This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Yiqiu Dong∗ Raymond H. Chan† Shufang Xu‡

Abstract This paper proposes a new image statistic for detecting random-valued impulse noise. Combining it with detail-preserving regularization, we obtain a powerful two-phase method for denoising even for noise level as high as 60%. Simulation results show that our method is signiﬁcantly better than a number of existing techniques in terms of image restoration and noise detection.

Key words. random-valued impulse noise, noise detector, detail-preserving regularization, image denoising.

1

Introduction

Images are often corrupted by impulse noise due to noisy sensors or communication channels [1]. There are two models of impulse noise: the easier-to-restore salt-and-pepper noise and the more diﬃcult random-valued impulse noise. This paper deals with the detection and denoising of random-valued impulse noise. Recently, a two-phase iterative method for removing random-valued impulse noise was proposed in [2]. In the ﬁrst phase, they use ACWM ﬁlter [3] to identify noisy pixels. In the second phase, these noise candidates are restored by a detail-preserving regularization. The capability of this method is mainly limited by the accuracy of the noise detector. In [4], Garnett et al. introduced a local image statistic ROAD to identify impulse. For detecting random-valued impulse noise, one drawback of ROAD is that some noise values may be very close to those of their neighbors, in which case, the ROAD values may not be large enough to distinguish them. In this paper, we deﬁne a new local image statistic based on ROAD. By this new statistic, the diﬀerences between noisy pixels and noise-free pixels will be ampliﬁed so that the noise detection will be more accurate. We use this new statistic in phase one of the two-phase method [2]. We have compared our method with a number of methods. It outperforms the others in both image restoration and noise detection. In particular, when the noise ratio is as high as 60%, it still can remove most of the noise while preserving image details. The outline of this paper is as follows. In Section II, we deﬁne the new statistic. Section III describes our method in detail. Section IV includes simulation results to demonstrate the performance of the new method. Finally conclusions are drawn in Section V.

2

Deﬁnition of ROLD

**Suppose the gray-level value yi,j is in [0, 1]. Let ΩN denotes the set of coordinates in a (2N + 1) × (2N + 1) window centered at (0, 0), i.e., ΩN = {(s, t)| − N ≤ s, t ≤ N },
**

∗ LMAM,

School of Mathematical Sciences, Peking University, Beijing, 100871, China(dyiqiu@math.pku.edu.cn). of Mathematics, The Chinese University of Hong Kong, Shatin, Hong Kong(rchan@math.cuhk.edu.hk). ‡ LMAM, School of Mathematical Sciences, Peking University, Beijing, 100871, China(xsf@pku.edu.cn).

† Department

1

Step 2: (Noise detection) If ROLD(ui. 1].e.n)∈Vi. 4 Simulations In this section.. and ROLD for short.j ) ≡ 1 + max{loga |yi+s.j ) = k=1 Rk (yi. At every iteration. N Then we use a logarithmic function to amplify the middle range of its values.j)∈Nk (m.j / = ui.j |. For all (i. and let Rk be the kth smallest Dst for all (s.n ) .j is noise. t) ∈ Ω0 . and (i. 3 Our method We combine the new noise detector ROLD with the detail-preserving regularization [6] to obtain a new two-phase method. We deﬁne our local image statistic as N m ROLDm (yi. The selections of the parameters a and b have great eﬀects on the accuracy of our detection. dst (yi. the noise candidate set. otherwise. the results for 512-by-512. ui. which is the maximum number of iterations. Otherwise. and ϕ is an edge-preserving potential (k+1) (k) function [7]. .j |. i. Step 3: (Noise restoration) Restore all pixels in Nk by minimizing the following function: F (u(k+1) ) = (i. take ui.Mathematical Model for Multi-Channel Image Processing (MultIm’2006) 2 and let Ω0 = ΩN \(0.j − u(k) ) + 2 m. To ensure high accuracy of detection. (k) where Vi.j+t − yi. t) ∈ Ω0 .n)∈Vi. j).j ). j) ∈ Nk . then ui. Note that the value of b decides the truncation position and the value of a controls the shape of the curve of the logarithmic function. 8-bit gray-level images “Lena” and “Bridge” are presented here.j is noise-free. Step 4: Stop the iteration as soon as k is larger than Kmax . ∀(s. We name this statistic as “Rank-Ordered Logarithmic Diﬀerence”. we use a simple truncation and a linear transformation: Dst (yi. j) ∈ N .j .j − ym. N where a. Deﬁne dst as the absolute diﬀerence between gray-level values of yi+s. we apply our method iteratively with decreasing threshold. For illustrations. −b}/b.j is the set of the four closest neighbors of (i. Our algorithm is as follows.j \Nk (k) ϕ(ui.j+t − yi. set k = k + 1. we decrease the threshold to include more noise candidates. Arrange all Dst in an increasing order. we choose a = 2 and b = 5. Suppose the noisy image is y.j ) = |yi+s. In order to keep it in [0.j . for an 8-bit graylevel image. b are positive parameters to be chosen.j ∩Nk (k) (k) (k) ϕ(ui.j+t and N yi. Based on [5]. t) ∈ Ω0 .j ) > Tk .n (m. and go back to Step 2. ∀(s. 0). Algorithm 1: Step 1: Set k = 0 and u(0) = y. we will compare the image restoration and noise detection capability of our method with a number of methods.

