Professional Documents
Culture Documents
minimizing the following objective function: noisy image patches almost never exactly lie on the same
manifold due to noise. A related work [26] shows that the
arg Mi n {Di X i 2 + i 1 }, (2) manifold smoothing is a usual trick for effectively removing
D,i i the noise. The weighted neighborhood graph, constructed
where D is an over-complete dictionary and each column from image patches, can approximate the intrinsic manifold
therein corresponds to an atom, and i is the sparse coding structure. The graph Laplacian operator is the generator
coefficient combination of all atoms for reconstructing the of the smoothing process on the neighborhood graph [25].
clean image patch from the noisy image patch X i under the Therefore, the recent promising graph Laplacian operator,
convex sparse priori regularization constraint .1 . in [26][29] and [31], for approximating the manifold
However, the above dictionary D is not easy to structure is leveraged as a generic smooth regularizer while
learn, and the corresponding denoising model uses a removing the noise of 2D image patches based on the sparse
1D vector, rather than the original 2D matrix to rep- coding model.
resent each image patch. Additionally, regarding the With the above considerations, we propose a Dictionary Pair
K-SVD basis, several effective, adaptive denoising methods, Learning model (DPL model) for image denoising. In the DPL
such as [11], [12], and [17][19] were also proposed in the model, the dictionary pair is used to capture the semantic
theme of converting image patches into 1D vectors and clus- features of 2D image patches, and the graph Laplacian
tering noisy image patches into regions with similar geometric operator guarantees a disciplined smoothing according to the
structures. Taking the NCSR algorithm [12] as a classical image patch geometric distribution in the 2D sparse coding
example, it unifies both priors in image local sparsity and non- space. However, we will face the NP-hardness of directly
local similarity via a clustering-based sparse representation. solving the dictionary pair and the 2D sparse coding matrixes
The NCSR algorithm incorporates considerable prior informa- for image denoising. In the NCSR model, the vectorized
tion to improve the denoising performance through introducing image patches are clustered into K subsets by K-means, and
sparse coding noise, (i.e., the third regularization term of the then one compact PCA sub-dictionary for each cluster is used.
following model, which is an extension of the model in Eq.(2)) So, in our DPL model, 2D image patches can, of course,
as follows: be clustered into some subsets with nonlocal similarities.
The 2D patches in a subset are very similar to each other.
arg Mi n {Di X i 2 + ai 1 + i i }, (3) Obviously, one needs only to extend the PCA sub-dictionary
D,i i to a 2DPCA sub-dictionary for each cluster. However, the
where i is a good estimation of the sparse codes i , and 2D image patches sampled from the noisy image with a
and are the balance factors of two regularization terms multi-resolution and sliding window in our DPL model are
(i.e., the convex sparse regularization term and sparse coding of a high quantity and have a non-linear distribution, such
noise term). that clustering faces a serious computational challenge.
In the NCSR model, while enforcing the sparsity of coding Fortunately, the literature [30] proposed a Subspace Indexing
coefficients, the sparse codes i s are also centralized to attain Model on Grassmann Manifold (SIM-GM) that can
a good estimations i s. Dictionary D is acquired by adopting top-to-bottom partition the non-linear space into local
an adaptive sparse domain selection strategy, which executes subspaces with a hierarchical tree structure. Mathematically,
K-Means clustering and then learns a PCA sub-dictionary for a Grassmann manifold is the set of all linear subspaces with a
each cluster. Nevertheless, this strategy still needs to convert fixed dimension [32], [33], and so an extracted PCA subspace
the noisy image patches into 1D vectors, so good estimations in each leaf node of the SIM-GM model corresponds to
i s are difficult to obtain. a point on a Grassmann manifold. To obtaining the most
To summarize, almost all patch-based and sparse effective local space, introducing the Grassmann manifold
coding-driven image denoising methods convert raw, distances (i.e., the angles between linear subspaces [34]),
2D matrix representations of image patches into 1D vectors the SIM-GM is able to automatically manipulate the leaf
for further processing, and thereby break down the inherent nodes in the data partition tree and build the most effective
2D geometric structure of the natural images. Moreover, the local subspace by using a bottom-up merging strategy.
