Professional Documents
Culture Documents
Image and Graphics 9th International Conference ICIG 2017 Shanghai China September 13 15 2017 Revised Selected Papers Part III 1st Edition Yao Zhao
Image and Graphics 9th International Conference ICIG 2017 Shanghai China September 13 15 2017 Revised Selected Papers Part III 1st Edition Yao Zhao
https://textbookfull.com/product/image-and-graphics-10th-
international-conference-icig-2019-beijing-china-
august-23-25-2019-proceedings-part-iii-yao-zhao/
https://textbookfull.com/product/image-and-graphics-8th-
international-conference-icig-2015-tianjin-china-
august-13-16-2015-proceedings-part-iii-1st-edition-yu-jin-zhang-
eds/
https://textbookfull.com/product/image-and-graphics-8th-
international-conference-icig-2015-tianjin-china-
august-13-16-2015-proceedings-part-i-1st-edition-yu-jin-zhang-
eds/
https://textbookfull.com/product/image-analysis-and-processing-
iciap-2017-19th-international-conference-catania-italy-
september-11-15-2017-proceedings-part-i-1st-edition-sebastiano-
Yao Zhao · Xiangwei Kong
David Taubman (Eds.)
LNCS 10668
Image
and Graphics
9th International Conference, ICIG 2017
Shanghai, China, September 13–15, 2017
Revised Selected Papers, Part III
123
Lecture Notes in Computer Science 10668
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7412
Yao Zhao Xiangwei Kong
•
Image
and Graphics
9th International Conference, ICIG 2017
Shanghai, China, September 13–15, 2017
Revised Selected Papers, Part III
123
Editors
Yao Zhao David Taubman
Beijing Jiaotong University UNSW
Beijing Sydney, NSW
China Australia
Xiangwei Kong
Dalian University of Technology
Dalian
China
LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics
These are the proceedings of the 8th International Conference on Image and Graphics
(ICIG 2017), held in Shanghai, China, during September 13–15, 2017.
The China Society of Image and Graphics (CSIG) has hosted this series of ICIG
conferences since 2000. ICIG is the biennial conference organized by the China
Society of Image and Graphics (CSIG), focusing on innovative technologies of image,
video, and graphics processing and fostering innovation, entrepreneurship, and net-
working. This time, Shanghai Jiaotong University was the organizer, and the Nanjing
Technology University and Zong Mu Technology Ltd. Company were the
co-organizers. Details about the past eight conferences, as well as the current one, are
as follows:
This time, the proceedings are published by Springer in the LNCS series. The titles,
abstracts, and biographies of the three invited speakers of plenary talks are presented
first. At ICIG 2017, 370 submissions were received, and 160 papers were accepted. To
ease in the search of a required paper in these proceedings, the 160 regular papers have
been arranged in alphabetical order according to their titles. Another 12 papers forming
a special topic are included at the end.
Our sincere thanks go to all the contributors (around 200), who came from around
the world to present their advanced works at this event. Special thanks go to the
members of Technical Program Committee, who carefully reviewed every single
submission and made their valuable comments for improving the accepted papers.
The proceedings could not have been produced without the invaluable efforts of the
publication chairs, the web chairs, and a number of active members of CSIG.
Honorary Chairs
Guanhua Xu MOST, China
Yuan F. Zheng Ohio State University, USA
General Chairs
Tieniu Tan Chinese Academy of Sciences, China
Hongkai Xiong Shanghai Jiaotong University, China
Zixiang Xiong Texas A&M University, USA
Finance Chairs
Zhihua Chen ECUST, China
Zhenwei Shi Beihang University, China
Award Chairs
Xin Li West Virginia University, USA
Shiqiang Yang Tsinghua University, China
VIII Organization
Publicity Chairs
Mingming Cheng Nankai University, China
Moncef Gabbouj TUT, Finland
Exhibits Chairs
Zhijun Fang Shanghai University of Engineering Science, China
Yan Lv Microsoft Research, China
Publication Chairs
Xiangwei Kong Dalian University of Technology, China
Jun Yan Journal of Image and Graphics, China
International Liaisons
Xiaoqian Jiang UCSD, USA
Huifang Sun MERL, USA
Local Chairs
Wenrui Dai UCSD, USA
Junni Zou Shanghai Jiaotong University, China
Registration Chair
Chen Ye Shanghai Jiaotong University, China
Webmasters
Chenglin Li EPFL, Switzerland
Yangmei Shen Shanghai Jiaotong University, China
Computational Imaging
Security
1 Introduction
operator
2
vect : RN ×N → RN ×1 : vect[f ] := (f1 , f2 , · · · , fN 2 )T = f , where
2
the N -elements are generated by re-ordering the N column vectors of f
sequently.
