Revision history [back]

• BRISK / FREAK / ORB were all intended for a fast matching for tracking and not for an image classification problem. SIFT is a strong descriptor but it is definitely slower than the mentioned ones, see http://computer-vision-talks.com/2012/08/a-battle-of-three-descriptors-surf-freak-and-brisk/ and http://computer-vision-talks.com/2011/08/feature-descriptor-comparison-report/ .
• For a classification using a large enough database of images, using FLANN will always be faster than with BruteForce. For binary features LSH is fine, you can also try Flann's hierarchical-clustering approach ( HierarchicalClusteringIndexParams ) of Muja and Lowe: "Fast Matching of Binary Features", see http://people.cs.ubc.ca/~mariusm/papers/binary_matching_crv2012.pdf
• For an image classification problem you shouldn't classify upon single matches. More common is a Bag-of-Words approach or other classification approaches, see http://answers.opencv.org/question/8677/image-comparison-with-a-database#8686 for more details on this topic.

If you still want to follow your (absolutely valid but woudln't be my first choice) approach, you can prune your matches with cross-check, ratio-test, or by geometry, all are explained here: http://answers.opencv.org/question/11840/false-positive-in-object-detection/