learned dictionary and sparse coding representations cannot Thus, by extending the kind of PCA subspace partitioning
capture the intrinsic position correlations between the pixels on a Grassmann manifold to a 2DPCA subspace pair
within each image patch. On the one hand, to preserve the partitioning on two Grassmann manifolds, we propose a
2D geometric structure of image patches in the transformation Dictionary Pair Learning algorithm on Grassmann-manifolds
domain, a bilinear transformation is particularly appropriate (DPLG algorithm in shorthand). Experimental results on
(for image patches in the matrix representation) for extracting benchmark images and Berkeley segmentation datasets show
the semantic features of the rows and columns from the that the proposed DPLG algorithm is more competitive
image matrixes [20], which is similar to 2DPCA [21] on than the state-of-the-art image denoising methods including
two directions or can also be viewed as a special case of the internal denoising methods and the external denoising
some existing tensor feature extraction methods such as methods.
TDCS [22], STDCS [23] and HOSVD [24]. On the other The rest of this paper is organized as follows: In Section II,
hand, we assume that image patches sampled from a denoised we build a novel dictionary pair learning model for
image lie on an intrinsic smooth manifold. However, the 2D image denoising. Section III first analyzes the learning
4558 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 11, NOVEMBER 2015
methods of the dictionary pair and sparse coding matrixes, pair of the learned dictionary pair, then, exploiting the graph
and then summarizes the dictionary pair learning algorithm Laplacian as a smoothing operator, both smoothing and
on Grassmann-manifolds for image denoising. In Section IV, co-sparsity can be simultaneously guaranteed while mini-
a series of experimental results are shown, and we present the mizing a penalty term on the weighted L 1 -norm divergence
concluding remarks and future work in Section V. between the coding matrix of a given image patch and
those coding matrixes of its nonlocal neighborhood patches,
II. D ICTIONARY PAIR L EARNING M ODEL as in:
According to the above discussion and analysis, to preserve wi j Si S j F,1 , (6)
the original 2D geometric structure and to construct a sparse i, j
coding model for image denoising, the 2D noisy image patches
where wi j is the similarity between the i th patch and its
are encoded by projections on a dictionary pair that correspond
j th neighbor.
to left multiplying a matrix and right multiplying a matrix.
According to our previous research in manifold learning,
Then by exploiting sparse coding and graph Laplacian operator
a patch similarity metric is selected to apply the generalized
smoothing to remove noises, we design a Dictionary Pair
Gaussian kernel function in literature [31]:
Learning model (DPL model) for image denoising in this 1
section.
ex p X i X j F /2i
wi j = i f X j i s k near est neighbor s o f X i , (7)
A. Dictionary Pair Learning Model for 2D Sparse Coding
To preserve the 2D geometrical structure with sparse sensing 0, other wi se.
in the transformation domain, we need only to find two linear where is the normalization factor, i is the variance
transformations for simultaneously mapping the columns and of neighborhood distribution and is the generalization
rows of image patches under the sparse constraint. Let the Gaussian exponent. In this paper, the neighborhood similarity
image patches set be {X 1 , X 2 , . . . , X i , . . . , X n }, X i MN ; is assumed to obey the super-Gaussian distribution:
our method computes the left and right 2D linear transforma- 1
tions to map the image patches into the 2D sparse matrix wi j = ex p X i X j F,1 / 2i . (8)
space. Thus, the corresponding objective function may be
defined as follows:
C. The Final Objective Function
arg Mi n {A T X i B Si F + Si F,1 }, (4) Combining the sparse coding term in Eq. (4) and the
A,B,S i smoothing term in Eq. (6), the final objective function of the
where A MM1 and B NN1 are respectively called DPL model is defined as follows:
the left coding dictionary and the right coding dictionary,
arg Mi n {A TX B S + Si F,1
i i F
S = {Si }, Si M1N1 is the sparse coefficient matrix,
A,B,S i
i
is the regularization parameter, . F denotes the matrix
+ wi j Si S j F,1 },
Frobenious norm, and . F,1 denotes the matrix L 1 -norm (9)
i, j
which is defined as the sum of the absolute values of all its
(1) wi j = 1
entries.
