Binarising SIFT-Descriptors to Reduce the Curse of Dimensionality in Histogram-Based Object Recognition

It is shown that distance computations between SIFT-descriptors using the Euclidean distance suffer from the curse of dimensionality. The search for exact matches is less affectedthan the generalisation of image patterns, e.g. by clustering methods. Experimental results indicate that for the case of generalisation, the Hamming distance on binarised SIFTdescriptorsis a much better choice. It is shown that the binary feature representation is visually plausible, numerically stable and information preserving. In an histogram-based object recognition system, the binary representation allows for the quick matching, compact storage and fast training of a code-book of features. A time-consuming clustering of the input data is redundant.