Jensen-Bregman LogDet Divergence with Application to Efficient Similarity Search for Covariance Matrices

Covariance matrices have found success in several computer vision applications, including activity recognition, visual surveillance, and diffusion tensor imaging. This is because they provide an easy platform for fusing multiple features compactly. An important task in all of these applications is to compare two covariance matrices using a (dis)similarity function, for which the common choice is the Riemannian metric on the manifold inhabited by these matrices. As this Riemannian manifold is not flat, the dissimilarities should take into account the curvature of the manifold. As a result, such distance computations tend to slow down, especially when the matrix dimensions are large or gradients are required. Further, suitability of the metric to enable efficient nearest neighbor retrieval is an important requirement in the contemporary times of big data analytics. To alleviate these difficulties, this paper proposes a novel dissimilarity measure for covariances, the Jensen-Bregman LogDet Divergence (JBLD). This divergence enjoys several desirable theoretical properties and at the same time is computationally less demanding (compared to standard measures). Utilizing the fact that the square root of JBLD is a metric, we address the problem of efficient nearest neighbor retrieval on large covariance datasets via a metric tree data structure. To this end, we propose a K-Means clustering algorithm on JBLD. We demonstrate the superior performance of JBLD on covariance datasets from several computer vision applications.

[1]  Alan L. Yuille,et al.  The Concave-Convex Procedure , 2003, Neural Computation.

[2]  Frank Nielsen,et al.  On the Centroids of Symmetrized Bregman Divergences , 2007, ArXiv.

[3]  Frank K. Soong,et al.  On divergence based clustering of normal distributions and its application to HMM adaptation , 2003, INTERSPEECH.

[4]  Ronen Basri,et al.  Actions as space-time shapes , 2005, Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1.

[5]  Erik Learned-Miller,et al.  FDDB: A benchmark for face detection in unconstrained settings , 2010 .

[6]  Inderjit S. Dhillon,et al.  Clustering with Bregman Divergences , 2005, J. Mach. Learn. Res..

[7]  J. Ibrahim,et al.  Statistical Analysis of Diffusion Tensors in Diffusion-Weighted Magnetic Resonance Imaging Data , 2007 .

[8]  Xavier Pennec,et al.  A Riemannian Framework for Tensor Computing , 2005, International Journal of Computer Vision.

[9]  Quanquan Gu,et al.  A similarity measure under Log-Euclidean metric for stereo matching , 2008, 2008 19th International Conference on Pattern Recognition.

[10]  Frank Nielsen,et al.  Bregman vantage point trees for efficient nearest Neighbor Queries , 2009, 2009 IEEE International Conference on Multimedia and Expo.

[11]  Nicholas Ayache,et al.  A Riemannian Framework for the Processing of Tensor-Valued Images , 2005, DSSCV.

[12]  Inderjit S. Dhillon,et al.  Low-Rank Kernel Learning with Bregman Matrix Divergences , 2009, J. Mach. Learn. Res..

[13]  Sergey Brin,et al.  Near Neighbor Search in Large Metric Spaces , 1995, VLDB.

[14]  Nicole Immorlica,et al.  Locality-sensitive hashing scheme based on p-stable distributions , 2004, SCG '04.

[15]  Gert R. G. Lanckriet,et al.  On the Convergence of the Concave-Convex Procedure , 2009, NIPS.

[16]  Janusz Konrad,et al.  Action Recognition Using Sparse Representation on Covariance Manifolds of Optical Flow , 2010, 2010 7th IEEE International Conference on Advanced Video and Signal Based Surveillance.

[17]  Rama Chellappa,et al.  Nearest-neighbor search algorithms on non-Euclidean manifolds for computer vision applications , 2010, ICVGIP '10.

[18]  Yogesh Rathi,et al.  A Graph Cut Approach to Image Segmentation in Tensor Space , 2007, 2007 IEEE Conference on Computer Vision and Pattern Recognition.

[19]  Andrzej Stachurski,et al.  Parallel Optimization: Theory, Algorithms and Applications , 2000, Scalable Comput. Pract. Exp..

[20]  Xuelong Li,et al.  Gabor-Based Region Covariance Matrices for Face Recognition , 2008, IEEE Transactions on Circuits and Systems for Video Technology.

