Discriminant Embedding for Local Image Descriptors

Gang Hua, Matthew Brown, and Simon Winder


Invariant feature descriptors such as SIFT and GLOH

have been demonstrated to be very robust for image matching

and visual recognition. However, such descriptors are

generally parameterised in very high dimensional spaces

e.g. 128 dimensions in the case of SIFT. This limits the

performance of feature matching techniques in terms of

speed and scalability. Furthermore, these descriptors have

traditionally been carefully hand crafted by manually tuning

many parameters. In this paper, we tackle both of

these problems by formulating descriptor design as a nonparametric

dimensionality reduction problem. In contrast

to previous approaches that use only the global statistics

of the inputs, we adopt a discriminative approach. Starting

from a large training set of labelled match/non-match

pairs, we pursue lower dimensional embeddings that are

optimised for their discriminative power. Extensive comparative

experiments demonstrate that we can exceed the

performance of the current state of the art techniques such

as SIFT with far fewer dimensions, and with virtually no

parameters to be tuned by hand.


Publication typeInproceedings
Published inInternational Conference on Computer Vision
> Publications > Discriminant Embedding for Local Image Descriptors