Hashing has been proven a promising technique for fast nearest neighbor search over massive databases. In many practical tasks it usually builds multiple hash tables for a desired level of recall performance. In the literature LSH-based multiple table indexing is usually adopted to independently build a set of hash tables using LSH functions, which can faithfully improve the recall performance (Lv et al. 2007; Norouzi, Punjani, and Fleet 2012; Xia et al. 2013; Cheng et al. 2014). However, without eliminating the table redundancy it often requires a huge number of tables, at the cost of significantly sacrificing precision.