Hashing with Binary Matrix Pursuit
This work addresses image retrieval efficiency for applications like search engines, but it appears incremental as it builds upon existing two-stage hashing frameworks.
The paper tackles the problem of improving two-stage hashing methods for image retrieval by proposing a residual learning scheme that can fit any neighborhood structure with arbitrary accuracy and simplifying binary code inference with high-capacity hash functions, resulting in a novel method that significantly outperforms previous studies on standard benchmarks.
We propose theoretical and empirical improvements for two-stage hashing methods. We first provide a theoretical analysis on the quality of the binary codes and show that, under mild assumptions, a residual learning scheme can construct binary codes that fit any neighborhood structure with arbitrary accuracy. Secondly, we show that with high-capacity hash functions such as CNNs, binary code inference can be greatly simplified for many standard neighborhood definitions, yielding smaller optimization problems and more robust codes. Incorporating our findings, we propose a novel two-stage hashing method that significantly outperforms previous hashing studies on widely used image retrieval benchmarks.