[21]  Frank Nielsen,et al.  Jensen Divergence-Based Means of SPD Matrices , 2013 .

[22]  Emilio Maggio,et al.  Particle PHD Filtering for Multi-Target Visual Tracking , 2007, 2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07.

[23]  I. Dryden,et al.  Non-Euclidean statistics for covariance matrices, with applications to diffusion tensor imaging , 2009, 0910.1656.

[24]  Pavel Zezula,et al.  M-tree: An Efficient Access Method for Similarity Search in Metric Spaces , 1997, VLDB.

[25]  Yuri Ivanov,et al.  Fast Approximate Nearest Neighbor Methods for Non-Euclidean Manifolds with Applications to Human Activity Analysis in Videos , 2010, ECCV.

[26]  David G. Lowe,et al.  Shape indexing using approximate nearest-neighbour search in high-dimensional spaces , 1997, Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[27]  Lawrence Cayton,et al.  Fast nearest neighbor retrieval for bregman divergences , 2008, ICML '08.

[28]  Paul M. Thompson,et al.  3D pattern of brain atrophy in HIV/AIDS visualized using tensor-based morphometry , 2007, NeuroImage.

[29]  Fatih Murat Porikli,et al.  Region Covariance: A Fast Descriptor for Detection and Classification , 2006, ECCV.

[30]  R. Bhatia Positive Definite Matrices , 2007 .

[31]  Jimmie D. Lawson,et al.  The Geometric Mean, Matrices, Metrics, and More , 2001, Am. Math. Mon..

[32]  Maher Moakher,et al.  Symmetric Positive-Definite Matrices: From Geometry to Applications and Visualization , 2006, Visualization and Processing of Tensor Fields.

[33]  Maher Moakher,et al.  Means of Hermitian positive-definite matrices based on the log-determinant α-divergence function , 2012 .

[34]  S. Sra Positive definite matrices and the Symmetric Stein Divergence , 2011 .

[35]  James C. Gee,et al.  Spatial transformations of diffusion tensor magnetic resonance images , 2001, IEEE Transactions on Medical Imaging.

[36]  Frank Nielsen,et al.  Shape Retrieval Using Hierarchical Total Bregman Soft Clustering , 2012, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[37]  Rachid Deriche,et al.  Unsupervised Segmentation Incorporating Colour, Texture, and Motion , 2003, CAIP.

[38]  Fatih Murat Porikli,et al.  Human Detection via Classification on Riemannian Manifolds , 2007, 2007 IEEE Conference on Computer Vision and Pattern Recognition.

[39]  Dario Bini,et al.  Computing the Karcher mean of symmetric positive definite matrices , 2013 .

[40]  N. Ayache,et al.  Log‐Euclidean metrics for fast and simple calculus on diffusion tensors , 2006, Magnetic resonance in medicine.

[41]  Shree K. Nayar,et al.  Reflectance and texture of real-world surfaces , 1999, TOGS.

[42]  Thomas S. Huang,et al.  Emotion Recognition from Arbitrary View Facial Images , 2010, ECCV.

[43]  Chun Chen,et al.  Speech Emotion Classification on a Riemannian Manifold , 2008, PCM.

[44]  Frank Nielsen,et al.  The Burbea-Rao and Bhattacharyya Centroids , 2010, IEEE Transactions on Information Theory.

[45]  Charles R. Johnson,et al.  Matrix analysis , 1985, Statistical Inference for Engineers and Data Scientists.

[46]  Xiaoqin Zhang,et al.  Visual tracking via incremental Log-Euclidean Riemannian subspace learning , 2008, 2008 IEEE Conference on Computer Vision and Pattern Recognition.

[47]  Rui Caseiro,et al.  Foreground Segmentation via Background Modeling on Riemannian Manifolds , 2010, 2010 20th International Conference on Pattern Recognition.

[48]  Gene H. Golub,et al.  Matrix computations (3rd ed.) , 1996 .

[49]  Zhizhou Wang,et al.  A constrained variational principle for direct estimation and smoothing of the diffusion tensor field from complex DWI , 2004, IEEE Transactions on Medical Imaging.

[50]  Robert H. Halstead,et al.  Matrix Computations , 2011, Encyclopedia of Parallel Computing.