How to select hashing bits? A direct measurement approach

Hashing, which encodes data points into binary codes, has become a popular approach for approximate nearest neighbor searching tasks because of its storage and retrieval efficiency. Given a pool of bits, we propose to select a set of bits according to a quality measurement directly related to the large scale approximate nearest neighbor searching problem. An alternating greedy optimization method is proposed to find a locally optimal solution. The experimental results show this optimization method is efficient and effective.