S.t.
j
In this paper, the left and right coding dictionaries are (2) AkT Ak = I, BkT Bk = I, k = 1, . . . , K
combined and called as the dictionary pair A, B. Once the
dictionary pair and the sparse representations are learned, where . F,1 denotes the matrix L 1 -norm which is defined
especially, the left and right dictionaries constrained by as the sum of the absolute values of all matrix elements, and
block orthogonality, each patch X i can be reconstructed by A and B are constrained to be block orthogonal matrices in
multiplying the selected sub-dictionary pair Aki , Bki with the following learning algorithm.
its sparse representation, that is: The above Eq. (9) is an accurate description of the
Dictionary Pair Learning model (DPL model), and Fig. 1
T
X i Aki Si Bki , (5) shows an illustration of the DPL model. In the DPL model,
where the orthogonal sub-dictionaries Aki , Bki are selected to two similar 2D image patches, X i and X j , extracted from
code the image patch X i , and ki is the index of the selected the given noisy image are encoded on two dictionaries
sub-dictionary pair. Note that the selection method of the (i.e., the left dictionary A and the right dictionary B),
ki th dictionary pair is described in Section III-B. which are respectively consisted of sub-dictionary sets
A = {A1 , . . . , Ak , . . . , A K } and B = {B1 , . . . , Bk , . . . , B K }
for computational simplicity, as analyzed in Section III-A. The
B. Graph Laplacian Operator Smoothing left coding dictionary A is used to extract the features of the
Nonlocal smoothing and co-sparsity are the prevailing tech- column vectors from the image patches, and the right coding
niques for removing noises. Clearly, a natural assumption is dictionary B is used to extract the features of the row vectors
that the coding matrixes of similar patches should be similar. from the image patches. For sparse response characteristics,
If similar image patches are encoded only on a sub-dictionary the two learned dictionaries are usually required to be
ZENG et al.: DPLG FOR IMAGE DENOISING 4559
A1T A2 = U V T , (11)
1
m
dictionaries from all leaf nodes lie on the other Grassmann
Si m(A1 , A2 ) = cos l . (12)
manifold. m
l=1
The angles between linear subspaces have intuitively
become a reasonable measure for describing the divergence Therefore, the larger Si m(Ai , A j ) are, the more similar the
between subspaces on a Grassmann manifold [32]. Thus, two subspaces are (i.e., the closer they are on the Grassmann
for computational convenience, the similarity metric between manifold). Those almost same subspaces should be merge into
two subspaces is typically defined by taking the cosines a single subspace. On the other hand, the same situation should
of the principal angles. Taking the left sub-dictionaries for be considered for the right sub-dictionaries Bi , i = 1, . . . , K .
example, the cosines of the principal angles are defined as The similarity metric between the right sub-dictionaries is
follows: defined in the same manner as the above method. Therefore,
Definition 1: Let A1 and A2 be two m-dimensional simultaneously taking the left sub-dictionaries and the right
subspaces corresponding to the two left sub-dictionaries. sub-dictionaries into account, our Sub-dictionary Merging
The cosine of the t th principal angle between the algorithm (SM algorithm) is described in Algorithm 2.
ZENG et al.: DPLG FOR IMAGE DENOISING 4561
Algorithm 2 (SM Algorithm) Sub-Dictionary Merging the non-local neighborhood similarity, and is the balance
Algorithm factor.
As for the balance factor , when the two terms of Eq. (14)
are simultaneously optimized, we can reach the following
conclusion (the proof is shown in Appendix B).
Theorem 2: If X i is the corrupted image patch by noise
N(0, ), and the non-local similarity obeys to the Laplacian
distribution with the parameter i , then the balance factor
= .
