Accelerated POP hashing for fast palmprint identification on large-scale databases

A typical palmprint identification system identifies a query palmprint by matching it with the templates in the database and comparing the result with the pre-defined similarity threshold. When applied on large-scale databases, it is often necessary to speed up this process. Previous study shows that POP hashing is very effective for fast and accurate palmprint identification. As state-of-the-art approach, its speedup is around 15 to 20 on databases with 10,000 to 100,000 templates. In this paper, we investigate the POP hashing algorithm in more depth, and propose to accelerate it by improved pre-processing approach, i.e. building the hash table using more consistent orientation features. Experimental results on several real and synthetic databases show that the proposed method can accelerate POP hashing up to 50%~85%, with negligible accuracy loss. Meanwhile, the space requirement of the hash table is reduced by a factor of 2, making it more appealing for practical applications.