Blind Image Restoration Based on Wiener Filtering
and Defocus Point Spread Function Estimation
Fengqing Qin
School of Computer and Information Engineering
Yibin University, 644007,
Yibin, China
Abstract—In order to improve the quality of the defocus blurred
image, the defocus point spread function (PSF) of the imaging
system needs to be estimated. A blind image restoration
algorithm was proposed, in which the defocus PSF of the blurred
image was estimated through error-parameter estimation method.
Firstly, the error-parameter curve was generated through
Wiener filtering algorithm. Then, by analyzing the error-
parameter curve, the defocus radius of the blurred image was
estimated. Finally, utilizing the estimated PSF, image restoration
was performed through Wiener filtering algorithm.
Experimental results showed that the defocus PSF was estimated
with high accuracy, and justified the fact that the defocus PSF
estimation plays a great important part in blind image
restoration.
Keywords- blind; image restoration; point spread function;
defocus; Wiener filtering.
I. INTRODUCTION
Image blur may be caused by relative motion between the
camera and the scene, or by an optical system that is out of
focus, atmospheric turbulence, aberrations in the optical system,
etc.
[1]
. In recent years, image restoration has been applied to
many areas such as astronomy, remote sensing imaging and
medical image restoration etc. The blurred image is expressed
as the convolution of the original high resolution image with
the point spread function (PSF) of the imaging system.
In most of the current image restoration algorithms, the PSF
of the imaging system is assumed to be a given type and
parameters. It does not meet the real imaging model of optical
devices and the performance of the restoration algorithm will
decrease. But in many applications, the PSF of the imaging
model is most likely unknown or is known only within a set of
parameters, and such image restoration is called blind
restoration. Thus, blind image restoration problem arises
naturally and is expressed as estimating a high-resolution
image and the PSF simultaneously, which has always been one
advanced issue and challenge in image processing
[2-4]
.
According to the mechanisms of image blurred,
researchers have proposed a series of blind image restoration
method. For example, the blind restoration based on Lucy-
Richardson method, employing the theory of the maximum
likelihood function to estimate the true image and PSF by
cross iteration under the noise of Poisson
[5-6]
,depends on the
initial value of PSF. It is effective only when the initial value
of PSF approaches the true PSF, otherwise the restoration
effect is poor .The accurate estimation of initial value of PSF,
however, is very difficult in the practical application, so the
method is inapplicable for the blind restoration of motion blur
images. Xue-lin Wang et al represent image restoration
method based on wavelet-domain local Gaussian model
[7]
,
assuming wavelet coefficients satisfy the local Gaussian
model. The model exists in some rationality for some images,
but usually the wavelet coefficients of astronomical images
and medical images do not satisfy the assumption, therefore
there are some limitations to the application scope. With a
view to the spectrum characteristic of motion-blur images, Ze-
feng Deng and You-lun Xiong have proposed a kind of
identification method of motion-blur direction based on
frequency-domain algorithm
[8]
, but it is not robust to noise
disturbance and may result in a great error in identification of
PSF if the motion-blur images are noised. Qin et al proposed a
blind sing-image super resolution reconstruction method based
on motion blur
[9]
, the motion PSF is estimated with high
accuracy when the signal-to-noise ratio is higher than 40dB.
The foremost difficulty of blind image restoration is rooted
in the fact that the observed image is an incomplete
convolution
[10]
. The convolution relationship around the
boundary is destroyed by the cut-off frequency, which makes it
much more difficult to identify the PSF of the imaging system.
In general, the Gaussian PSF and motion PSF are mainly
considered in many papers, but the defocus PSF is less
discussed.
In this paper, a blind image restoration method based on
Wiener filtering and defocus PSF estimation was proposed. For
a given defocused image, the defocus PSF was estimated
through error-parameter analysis method. Image restoration
was performed through Wiener filtering algorithm. The
importance of defocus PSF estimation in blind image
restoration was tested through experiments.
II.
WIENER FILTERING IMAGE RESTORATION
METHOD
Wiener filtering is broadly applied in signal and image
processing. Practical experience shows that it is a classical