2
2i
Clearly, the objective function of Si in Eq. (14) is convex
and can be efficiently solved. The first term is to minimize the
reconstruction error on the sub-dictionary pair Aki , Bki , and
the second term is to ensure the smoothing and co-sparsity
in coefficient matrix space. We initialize the coding matrix
Si and S j by the projections of the image patch X i and its
neighbors X j on the selected sub-dictionary pair Aki , Bki ,
that is:
wi j = 1
S.t. approach, a given noisy image needs to be divided into many
j
overlapping small image patches. The corresponding denoised
where S j is the sparse coding matrix of the j th nearest image is obtained by combining all of the denoised image
image patch on the sub-dictionary pair Aki , Bki , wi j is patches. Let x denote a noisy image, and let the binary
4562 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 11, NOVEMBER 2015
TABLE I
T IME C OMPLEXITY FOR O NE U PDATE OF T WO BASIC S TEPS IN T HREE
D ICTIONARY L EARNING A LGORITHMS : DPLG, NCSR AND K-SVD
TABLE II
T HE SSIM VALUES BY D ENOISING 11 I MAGES AT D IFFERENT N OISE VARIANCE
ZENG et al.: DPLG FOR IMAGE DENOISING 4565
TABLE III
T HE PSNR VALUES BY D ENOISING 11 I MAGES AT D IFFERENT N OISE VARIANCE
4566 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 24, NO. 11, NOVEMBER 2015
TABLE IV
C OMPARISON OF DPLG W ITH S EVERAL D ENOISING M ETHODS
U SING E XTERNAL T RAINING I MAGES
Fig. 5. The average SSIM values and average PSNR values of 11 denoised
images at different noise variance . (a) Average SSIM values of 11 denoised
images. (b) Average PSNR values of 11 denoised images.
Fig. 8. The distribution of the SSIM and PSNR values of denoising 100 noisy
images corrupted by different noises by using NCSR, BM3D, KSVD and our
DPLG algorithm. (a) The distribution of SSIM values. (b) The distribution of
PSNR values.
In this paper, we proposed a DPLG algorithm which is a clear image patch, {Aki , Bki } are the orthogonal dictionary
novel 2D image denoising method working on Grassmann pair.
manifolds and leading to state-of-the-art performance. The E AkiT X B S = 2
i ki i
DPLG algorithm has three primary advantages: 1) the adaptive As for the second term of Eq. (14). According to that the
dictionary pairs are rapidly learned via subspace partitioning F1 -norm Si S j F,1 obeys to the Laplacain distribution
and sub-dictionary pair merging on the Grassmann manifolds; in Eq. (6)
2) 2D sparse representations are notably easy to obtain;
and 3) a graph Laplacian operator makes 2D sparse code E( wi j Si S j F,1 ) = E( wi j E(Si S j F,1 ))
representations vary smoothly for denoising. Moreover, j
j
extensive experimental results achieved on the benchmark = E( wi j 2i )
images and the Berkeley segmentation datasets demonstrated
j
that our DPLG algorithm can obtain better-than-average = E( 2i wi j )
performance for restoring the perceptual visual effect than j
the state-of-the-art internal denoising methods. In the future, = E( 2i ), ( wi j = 1)
we would consider several potential problems, such as learning
j
3D multiple dictionaries for video denoising and exploring = 2i
the fusion of manifold denoising and multi-dimensional
sparse coding techniques. For preserving the scaling consistency, the ratio of two terms
should be equal to 1.
2
A PPENDIX A = 1 = 2 i
2 .
2i
T HE P ROOF OF T HEOREM 1 IN S ECTION III-A
We give a simple proof of Theorem 1.
Proof: A1 and A2 are two D m-dimensional column- ACKNOWLEDGMENT
orthogonal matrixes. The authors would like to thank the anonymous reviewers
A1T A2 is a m-dimensional matrix. for their help.
According to the SVD decomposition of the matrix A1T A2 ,
if 1 , 2 , . . . , m are m eigen-values from the largest to the
R EFERENCES
smallest, and U, V are two m-dimensional orthogonal matrixes
corresponding to the eigen-values 1 , 2 , . . . , m , then [1] P. Chatterjee and P. Milanfar, Is denoising dead? IEEE Trans. Image
Process., vol. 19, no. 4, pp. 895911, Apr. 2010.
U T A1T A2 V = di ag(1 , 2 , . . . , m ) [2] C. Sutour, C.-A. Deledalle, and J.-F. Aujol, Adaptive regularization of
the NL-means: Application to image and video denoising, IEEE Trans.
And V is a m m orthogonal matrix. Image Process., vol. 23, no. 8, pp. 35063521, Aug. 2014.
