A review of feature indexing methods for fast approximate nearest neighbor search

Fast feature matching is of crucial importance for time-critical applications in computer vision. The main goal of this work is to provide a comprehensive review of the state-of-the-art approaches dealing with the problem of feature indexing. Crucially, indexing methods can be grouped into four classes, including space partitioning, clustering, hashing, and product quantization. The methods are deeply presented, discussed, and linked to each other. An empirical report of performance analysis is also provided to characterize the studied methods. Lastly, we give comments on possible room of improvements for some indexing schemes.