operator array: the inverse of the vect, i.e., array[f ] = f .
Two-dimensional discrete Fourier transform (DFT) matrix F:
f̂ := vect[F T f F ] = (F ⊗ F )f := Ff ,
By partial frequency data of fˆ, we mean that only parts of the elements of
fˆ are sampled, with other elements considered as 0. Denoted by P the N × N
matrix generating from the identity matrix I by setting its N –M rows as null
vectors, i.e., P = diag(p11 , p22 , · · · , pN N ) with pii being 1 or 0. Then P fˆ means
that we only take M (≤ N ) rows of fˆ as our partial data. To the vector form f ,
the sampling matrix should be modified as an N 2 × N 2 matrix. Then we have
the following notations with tensor product:
vect[P fˆ] = (I ⊗ P )vect[fˆ] = (I ⊗ P )f̂ := Pf̂ . (2.2)
Obviously, the matrix P is the sampling matrix in algorithm domain which is
chosen before reconstruction.
Generally, the frequency data of an image f are obtained by some scanning
process, which are of unavoidable error, i.e., our inversion data for image recovery
are in fact P ĝ δ with noisy data ĝ δ of fˆ satisfying
P fˆ − P ĝ δ F ≤ fˆ − ĝ δ F ≤ δ, (2.3)
where · F is the Frobenius norm for an N × N matrix, corresponding to the
2-norm of an N 2 -dimensional vector after stacking the elements of an N × N
matrix into an N 2 -dimensional vector. Hence the data matching term can be
written as
P fˆ − P ĝ δ 2F = P(Ff ) − Pĝδ 22 . (2.4)
There are various sampling matrices, like those in radial lines sampling
method [12], band sampling method [6] which is much efficient in numerical
experiments. In this paper, we apply the random band sampling process which
samples some rows randomly. Denoted by cenR the central ratio, i.e., to the fre-
quency image there are only cenR × N rows in the central parts of the frequency
image (the centre is between No. N/2−1 and N /2 rows) in Cartesian coordinate
system or in the natural domain coordinate which is shown in Fig. 1(b), and the
others sampling rows are the random ones. In the algorithm domain shown in
Fig. 1(a), the sampling rows are distributed on the four corners, i.e., it is the
same distribution as the mask as usual.
Recall that the most important issue for image restorations is the edge-
preservation property of the image, which means that we are essentially inter-
ested in the efficient reconstruction for a piecewise constant image with the
6 X. Liu and J. Liu
main interests in detecting image edges. Since the total variation (TV) of a two-
dimensional function can describe the function jumps efficiently, we are then led
to the following constraint optimization problem
min |f |T V : P F T f F − P ĝ δ 2F ≤ δ , (2.5)
f ∈RN ×N
N
2 2
|f |T V,β = (∇xm,n
1
f ) + (∇xm,n
2
f) + β (2.6)
m,n=1
for small constant β > 0, where ∇m,n f := ∇xm,n
1
f, ∇xm,n
2
f with two components
fm+1,n − fm,n , if m < N, x2 fm,n+1 − fm,n , if n < N,
∇xm,n
1
f = ∇m,n f =
f1,n − fm,n , if m = N, fm,1 − fm,n , if n = N
1
P F T fα∗1 ,α2 ,β,δ F − P ĝ δ 2F + α1 fα∗1 ,α2 ,β,δ 2F + α2 |fα∗1 ,α2 ,β,δ |T V,β
2
1
≤ P F T f † F − P ĝ δ 2F + α1 f † 2F + α2 |f † |T V,β
2
1 2
≤ δ + α1 f † 2F + α2 (|f † |T V,β − |f † |T V ) + α2 |f † |T V
2
1 N
β
= δ 2 + α1 f † 2F + α2 + α2 |f † |T V
2 † †
m,n=1 |fm,n |2 + β + |fm,n |2
1 2
≤ δ + α1 f † 2F + α2 N 2 β + α2 |f † |T V . (2.11)
2
Since |fα∗1 ,α2 ,β,δ |T V ≤ |fα∗1 ,α2 ,β,δ |T V,β , the proof is complete by the triangle
inequality.