A2 V is a rotation transformation of subspace A2 by the [3] S. G. Chang, B. Yu, and M. Vetterli, Adaptive wavelet thresholding for
image denoising and compression, IEEE Trans. Image Process., vol. 9,
rotation matrix V , that is: no. 9, pp. 15321546, Sep. 2000.
span(A2) = span(A2 V ) [4] C.-H. Xie, J.-Y. Chang, and W.-B. Xu, Medical image denoising by
generalised Gaussian mixture modelling with edge information, IET
Similarly, the same equation span(A1) = span(A1U ). Image Process., vol. 8, no. 8, pp. 464476, Aug. 2014.
[5] J.-L. Starck, E. J. Candes, and D. L. Donoho, The curvelet transform
Let u k and v k of Eq.(11) respectively be the k th column for image denoising, IEEE Trans. Image Process., vol. 11, no. 6,
of matrixes A1 and A2 , that is: pp. 670684, Jun. 2002.
[6] A. Buades, B. Coll, and J. M. Morel, A review of image denoising
[u 1 , u 2 , . . . , u k , . . . , u m ] = A1 U algorithms, with a new one, Multiscale Model. Simul., vol. 4, no. 2,
[v 1 , v 2 , . . . , v k , . . . , v m ] = A2 V pp. 490530, Feb. 2005.
[7] J. Portilla, V. Strela, M. J. Wainwright, and E. P. Simoncelli, Image
then, denoising using scale mixtures of Gaussians in the wavelet domain,
IEEE Trans. Image Process., vol. 12, no. 11, pp. 13381351, Nov. 2003.
di ag(cos 1 , . . . , cos k , . . . , cos m ) [8] A. Buades, B. Coll, and J.-M. Morel, A non-local algorithm for image
denoising, in Proc. IEEE Comput. Soc. Conf. CVPR, vol. 2. Jun. 2005,
= [u 1 , u 2 , . . . , u k , . . . , u m ]T [v 1 , v 2 , . . . , v k , . . . , v m ] pp. 6065.
= U T A1T A2 V [9] A. Rajwade, A. Rangarajan, and A. Banerjee, Image denoising using
the higher order singular value decomposition, IEEE Trans. Pattern
= di ag(1 , 2 , . . . , m ). Anal. Mach. Intell., vol. 35, no. 4, pp. 849862, Apr. 2013.
[10] K. Dabov, A. Foi, V. Katkovnik, and K. Egiazarian, Image denoising
by sparse 3-D transform-domain collaborative filtering, IEEE Trans.
Image Process., vol. 16, no. 8, pp. 20802095, Aug. 2007.
A PPENDIX B [11] M. Elad and M. Aharon, Image denoising via sparse and redundant
T HE P ROOF OF T HEOREM 2 IN S ECTION III-B representations over learned dictionaries, IEEE Trans. Image Process.,
Proof: Considering the first term of Eq. (14). vol. 15, no. 12, pp. 37363745, Dec. 2006.
[12] W. Dong, L. Zhang, G. Shi, and X. Li, Nonlocally centralized sparse
X i is the corrupted image patch by noise N(0, ), and representation for image restoration, IEEE Trans. Image Process.,
Si is the code of the corresponding clear patch. vol. 22, no. 4, pp. 16201630, Apr. 2013.
And [13] J. Xie, L. Xu, and E. Chen, Image denoising and inpainting with deep
neural networks, in Proc. Adv. NIPS, Dec. 2012, pp. 350358.
E Aki T
X i Bki Si = E Aki T
X i Aki Si Bki
T
Bki F [14] H. Li, Deep learning for image denoising, Int. J. Signal Process.,
Image Process. Pattern Recognit., vol. 7, no. 3, pp. 171180, Mar. 2014.
ZENG et al.: DPLG FOR IMAGE DENOISING 4569
[15] Z. Y. Wang, Y. Z. Yang, J. C. Yang, and T. S. Huang, Design- Xianhua Zeng received the Ph.D. degree in com-
ing a composite dictionary adaptively from joint examples, in Proc. puter software and theory from Beijing Jiaotong
IEEE Comput. Soc. Conf. CVPR, Jun. 2015. [Online]. Available: University, in 2009. He is currently an Associate
http://arxiv.org/pdf/1503.03621.pdf Professor with the Chongqing Key Laboratory of
[16] L. Chen and X. Liu, Nonlocal similarity based coupled dictionary Computational Intelligence, College of Computer
learning for image denoising, J. Comput. Inf. Syst., vol. 9, no. 11, Science and Technology, Chongqing University of
pp. 44514458, Nov. 2013. Posts and Telecommunications, Chongqing, China.
