Multi-index hashing with repeat-bits in hamming space

Hashing approaches play an essential role in large-scale visual recognition and classification. It maps the high-dimensional image descriptors to hamming space, and the image can be represented as compact binary codes which can enable huge efficiency in storage and computation cost. There is little attention drawn to explore effective query and indexing mechanism for the hashed hamming space. A recently proposed method generates multi-index hashing tables based on substring of binary codes to address this issue. However, it performs well only in the non-uniformly distributed data set. By analyzing the performance of classic algorithm, we found that query speed is more important than the space cost in some applications. In this paper, we propose a new and faster query algorithm in hamming space, even if it take more storage space.