1
INTRODUCTION
Image signal is often affected by various noises during
generation, transmission and recording. Therefore, how to
recover the original image from a degenerated image
become an important research topic [1]. However, this is an
ill-posed problem [2,3]. There is no analytic solution to this
problem, which means that we can only get numerical
solution. We only focus on the fractional form of the total
variation (TV) method in this paper.
In 1992, Rudin et al. firstly apply the total variation model
to solve image denoising problem [4]. This model has been
proved to be effective in preserving details and removing
noise. The TV model is defined in bounded variation space
(BVS) which can improve smooth of image edge. However,
BVS is piecewise smooth that brings obvious staircase
artifact. After that, many improved methods are proposed
for this model [5]. In 2001, Meyer proposed an oscillation
function model which an approximate dual space based on
BV spaces. The model described the components of the
image vibration from theoretical analysis and get
widespread attention. However, the computation of this
model is rather complicated, especially the process of
solving the Euler equation [6,7]. In 2009, aiming at the
shortcomings of slow convergence of ROF model,
Goldstein and Osher proposed the famous Split Bregman
Iteration algorithm [8]. This algorithm not only solved the
solution of ROF total variation model with
1
l
regularization, but also provided new ideas for other
1
l
regularization problem[9,10,11].
The theory of fractional calculus has appeared for more
than 300 years, but it has not been applied to practical
This work is supported by National Natural Science Foundation of
China under Grant 61873022, 61573052, Beijing Natural Science
Foundation under Grant 4182045 and the Fundamental Research Funds for
the Central Universities under Grant XK1802-4.
engineering. Until recent years, fractional calculus was
successfully applied to many fields, such as automatic
control, signal processing, image processing, biomedicine
and so on. The application of fractional calculus in image
processing has been a lot of success [12]. It is mainly used
to solve the problems of image denoising and enhancement
[13,14,15].
At present, there are many achievements based on
fractional calculus theory in image denoising. In [16], Chen
et al. introduce a new fractional variation optical flow
model, which use the
2
discrete Fourier transform to
implement the numerical algorithm. In 2015, a
fractional-order total variation image denoising algorithm
based on the primal-dual method is proposed [17,18]. Li et
al. proposed an eight degree of freedom fractional calculus
mask
()
DKM
which is used
SBI
to solve the fractional
ROF model [19]. In 2016, Li et al. proposed an adaptive
fractional-order total algorithm
1
l
regularization
1
()
FOTV l−
model which developed a 16 degree of
freedom fractional calculus mask based on Li’s
work[19,20]. The improved fractional-order differential
kernel mask (IFODKM) made great progress in removing
image noise and avoiding staircase artifacts [21]. In this
paper, we propose a new weighted
SBI
algorithm based
on
1
AFOTV l−
, which speeds up the convergence rate and
improves the PSNR of the image. It provides a new way to
solve the
1
OTV l−
model [22].
The remainder of this paper is organized as follows. Section
2 introduce some basic knowledge about fractional calculus
and split Bregman iteration algorithm. In Section 3, a
weighted split Bregman iteration algorithm for
1
AFOTV l−
is proposed. The details of the proposed algorithm will also
be given. The numerical experiment result and discussion
are showed in Section 4. Some conclusions are drawn for
this paper in Section 5.
A weighted split Bregman iteration for adaptive fractional order total variation
model
Dazi Li*; Daozhong Jiang; Qibing Jin
College of information Science and Technology, Beijing University of Chemical Technology, Beijing 100029, PR China
E-mail:lidz@mail.buct.edu.cn
Abstract: Image denoising is an important branch in the process of image processing which has been widely used in
various fields. This paper proposes a weighted split Bregman iteration
()WSBI
algorithm for adaptive fractional order
total variation model, which provides an effective method to deal with the image denoising problem. A weight
coefficient
w
is added to the split Bregman iteration ( )
SBI
algorithm. Compared with the ordinary
SBI
algorithm,
this improved algorithm can achieve faster convergence and higher ( )
SNR
of the image by experiments. This
algorithm also can well preserve the texture details of the image and avoid the staircase artifact.
Key Words:
Split Bregman iteration, Image denoising, Fractional calculus, AFOTV
2036
978-1-7281-0106-4/19/$31.00
c
2019 IEEE