The above decompositions are important for seeking the minimizer of our
cost functional, which is taken as our reconstruction of image. This result gen-
erates the resolution analysis for our reconstruction scheme in terms of the
data-matching and regularity-matching for the image, i.e., the quantitative error
descriptions on these two terms are given. We can adjust the parameters α1 , α2
analytically such that our reconstruction fits our concerns for either image details
(data-matching) or image sparsity (TV difference).
1
min Jβ (f ) = min P F T f F − P ĝ δ 2F + α1 f 2F + α2 |f |T V,β (3.1)
f f 2
finding the minimizer f ∗ approximately, the Bregman iterative algorithm is given
in [13], which is established in terms of Bregman distance [14]. In order to solve
the optimization problem (3.1) iteratively, f (k+1) is yielded by solving its Euler-
Lagrange equation [15]. Due to the penalty terms f F and |f |T V,β , the corre-
sponding Euler-Lagrange equation for the minimizer is nonlinear. So we propose
8 X. Liu and J. Liu
to find the minimizer by the lagged diffusivity fixed point method [16]. Consid-
ering the optimization problem with respect to the image vector f as
1
min Jβ (f ) := min P(Ff ) − Pĝδ 22 + α1 f 22 + α2 |f |T V,β . (3.2)
f f 2
In order to solve the Euler-Lagrange equation of (3.1), we need the derivatives
of data matching term and the penalty terms in (3.2). By straightforward com-
putations, these derivatives have the following expressions:
⎧
⎨ ∇f 12 P F T f F − P ĝ δ 2F = F∗ P∗ PFf − F∗ P∗ Pĝδ ,
∇f f 2F = 2I ⊗ If , (3.3)
⎩
∇f |f |T V,β = L[f ]f ,
with
⎧
⎨ Λ[f ] := diag
d1 [f ] , · · · , dN 2 [f ] ,
1 1
⎩ d [f ] := (N 2 (I ⊗ D) f )2 + (N 2 (D ⊗ I) f )2 + β,
i l =1 i,l l l =1 i,l l
with sampling data P ĝ δ and the spatial approximation f (l) at the (l − 1)th step,
which constitutes the standard Bregman iterative algorithm. Now we propose a
new algorithm based on the Bregman iteration by introducing an inner recursion.
Notice, the real symmetric matrix I ⊗ P may not be invertible due to our
finite sampling matrix P . Therefore an efficient algorithm should be developed
2
for solving the nonlinear system (3.5) with unknown f ∈ RN ×1 . We apply the
lagged diffusivity fixed point method [16].
1
Define Λn [f ] := diag( d(n−1)N +1 [f ]
, · · · , d(n−1)N
1
+N [f ]
), then Λ[f ] = diag(Λ1 [f ],
· · · , ΛN [f ]). Since
L1 [f ] := 2α1 (I ⊗ I) + α2 (I ⊗ D)T diag(Λ1 [f ], · · · , ΛN [f ])(I ⊗ D)
is a real positive block diagonal matrix and
L2 [f ] := α2 (D ⊗ I)T Λ[f ](D ⊗ I)
⎛ 1 ⎞
1
d1 [f ] + dN 2 [f ] − d11[f ] ··· 0 −d 1
N 2 [f ]
⎜ − d1 [f ] ⎟
d1 [f ] + d2 [f ] · · ·
1 1 1
⎜ 0 0 ⎟
⎜ . .. .. .. ⎟
= α2 ⎜⎜
.. .