[17] S. Hawe, M. Kleinsteuber, and K. Diepold, Analysis operator learn- He was a Visiting Scholar with the University of
ing and its application to image reconstruction, IEEE Trans. Image Technology at Sydney, Sydney, from 2013 to 2014.
Process., vol. 22, no. 6, pp. 21382150, Jun. 2012. His main research interests include image process-
[18] P. Chatterjee and P. Milanfar, Clustering-based denoising with locally ing, machine learning, and data mining.
learned dictionaries, IEEE Trans. Image Process., vol. 18, no. 7,
pp. 14381451, Jul. 2009.
[19] W. Zuo, L. Zhang, C. Song, and D. Zhang, Texture enhanced image Wei Bian (M14) received the B.Eng. degree in
denoising via gradient histogram preservation, in Proc. IEEE Comput. electronic engineering, the B.Sc. degree in applied
Soc. Conf. CVPR, Jun. 2013, pp. 12031210. mathematics, and the M.Eng. degree in electronic
[20] J. Ye, Generalized low rank approximations of matrices, Mach. Learn.,
engineering from the Harbin Institute of Technology,
vol. 61, nos. 13, pp. 167191, Nov. 2005.
China, in 2005 and 2007, respectively, and the
[21] J. Yang, D. Zhang, A. F. Frangi, and J.-Y. Yang, Two-dimensional
Ph.D. degree in computer science from the Uni-
PCA: A new approach to appearance-based face representation and
versity of Technology, Sydney, Australia, in 2012.
recognition, IEEE Trans. Pattern Anal. Mach. Intell., vol. 26, no. 1,
His research interests include pattern recognition and
pp. 131137, Jan. 2004.
[22] S.-J. Wang, J. Yang, M.-F. Sun, X.-J. Peng, M.-M. Sun, and C.-G. Zhou, machine learning.
Sparse tensor discriminant color space for face verification, IEEE
Trans. Neural Netw. Learn. Syst., vol. 23, no. 6, pp. 876888, Jun. 2012.
[23] S.-J. Wang, J. Yang, N. Zhang, and C.-G. Zhou, Tensor discriminant
color space for face recognition, IEEE Trans. Image Process., vol. 20, Wei Liu (M14) is Professor of the School of
no. 9, pp. 24902501, Sep. 2011. Electronic Engineering, Xidian University, China,
[24] J. Liang, Y. He, D. Liu, and X. Zeng,Image fusion using higher order and has been a research scientist of IBM T. J.
singular value decomposition, IEEE Trans. Image Process., vol. 21, Watson Research Center, Yorktown Heights, NY,
no. 5, pp. 28982909, May 2012. USA from 2012 to 2015. He received the M.Phil.
[25] A. Elmoataz, O. Lezoray, and S. Bougleux, Nonlocal discrete regu- and Ph.D. degrees in electrical engineering from
larization on weighted graphs: A framework for image and manifold Columbia University, New York, NY, USA in 2012.
processing, IEEE Trans. Image Process., vol. 17, no. 7, pp. 10471060, Dr. Liu is the recipient of 2011-2012 Facebook
Jul. 2008. Fellowship and 2013 Jury Award for best thesis
[26] M. Hein and M. Maier, Manifold denoising, in Proc. Adv. NIPS, of Department of Electrical Engineering, Columbia
Jun. 2006, pp. 561568. University. He holds adjunct faculty positions in
[27] M. Zheng et al., Graph regularized sparse coding for image represen- Rensselaer Polytechnic Institute (RPI) and Stevens Institute of Technology.
