BoostMap: An Embedding Method for Efficient Nearest Neighbor Retrieval

Athitsos, Vassilis; Alon, Jonathan; Sclaroff, Stan; Kollios, George

doi:10.1109/tpami.2007.1140

Cited by 77 publications

(70 citation statements)

References 57 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most related to some of the techniques here, Athitsos et al [2,3] propose a boosting-based approach which gives a parametric function for mapping points to binary vectors, and can accommodate metric and non-metric target similarity functions. Salakhutdinov and Hinton [56] use a neural network trained with an NCA objective [26] to build codes for textdocuments.…”

Section: Other Unsupervised Methodsmentioning

confidence: 99%

“…Memory usage with LSH is typically greater, however, assuming one opts to mitigate the 0-threshold Hamming distance by expanding the search to multiple independently generated hash tables. 3 Furthermore, whereas a user of Semantic Hashing specifies a radius of interest in the embedded Hamming space, a user of LSH (for the radius-based search variant) specifies the radius of interest in the original feature space.…”

Section: Recap Of Search Strategy Tradeoffsmentioning

confidence: 99%

See 1 more Smart Citation

Learning Binary Hash Codes for Large-Scale Image Search

Grauman

Fergus

2013

Studies in Computational Intelligence

View full text Add to dashboard Cite

Algorithms to rapidly search massive image or video collections are critical for many vision applications, including visual search, content-based retrieval, and non-parametric models for object recognition. Recent work shows that learned binary projections are a powerful way to index large collections according to their content. The basic idea is to formulate the projections so as to approximately preserve a given similarity function of interest. Having done so, one can then search the data efficiently using hash tables, or by exploring the Hamming ball volume around a novel query. Both enable sub-linear time retrieval with respect to the database size. Further, depending on the design of the projections, in some cases it is possible to bound the number of database examples that must be searched in order to achieve a given level of accuracy.This chapter overviews data structures for fast search with binary codes, and then describes several supervised and unsupervised strategies for generating the codes. In particular, we review supervised methods that integrate metric learning, boosting, and neural networks into the hash key construction, and unsupervised methods based on spectral analysis or kernelized random projections that compute affinitypreserving binary codes. Whether learning from explicit semantic supervision or exploiting the structure among unlabeled data, these methods make scalable retrieval possible for a variety of robust visual similarity measures. We focus on defining the algorithms, and illustrate the main points with results using millions of images.

show abstract

Section: Other Unsupervised Methodsmentioning

confidence: 99%

Section: Recap Of Search Strategy Tradeoffsmentioning

confidence: 99%

Learning Binary Hash Codes for Large-Scale Image Search

Grauman

Fergus

2013

Studies in Computational Intelligence

View full text Add to dashboard Cite

show abstract

“…Among them, we focus on pseudo-score based indexing schemes [7]- [10], and use the standard pivot-based indexing scheme [7] and the permutation-based indexing scheme [7], [8] in our experiments in Sect. 5.…”

Section: Metric Space Indexingmentioning

confidence: 99%

A General Framework and Algorithms for Score Level Indexing and Fusion in Biometric Identification

Murakami

Takahashi

Matsuura

2014

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYBiometric identification has recently attracted attention because of its convenience: it does not require a user ID nor a smart card. However, both the identification error rate and response time increase as the number of enrollees increases. In this paper, we combine a score level fusion scheme and a metric space indexing scheme to improve the accuracy and response time in biometric identification, using only scores as information sources. We firstly propose a score level indexing and fusion framework which can be constructed from the following three schemes: (I) a pseudo-score based indexing scheme, (II) a multi-biometric search scheme, and (III) a score level fusion scheme which handles missing scores. A multi-biometric search scheme can be newly obtained by applying a pseudo-score based indexing scheme to multi-biometric identification. We secondly propose the NBS (Naive Bayes search) scheme as a multi-biometric search scheme and discuss its optimality with respect to the retrieval error rate. We evaluated our proposal using the datasets of multiple fingerprints and face scores from multiple matchers. The results showed that our proposal significantly improved the accuracy of the unimodal biometrics while reducing the average number of score computations in both the datasets. key words: biometric identification, score level fusion, missing score, metric space indexing, pseudo-score, multi-biometric search

show abstract

“…As there is no external conditions or parameters of the dataset used, we directly used the values reported in the BoostMap paper [3] for other algorithms namely RRO, RLP, FastMap, VP-Trees 1 . Each subplot shows the exact number of DTW distances that needs to be computed against different values of nearest neighbors to be retrieved, for different accuracies on input dataset.…”

Section: Unipen Handwriting Databasementioning

confidence: 99%

“…Although these methods assume that triangular inequality holds, they work for non-metric distances as well with certain amount of distortion in embedding. Athitsos [3] framed embedding construction as a machine learning task, where AdaBoost is used to combine many simple, 1D embeddings into a multidimensional embedding that preserves a significant amount of the proximity structure in original space. These five techniques are most related to our approach as their main target is to learn embeddings for fast retrieval of nearest neighbors.…”

Section: Introductionmentioning

confidence: 99%

Hierarchical Local Maps for Robust Approximate Nearest Neighbor Computation

Bhatt

Namboodiri

2009

2009 Seventh International Conference on Advances in Pattern Recognition

View full text Add to dashboard Cite

In this paper, we propose a novel method for fast nearest neighbors retrieval in non-Euclidean and non-metric spaces. We organize the data into a hierarchical fashion that preserves the local similarity structure. A method to find the approximate nearest neighbor of a query is proposed, that drastically reduces the total number of explicit distance measures that need to be computed. The representation overcomes the restrictive assumptions in traditional manifold mappings, while enabling fast nearest neighbor's search. Experimental results on the Unipen and CASIA Iris datasets clearly demonstrates the advantages of the approach and improvements over state of the art algorithms. The algorithm can work in batch mode as well as in sequential mode and is highly scalable.

show abstract

BoostMap: An Embedding Method for Efficient Nearest Neighbor Retrieval

Cited by 77 publications

References 57 publications

Learning Binary Hash Codes for Large-Scale Image Search

Learning Binary Hash Codes for Large-Scale Image Search

A General Framework and Algorithms for Score Level Indexing and Fusion in Biometric Identification

Hierarchical Local Maps for Robust Approximate Nearest Neighbor Computation

Contact Info

Product

Resources

About