SIFT-based object recognition with fast alphabet creation and reduced curse of dimensionality

This article presents a SIFT-based object recognition method that avoids the typical problems of time-consuming code-book generation and curse of dimensionality in feature comparison. The first problem is solved by using an alphabet of completely synthetic feature vectors. The second problem is solved by using the Hamming-distance on binarised SIFT-features. The method is supported by competitive results on the Graz'02 image data base.