Abstract:Hashing, a widely studied solution to the approximate nearest neighbor search, aims to map data points in the high-dimensional Euclidean space to the low-dimensional Hamming space while preserving the similarity between original points. As directly learning binary codes can be NP-hard due to discrete constraints, a two-stage scheme, namely, “projection and quantization”, has already become a standard paradigm for learning similarity-preserving hash codes. However, most existing hashing methods typically separate these two stages and thus fail to investigate complementary effects of both stages. In this paper, we systematically study the relationship between “projection and quantization”, and propose a novel minimal reconstruction bias hashing (MRH) method to learn compact binary codes, in which the projection learning and quantization optimizing are jointly performed. By introducing a lower bound analysis, we design an effective ternary search algorithm to solve the corresponding optimization problem. Furthermore, we conduct some insightful discussions on the proposed MRH approach, including the theoretical proof, and computational complexity. Distinct from previous works, the MRH can adaptively adjust the projection dimensionality to balance the information loss between the projection and quantization. The proposed framework not only provides a unique perspective to view traditional hashing methods, but also evokes some other researches, e.g., guiding the design of the loss functions in deep networks. Extensive experiment results have shown that the proposed MRH significantly outperforms a variety of state-of-the-art methods over eight widely used benchmarks.

Balanced Hashing

A Balanced Semi-Supervised Hashing Method for CBIR

Harmonious Hashing

Distributed Discrete Hashing by Equivalent Continuous Formulation.

Discrete Double-Bit Hashing

Image Retrieval via Balanced and Maximum Variance Deep Hashing.

Double-Bit Quantization for Hashing

Hamming Compatible Quantization for Hashing

Adaptive Multi-Bit Quantization for Hashing.

Complementary Projection Hashing

Distributed Adaptive Binary Quantization for Fast Nearest Neighbor Search.

Density Sensitive Hashing

Query-Adaptive Hash Code Ranking For Fast Nearest Neighbor Search

Optimized Projection for Hashing.

Accelerated Manhattan Hashing Via Bit-Remapping with Location Information.

Hash Bit Selection for Nearest Neighbor Search

Balanced Deep Supervised Hashing

Fast and Accurate Hashing Via Iterative Nearest Neighbors Expansion.

Minimizing Reconstruction Bias Hashing Via Joint Projection Learning and Quantization

Double-Bit Quantization and Index Hashing for Nearest Neighbor Search

Multi-bit Quantisation for Similarity-Preserving Hashing