Efficient and compact visual feature descriptors hashing using hierarchical multiple assignment k-means

In this paper we present an efficient and effective method for visual descriptors hashing based on hierarchical multiple assignment within a k-means framework. The method has been used to address the problem of approximate nearest neighbor (ANN) retrieval, and it has been tested on local and global visual content descriptors, either engineered or learned. The proposed method has been compared to state-of-the-art methods on different standard large-scale datasets composed by millions of visual features: SIFT 1M and GIST 1M (BIGANN), and also on the recent DEEP1M dataset, composed by one million CNN-based features. Experimental results show that, despite its simplicity, the proposed method obtains an excellent performance.