..
. . . ⎟
⎟
⎜ ··· 1 1
−d 1 ⎟
⎝ 0 0 dN 2 −2 [f ] + dN 2 −1 [f ] N 2 −1 [f ]
⎠
− d 12 [f ] 0 ··· −d 1 1
+ 1
N N 2 −1 [f ] dN 2 −1 [f ] dN 2 [f ]
is a symmetric block matrix, we construct the inner iteration scheme from l-step
for the nonlinear system (3.5) as
L1 [f (l) ]f (l+1) = − N 2 (I ⊗ P ) + L2 [f (l) ] f (l)
+ I ⊗ P F∗ ĝδ − F∗ Ff (l) (3.6)
(l)
where κk is the step-size at k-step in inner recursion defining as
(l) (l)
(l) −rk , (μI + L1 [f (l) ]T L1 [f (l) ])(−rk )T
κk := (l)
,
(μI + L1 [f (l) ]T L1 [f (l) ])(−rk )22
from the classic successive over relaxation method (SOR), , is the L2 inner
(l) (l) (l)
product, and rk := μfk + L1 [f (l) ]T (L1 [f (l) ]fk − b(l) ).
10 X. Liu and J. Liu
(l)
Notice, the initial value f0 in the inner recursion can be chosen as 0 or f (l) ,
and the stopping criterion may be the maximum iteration number K0 or some
others related to the small values of the cost function or the small difference of
the iterative sequences. Here we stop the iteration process when the difference
between f (l+1) and f (l) is smaller than 10−3 . Finally we have the scheme to
find the approximate minimizer by the following iterative algorithm with inner
recursion.
4 Numerical Experiments
All the numerical tests are performed in MATLAB 7.10 on a laptop with an
Intel Core i5 CPU M460 processor and 2 GB memory.
We consider a model problem with Ω = [0, 1]2 and N = 128. Define
2 2
1 1 1
D1 := x = (x1 , x2 ) : x1 − + x2 − ≤ ,
4 2 64
3 1 1 1
D2 := x = (x1 , x2 ) : x1 − ≤ , x2 − ≤ , (4.1)
4 8 2 4
and
⎧
⎪
⎨1, x ∈ D1 ,
f (x) := 2, x ∈ D2 , (4.2)
⎪
⎩
0, x ∈ Ω \ (D1 ∪ D2 ).
The functions f (x) together with its frequency function log(|fˆ(ω)|) in algorithm
domain (i.e., after shifting as Fig. 1) is shown in Fig. 2(a) and (b). Obviously, the
A Double Recursion Algorithm to Image Restoration 11
frequency data in the center (or in the four corners before shifting) consist of
the main information about the image, so we should sample these data as much
as possible.
Fig. 2. (a) f (x); (b) Frequency function log(|fˆ(ω)|) after shifting; (c) and (d) Sampling
noisy frequency data P60 log(|ĝ δ (ω)|) and P90 log(|ĝ δ (ω)|) after shifting.
δ
Firstly we yield the full noisy data gm,n from the exact image fm,n by
δ
gm,n = fm,n + δ × rand(m, n),
where m, n = 1, · · · , N and rand(m, n) are the random numbers in [−1, 1]. The
mesh image of initial image and the noisy image are shown in Fig. 3. Then the
full noisy frequency data are simulated by
,n = F[gm,n ], m , n = 1, · · · , N.
δ δ
ĝm (4.3)
So with the random band row sampling method using sampling matrix P ,
δ
P ĝm ,n is the input incomplete noisy data by row sampling process.
