Performance study of multi-bit vote count with weak range

Approximate nearest neighbors search can be implemented by hashing based vote counting. This implementation can be extended to multi-bit case when vectors are hashed to multi-bit in each hashing. This improves the retrieval efficiency and saves the computation on hashing operation. The performance of multi-bit vote count can be further improved by the concept of weak range. In this manuscript, multi-bit Vote Count and reference side weak range for multi-bit vote count is discussed and its performance is compared with traditional Vote Count and enhanced Vote Count implemented by binary quantization. Simulation results show that, multi-bit Vote Count significantly improves the kNN accuracy and weak range implementation can further enhance the multi-bit Vote Count performance. Multi-bit Vote Count can also reduce the computational complexity by means of less hashing number required.