SIFT always produce same size descriptor vector which is 128 bit long therefore, matching is always done for the same size descriptor vectors.

But there is 5 feature vectors (keypoint, keypoint descriptor, keypoint magnitude, keypoints orientation and key point location) I thought that we should fusionne this 5 features not only keypoint descriptor ps: i m new in SIFT

There is no need to use all 5 parameters. As Nadia mentioned, you should just matched descriptors with each other. Nearest neighbor method can be used to find matched keypoints.

You can take a look at how matching is done in OpenCV:

http://docs.opencv.org/doc/tutorials/features2d/feature_flann_matcher/feature_flann_matcher.html

The same in python:

http://docs.opencv.org/trunk/doc/py_tutorials/py_feature2d/py_matcher/py_matcher.html

Basically a fast approximation of nearest neighbors is used in both cases. All the best.

Depending on what are you intending to do with SIFT, you might be interested in using aggregated methods (generating a single descriptor for the entire image), as BoW, BossaNova, VLAD, FisherVectors, and other.

(http://www.codeproject.com/Articles/619039/Bag-of-Features-Descriptor-on-SIFT-Features-with-O)

http://webia.lip6.fr/~cord/Actu_files/cord.pdf

If you want to match specific objects inside your image you probably would like to use approximate nearest neighbors search, as it was pointed out by Hugues Talbot.

Use Vlfeat to understand SIFT

http://www.vlfeat.org/overview/sift.html

Thanks everybody, i read the article of david low in his website http://www.cs.ubc.ca/~lowe/keypoints/ and I understood the basis of this method website