tation, IEEE Trans. Image Process., vol. 20, no. 5, pp. 13271336, His research interests include machine learning, data mining, information
May 2011. retrieval, computer vision, pattern recognition, and image processing. Dr. Liu
[28] S.-J. Wang, S. Yan, J. Yang, C.-G. Zhou, and X. Fu, A general expo- has published more than 80 peer-reviewed journal and conference papers,
nential framework for dimensionality reduction, IEEE Trans. Image including Proceedings of IEEE, IEEE TIP, IEEE TMI, IEEE TCSVT, ACM
Process., vol. 23, no. 2, pp. 920930, Feb. 2014. TIST, NIPS, ICML, KDD, CVPR, ICCV, ECCV, IJCAI, AAAI, MICCAI,
[29] M. Belkin and P. Niyogi, Laplacian eigenmaps for dimensionality SIGIR, SIGCHI, ACL, DCC, etc. His current research work is geared to
reduction and data representation, Neural Comput., vol. 15, no. 6, large-scale machine learning, big data analytics, multimedia search engine,
pp. 13731396, Jun. 2003. mobile computing, and parallel and distributed computing.
[30] X. Wang, Z. Li, and D. Tao, Subspaces indexing model on Grassmann
manifold for image search, IEEE Trans. Image Process., vol. 20, no. 9,
pp. 26272635, Sep. 2011.
[31] X. H. Zeng, S. W. Luo, J. Wang, and J. L. Zhao, Geodesic distance- Jialie Shen received the Ph.D. in computer science
based generalized Gaussian Laplacian eigenmap, Ruan Jian Xue Bao/J. from the University of New South Wales (UNSW),
Softw., vol. 20, no. 4, pp. 815824, Apr. 2009. Australia, in large-scale media retrieval and
[32] J. Hamm, Subspace-based learning with Grassmann manifolds, database access methods. He was a Faculty Member
Ph.D. dissertation, Dept. Elect. Syst. Eng., Univ. Pennsylvania, Philadel- at UNSW, Sydney, and a Researcher with the
phia, PA, USA, 2008. Information Retrieval Research Group, University
[33] P. Turaga, A. Veeraraghavan, A. Srivastava, and R. Chellappa, Sta- of Glasgow, for a few years. He is currently a
tistical computations on Grassmann and Stiefel manifolds for image Faculty Member of the School of Information
and video-based recognition, IEEE Trans. Pattern Anal. Mach. Intell., Systems with Singapore Management University,
vol. 33, no. 11, pp. 22732286, Nov. 2011. Singapore.
[34] G. H. Golub and C. F. Van Loan, Matrix Computations, 3rd ed.
Baltimore, MD, USA: The Johns Hopkins Univ. Press, 1996.
[35] I. Daubechies, M. Defrise, and C. De Mol, An iterative thresholding
algorithm for linear inverse problems with a sparsity constraint, Com- Dacheng Tao (F15) is Professor of Computer Sci-
mun. Pure Appl. Math., vol. 57, no. 11, pp. 14131457, Nov. 2004. ence with the Centre for Quantum Computation &
[36] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, Image
Intelligent Systems, and the Faculty of Engineering
quality assessment: From error visibility to structural similarity, IEEE
and Information Technology in the University of
Trans. Image Process., vol. 13, no. 4, pp. 600612, Apr. 2004.
[37] P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P.-A. Manzagol, Technology, Sydney. He mainly applies statistics
Stacked denoising autoencoders: Learning useful representations in a and mathematics to data analytics problems and
deep network with a local denoising criterion, J. Mach. Learn. Res., his research interests spread across computer vision,
vol. 11, no. 3, pp. 33713408, Mar. 2010. data science, image processing, machine learning,
[38] K-SVD. [Online]. Available: http://www.cs.technion.ac.il/~elad/software/, and video surveillance. His research results have
accessed Mar. 14, 2014. expounded in one monograph and 100+ publications
[39] Lander and Rover Capture Photos of Each Other. [Online]. at prestigious journals and prominent conferences,
Available: http://english.cntv.cn/20131215/104015_1.shtml, accessed such as IEEE T-PAMI, T-NNLS, T-IP, JMLR, IJCV, NIPS, ICML, CVPR,
May 20, 2014. ICCV, ECCV, AISTATS, ICDM; and ACM SIGKDD, with several best paper
[40] Berkeley Segmentation DataSet (BSDS). [Online]. Available: awards, such as the best theory/algorithm paper runner up award in IEEE
http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/fg/, ICDM07, the best student paper award in IEEE ICDM13, and the 2014
accessed Jun. 13, 2014. ICDM 10 Year Highest-Impact Paper Award.