Cross-View Feature Hashing for Image Retrieval

Traditional cross-view information retrieval mainly rests on correlating two sets of features in different views. However, features in different views usually have different physical interpretations. It may be inappropriate to map multiple views of data onto a shared feature space and directly compare them. In this paper, we propose a simple yet effective Cross-View Feature Hashing (CVFH) algorithm via a “partition and match” approach. The feature space for each view is bi-partitioned multiple times using B hash functions and the resulting binary codes for all the views can thus be represented in a compatible B-bit Hamming space. To ensure that hashed feature space is effective for supporting generic machine learning and information retrieval functionalities, the hash functions are learned to satisfy two criteria: (1) the neighbors in the original feature spaces should be also close in the Hamming space; and (2) the binary codes for multiple views of the same sample should be similar in the shared Hamming space. We apply CVFH to cross-view image retrieval. The experimental results show that CVFH can outperform the Canonical Component Analysis (CCA) based cross-view method.

[1]  Qiang Yang,et al.  Scalable heterogeneous translated hashing , 2014, KDD.

[2]  Antonio Torralba,et al.  Spectral Hashing , 2008, NIPS.

[3]  Guandong Xu,et al.  Coupling Multiple Views of Relations for Recommendation , 2015, PAKDD.

[4]  Wasin Sinthupinyo,et al.  Improved Spectral Hashing , 2011, PAKDD.

[5]  Trevor Darrell,et al.  Learning to Hash with Binary Reconstructive Embeddings , 2009, NIPS.

[6]  Jun Wang,et al.  Comparing apples to oranges: a scalable solution with heterogeneous hashing , 2013, KDD.

[7]  Philip S. Yu,et al.  A General Model for Multiple View Unsupervised Learning , 2008, SDM.

[8]  Piotr Indyk,et al.  Similarity Search in High Dimensions via Hashing , 1999, VLDB.

[9]  Mikhail Belkin,et al.  Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering , 2001, NIPS.

[10]  Yongdong Zhang,et al.  Multiview Spectral Embedding , 2010, IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics).

[11]  Alan M. Frieze,et al.  Min-wise independent permutations (extended abstract) , 1998, STOC '98.

[12]  Alan M. Frieze,et al.  Min-Wise Independent Permutations , 2000, J. Comput. Syst. Sci..

[13]  Chong-Wah Ngo,et al.  Learning Query and Image Similarities with Ranking Canonical Correlation Analysis , 2015, 2015 IEEE International Conference on Computer Vision (ICCV).

[14]  Chengqi Zhang,et al.  Nested Subtree Hash Kernels for Large-Scale Graph Classification over Streams , 2012, 2012 IEEE 12th International Conference on Data Mining.

[15]  Jitendra Malik,et al.  Normalized cuts and image segmentation , 1997, Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[16]  Susan T. Dumais,et al.  Automatic Cross-Language Information Retrieval Using Latent Semantic Indexing , 1998 .

[17]  Nello Cristianini,et al.  Inferring a Semantic Representation of Text via Cross-Language Correlation Analysis , 2002, NIPS.

[18]  Christoph H. Lampert,et al.  Correlational spectral clustering , 2008, 2008 IEEE Conference on Computer Vision and Pattern Recognition.

[19]  Sham M. Kakade,et al.  Multi-view clustering via canonical correlation analysis , 2009, ICML '09.

[20]  Xiaofei He,et al.  Locality Preserving Projections , 2003, NIPS.

[21]  Christoph H. Lampert,et al.  Learning Multi-View Neighborhood Preserving Projections , 2011, ICML.

[22]  Bin Li,et al.  Context-Preserving Hashing for Fast Text Classification , 2014, SDM.

[23]  Hiroshi Nakagawa,et al.  Locally Optimized Hashing for Nearest Neighbor Search , 2015, PAKDD.

[24]  John Shawe-Taylor,et al.  Canonical Correlation Analysis: An Overview with Application to Learning Methods , 2004, Neural Computation.