opencv - Comparing feature descriptors of large set of images -

I have a set of thousands of images, and for each image I have drawn a set of SIFT feature descriptor (current Is binding in 200 per picture).

I need to create a whole graph of the distance between each picture. I.e., I need to work through every metric through every metric, through every second image of the distance.

So far, I have tried to use FLANN to calculate 20 neighboring descriptors between two nodes, and then the average distance between all calculations matched, unfortunately, this process is executed It's taking too much time to do.

Is there any way to compare these images more efficiently to the detail?

You can use your SIFT decoder by the Think-Vis-Visual Word (BV) or locally consolidated descriptor ( VLAD) can be considered for collecting in vector. Originally:

1 - Calculate a textbook (K SIFT descriptor) such as - K-Earth

2 - For each image, remove the SIF descriptor, then each See the closest neighbor in the codebook, so calculate the histogram of the image shift according to the codebook. This is the easiest way (hard coding, even pooling), but there are alternatives (and often give better results for computer vision problems) <3> - Therefore, the size of each image is shown with a unique peculiar K histogram) . After this, you can calculate the distance between the images between the histogram (example Euclidean).