SPEC hashing: Similarity preserving algorithm for entropy-based coding

Abstract

Searching approximate nearest neighbors in large scale high dimensional data set has been a challenging problem. This paper presents a novel and fast algorithm for learning binary hash functions for fast nearest neighbor retrieval. The nearest neighbors are defined according to the semantic similarity between the objects. Our method uses the information of these semantic similarities and learns a hash function with binary code such that only objects with high similarity have small Hamming distance. The hash function is incrementally trained one bit at a time, and as bits are added to the hash code Hamming distances between dissimilar objects increase. We further link our method to the idea of maximizing conditional entropy among pair of bits and derive an extremely efficient linear time hash learning algorithm. Experiments on similar image retrieval and celebrity face recognition show that our method produces apparent improvement in performance over some state-ofthe-art methods.

A probabilistic model for multimodal hash function learning

“…Although spectral hashing is faster and more effective than the previous two methods, it takes a restrictive and unrealistic assumption that the data are uniformly distributed in a hyper-rectangle. Several new methods have since been proposed to relax this restrictive assumption, such as self-taught hashing [44], binary reconstructive embeddings [18], distribution matching [22], and shift-invariant kernel hashing [29]. Compared to spectral hashing, they have shown superior performance.…”

Sparse coding of auditory features for machine hearing in interference

“…For example, some of the methods developed for locality-sensitive hashing, such as SPEC-hashing, may be good alternatives [9]. These are also much faster than vector quantization, which is one reason we are looking into them.…”

Learning Binary Hash Codes for Large-Scale Image Search

“…Mu and colleagues develop a kernel-based maximum margin approach to select hash functions [46], and a semi-supervised approach that minimizes empirical error on a labeled constraint set while promoting independence between bits and balanced partitions is described in [69]. The SPEC hashing approach [42] uses a conditional entropy measure to add binary functions in a way that matches a desired similarity function, but approximately so as to ensure linear run-time. Jain and colleagues develop a dynamic hashing idea to accommodate metrics learned in an online manner, where similarity constraints are accumulated over time rather than made available at once in batch [36].…”