image

The basic thing is to be smart about your storage, and choose your descriptor wisely. Structures like the k-d tree show methods for searching high-dimensional data quickly. Your choice of descriptor matters as well, the better you can compress the image into a descriptor the better, because smaller descriptors are faster to check. Of course, the speed of calculating the descriptor matters too.

This is a very active area of research, typically with the keywords "image retrieval". Search Google Scholar with those keywords and begin reading. You can limit it to newer papers to get the best, but be sure to look at the citations to find good papers from the past that explain the basics.

edit flag offensive delete link more

Question Tools

1 follower


Asked: 2017-06-14 11:39:49 -0500

Seen: 188 times

Last updated: Jun 14 '17