19-02-2013, 03:30 PM
An Efficient Edge-based Bilateral Filter for Restoring Real Noisy Image
[attachment=51366]
Abstract
This paper presents an efficient edge-based
bilateral filter for real noisy image restoration. By dividing all
pixels of a noisy image into edge region or non-edge region, the
different strategies and parameters are adopted in the edgebased
bilateral filter to balance the conservation of image
features and the reduction of noise level. Extensive
experimental results are shown that this filter achieves very
competitive performance in restoring real noisy images,
compared with other state-of-the-art denoising algorithms1
INTRODUCTION
During acquisition and transmission, images are often
corrupted by noise usually modeled as Gaussian type. In order
to reduce the affect of the noise, many wavelet based
denoising methods have been developed over the past years,
in which the wavelet threshold is one of the most popular
approaches [1]. Accurate and efficient wavelet coefficients
modeling, whether done explicitly or implicitly, is a critical
component of image denoising. In [2], the wavelet coefficients
were modeled with Gaussian density, and locally adaptive
Wiener estimates were used for coefficient variances. Portilla
et al. [3] presented Gaussian scale mixture model and
Bayesian least squares estimator in the wavelet domain. Non-
Gaussian bivariate distributions were introduced in [4] by
taking into account the dependency among the wavelet
coefficients. A new wavelet-based noise reduction algorithm
which used both enhanced directional wavelet bases and
multivariate shrinkage method was introduced in [5]. The
main problem of wavelet-based methods is that they are prone
to produce salient artifacts such as the low frequency noise
and the edge ringing which are related to the structure of the
underlying wavelet.
EXPERIMENTAL RESULTS
In this section, the proposed edge-based bilateral filter (EBF) is
evaluated and compared with other existing techniques. We
compare the EBF algorithm with that of non-local means (NLmeans)
algorithm [12] and multiresolution bilateral filter (MBF)
[10] in restoring real color noisy images. The noisy image in Fig.
3(a) and Fig. 4(a) were captured with two different types of digital
camera at ISO 1250 and ISO 1600 respectively, Fig. 5(a) and Fig.
6(a) were downloaded from [16]. In our color denoising algorithm,
we use the edge-based bilateral filter in luminance-chrominance
(YUV) color space. There are two important criteria that are widely
used to judge the visual quality of restoration of real noisy image:
the conservation of image features and the reduction of noise level.
CONCLUSION
In this paper, an efficient bilateral filter is proposed for
restoring real noisy images, in which the strategies and
parameters are directly based on the edge detection results.
Extensive experimental results are shown that the new method
is competitive with other denoising techniques.
[attachment=51366]
Abstract
This paper presents an efficient edge-based
bilateral filter for real noisy image restoration. By dividing all
pixels of a noisy image into edge region or non-edge region, the
different strategies and parameters are adopted in the edgebased
bilateral filter to balance the conservation of image
features and the reduction of noise level. Extensive
experimental results are shown that this filter achieves very
competitive performance in restoring real noisy images,
compared with other state-of-the-art denoising algorithms1
INTRODUCTION
During acquisition and transmission, images are often
corrupted by noise usually modeled as Gaussian type. In order
to reduce the affect of the noise, many wavelet based
denoising methods have been developed over the past years,
in which the wavelet threshold is one of the most popular
approaches [1]. Accurate and efficient wavelet coefficients
modeling, whether done explicitly or implicitly, is a critical
component of image denoising. In [2], the wavelet coefficients
were modeled with Gaussian density, and locally adaptive
Wiener estimates were used for coefficient variances. Portilla
et al. [3] presented Gaussian scale mixture model and
Bayesian least squares estimator in the wavelet domain. Non-
Gaussian bivariate distributions were introduced in [4] by
taking into account the dependency among the wavelet
coefficients. A new wavelet-based noise reduction algorithm
which used both enhanced directional wavelet bases and
multivariate shrinkage method was introduced in [5]. The
main problem of wavelet-based methods is that they are prone
to produce salient artifacts such as the low frequency noise
and the edge ringing which are related to the structure of the
underlying wavelet.
EXPERIMENTAL RESULTS
In this section, the proposed edge-based bilateral filter (EBF) is
evaluated and compared with other existing techniques. We
compare the EBF algorithm with that of non-local means (NLmeans)
algorithm [12] and multiresolution bilateral filter (MBF)
[10] in restoring real color noisy images. The noisy image in Fig.
3(a) and Fig. 4(a) were captured with two different types of digital
camera at ISO 1250 and ISO 1600 respectively, Fig. 5(a) and Fig.
6(a) were downloaded from [16]. In our color denoising algorithm,
we use the edge-based bilateral filter in luminance-chrominance
(YUV) color space. There are two important criteria that are widely
used to judge the visual quality of restoration of real noisy image:
the conservation of image features and the reduction of noise level.
CONCLUSION
In this paper, an efficient bilateral filter is proposed for
restoring real noisy images, in which the strategies and
parameters are directly based on the edge detection results.
Extensive experimental results are shown that the new method
is competitive with other denoising techniques.