Fig. 3. The mesh of initial image f (x) and noisy image g δ (x).
i ∈ {1 − 5, 16, 17, 18, 23, 34, 37, 39, 40, 43, 44, 45, 47 − 55,
58, 60, 61, 63, 64, 70, 76 − 79, 81, 82, 83, 86, 88 − 91,
94, 95, 97, 98, 100, 101, 103, 105, 107 − 112, 125 − 128} (4.4)
12 X. Liu and J. Liu
and
i ∈ {1 − 14, 16, 17, 18, 23, 27 − 32, 34, 37, 39, 40, 43, 44, 45, 47 − 55,
58, 60, 61, 63, 64, 70, 72, 73, 74, 76 − 79, 81, 82, 83, 86, 88, 89 − 91,
94, 95, 97, 98, 100, 101, 103, 105, 107 − 112, 114 − 128} (4.5)
respectively. Figure 2(c) and (d) show the two-dimensional image of full fre-
quency data, the incomplete noisy frequency data with P60 , P90 after shifting
respectively.
In our iteration process, the Bregman iterative number is L0 = 20, and
the initial value in ACGM inner recursion is f (l) . We compare Algorithm 2
with Algorithm 1, i.e., comparing the proposed scheme to the Bregman iterative
algorithm without inner recursion. Figure 4(a) and (b) give the reconstructed
image f ∗ with P60 , P90 by our proposed algorithm, while Fig. 4(c) shows the
reconstructed image f ∗ with P90 by standard Bregman iterative algorithm.
Fig. 4. (a), (b) The reconstruction of f ∗ with P60 , P90 by our Bregman iterative algo-
rithm with ACGM inner recursion; (c) The reconstruction of f ∗ with P90 by standard
Bregman iterative algorithm without inner recursion.
5 Conclusion
An efficient algorithm to restore image based on L2 − T V regularization penalty
terms is established. The data matching term of the optimizing model is only
used limited data in frequency domain, which are obtained by random band
sampling process. The new idea is that the model is included with two iteration:
Bregman iteration and adjoint conjugate gradient method as inner recursion.
In order to solve the optimizing problem, the Bregman iteration with lagged
Another random document with
no related content on Scribd:
the world. He has carried on that controversy since his Balliol days.
The exigencies that oblige him to pretend, against his better
knowledge and common civility, that I am petty and provincial and
patriotic and wilfully ignorant and pitifully out-of-date, oblige him to
pretend as much about most of those who stand for modern science
and a modern interpretation of history. He would pretend as hard
about Sir Ray Lankester, for example, or about Professor Gilbert
Murray or Sir Harry Johnston or Professor Barker, as he does about
me. It is a general system of pretence. It is a necessary part of—I will
not say of the Catholic attitude, but of his Catholic attitude towards
modern knowledge.
The necessity for a pose involving this pretence is not very
difficult to understand. Long before Mr. Belloc embarked upon the
present dispute he had become the slave of a tactical fiction, which
reiteration had made a reality for him. He evoked the fiction as early,
I believe, as his Oxford days. It may have been very effective at
Oxford—among the undergraduates. Then perhaps it was
consciously a defensive bluff, but certainly it is no longer that. He
has come at last to believe absolutely in this creature of his
imagination. He has come to believe this: that there is a vast
“modern European” culture of which the English-speaking world
knows nothing, of which the non-Catholic world knows nothing, and
with which he is familiar. It is on his side. It is always on his side. It is
simply and purely Belloccian. He certainly believes it is there. It
sustains his faith. It assuages the gnawing attacks of self-criticism
that must come to him in the night. Throughout these papers he is
constantly referring to this imaginary stuff—without ever coming to
precisions. Again and again and again and again—and again and
again and again, he alludes to this marvellous “European” science
and literature, beyond our ken.
He does not quote it; it does not exist for him to quote; but he
believes that it exists. He waves his hand impressively in the
direction in which it is supposed to be. It is his stand-by, his refuge,
his abiding fortress. But, in order to believe in it, it is necessary for
him to believe that no other English-speaking men can even read
French, and that their scepticism about it is based on some
“provincial” prejudice or some hatred of Catholics, or southern
people, or “Dagoes,” or “foreigners,” or what you will. That is why
Nature wilfully ignores the wonderful science of this “Europe”; and
why our Royal Society has no correspondence with it. But he has to
imagine it is there and make his readers imagine it is there, and that
there is this conspiracy of prejudice to ignore it, before he can even
begin to put up any appearance of a case against such a résumé of
current knowledge as the Outline of History.