168
Transform Based Image Denoising:A Review
Nidhi Soni
Department of E&C
Samrat Ashok Technological Institute, Vidisha
soni.nidhi0108@gmail.com
K.G. Kirar
Department of E&I
Samrat Ashok Technological Institute, Vidisha
kg.kirar@gmail.com
__________________________________________________________________________________________
Abstract – The challenge to remove noise from
original image still exists. Over the past two decades,
different kinds of noise reduction techniques have
been developed. This paper reviews the transform
based denoising techniques and performs their
comparative study. Here we put results of different
approaches including general ridgelets and curvelets,
Empirical Mode Decomposition and Empirical
ridgelets and curvelets. A quantitative measure of
comparisons is presented in terms of PSNR.
Keywords – Ridgelet Transform, Curvelet Transform,
Empirical Mode Decomposition, Intrinsic Mode
Functions, Empirical Ridgelet Transform, Empirical
Curvelet Transform
I. INTRODUCTION
Image noise is the random variation of brightness
or color information in images, which alters the
original information of an image. Image denoising
is done to remove this alteration and restore the
original image. The ridgelet transform discussed in
section II was introduced by Candes and Donoho
[1, 2, 3] and was one of the first directional 2D
wavelet type transform. It considers the usual
wavelet transform on a collection of a 1D signal.
While in the empirical ridgelet transform
elaborated in section IV, we use empirical wavelet
transform instead of standard wavelet transform [4]
The curvelet transform was introduced by Candes
et al. [5, 6, 7]. Curvelet transform can provide a
stable, efficient, and near-optimal representation of
smooth objects discontinuities along smooth
curves. It is a multiscale transform which is more
adapted to the image processing after wavelet
transform. Empirical mode decomposition(EMD)
concept that utilizes the Hilbert–Huang transform
(HHT) is expanded into two dimensions for the
images[8]. It provides a tool for image processing,
further discussed in Section III. The conclusion and
results are discussed in section V and section VI
respectively.
II. RIDGELETS AND CURVELETS
The ridgelet and curvelet transforms were
developed to overcome the limitations of the
wavelet transform. The limitation arises from the
fact that along the important edges in the image, the
two-dimensional (2-D) wavelet transform contains
large wavelet coefficients. That means many
wavelet coefficients are required to reconstruct the
edges in an image properly. Denoising also
becomes challenging with so many coefficients to
estimate. So, to represent smooth function and
edges with a few nonzero coefficients accurately,
the ridgelet transform and curvelet transform were
developed.
Consider the usual wavelet transform on a
collection of a 1D signal. This 1D signal collection
is generated by an inverse 1D Fast Fourier
Transform (FFT) of the 1D spectrum problem
extracted from the 2D FFT of input image along
lines going through the origin and with angle Ɵ. By
using our notations, the classical ridgelet transform
can be written, 0 < n < N (where n = 0 corresponds
to the approximation sub band) [4],
ѡ
R
f(
n, θ, t)= ѡ
1,t
(Ғ*
1,|ω|
( Ғ
p
(f)( θ,|ω|))) (2.1)
And it’s inverse
f(x)= Ғ*
P
(Ғ
1,t
(ѡ*
1,t
(ѡ
R
f
)) (2.2)
ISBN 978-1-5090-4760-4/17/$31.00©2017 IEEE
Proceeding International conference on Recent Innovations is Signal Processing and Embedded Systems (RISE-2017) 27-29 October,2017
Authorized licensed use limited to: Spreadtrum Communications (Shanghai) Co.Ltd. Downloaded on August 03,2020 at 02:43:59 UTC from IEEE Xplore. Restrictions apply.