there are still many noticeable noise patches. In the results of the old two-phase iterative method. our method outperforms all the others by more than 1 dB. such as edges and lines.95 (a) (b) (c) (d) (e) (f) (g) (h) Figure 1: Results of diﬀerent methods in restoring 60% corrupted images: (a) and (e) original image. we compare our method with all the methods in Table 1 that have noise detectors. we give in Figures 1. Here.14 36. Although no noticeable noise is obtained by the trilateral ﬁlter.21 24.01 20. the visual qualities of our restored images are quite good.2 Comparison of noise detection For good performance. Table 2 lists the number of undetected noisy pixels (“miss” term) and the number of noise-free pixels which are identiﬁed as noise (“false-hit” term). To compare the results subjectively.59 21.32 “Bridge” image 20% 40% 60% 27.84 24.55 20. .57 32.08 36. For 60% noise ratio.12 26.08 23.31 24.07 28.1 Comparison of image restoration In Table 1.29 28.64 35.26 22. From Table 1. our method performs better.84 27. Considering the abundance of image details and the high noise level.07 27.01 27.27 23. the capability of noise detection is very important.60 20.43 27.33 19. In contrast. and can suppress the noise successfully while preserving more details. are not restored well. it can be seen that in all cases our method provides the best results. (b) and (f) the old two-phase iterative method.79 21. 4.19 36. enlarged areas of the images restored by diﬀerent methods. we list the best results in PSNR for the two images with diﬀerent noise ratios.23 19.04 23.70 31. (c) and (g) the trilateral ﬁlter. (d) and (h) our method.44 29.62 37.46 33.27 27. the details.86 21. Because some of the random-valued impulse noise values are not so diﬀerent from their neighbors as in salt-and-pepper noise.89 27. 4.66 24.Mathematical Model for Multi-Channel Image Processing (MultIm’2006) 3 Table 1: Comparison of restoration results in PSNR (dB) Method SD-ROM Filter [8] MSM Filter [9] ACWM Filter [3] Trilateral Filter [4] Two-Phase Iterative Method [2] Our Method “Lena” image 20% 40% 60% 35. there may be much more noise-free pixels detected as noise when detecting .

11:921–924. Charbonnier. 2005.-W. Barlaud. G. 5:1012–1025. H. [8] E. Leung and M. S. Deterministic edge-preserving regulare ization in computed imaging. Gonzalez and R. Ho. Chen and H. 2002. References [1] R. C. A universal noise removal algorithm with an impulse detector. our method can distinguish more noise pixels with fewer mistakes. Mitra and K. for a good random-valued impulse noise detector. H. M. Chan and Shufang Xu. Wu. Chui and W. 2004. Therefore. by which we can identify more noisy pixels with less false-hit pixels. A new detection statistic for random-valued impulse noise removal. K. . C. Chen and H. Aubert and M. C. 1996. but on the other hand. He. T. A new eﬃcient approach for the removal of impulse noise from highly corrupted images. IEEE Transactions on Circuits and Systems II. on one hand it should be able to identify most of the noisy pixels. Nikolova. Simulation results show that our method outperforms a number of existing methods both visually and quantitatively. 48:784–789. Hu and M. Garnett. [5] Yiqiu Dong. Minimization of detail-preserving regularization functional by Newton’s method with continuation. [3] T. [7] P. IEEE Transactions on Image Processing. Nikolova. Arakawa. 1997. its “false-hit” value should be as small as possible. we propose a new local image statistic ROLD. Chan. our method can still identify most of the noisy pixels. R. R. Unpublished. An iterative procedure for removing random-valued impulse noise. IEEE Transactions on Image Processing. IEEE Signal Processing Letters. Wu. Digital Image Processing. Blanc-F´raud. Even when the noise level is as high as 60%. E. Chan. Huegerich. Pearson Education. In ICIP. [6] R. Adaptive impulse detection using center-weighted median ﬁlters. Comparing with others. C. Space variant median ﬁlters for the restoration of impulse noise corrupted images. IEEE Signal Processing Letters. 8:1–3. 14:1747–1754. 2001. Lightstone. We combine it with the detail-preserving regularization to get a powerful method for removing random-valued impulse noise. Woods.Mathematical Model for Multi-Channel Image Processing (MultIm’2006) 4 Table 2: Comparison of noise detection results for image “Lena” Method SD-ROM Filter [8] MSM Filter [9] ACWM Filter [3] Two-Phase Iterative Method [2] Our Method miss 22842 16582 16052 13657 11459 40% false-hit 411 7258 1759 6192 6531 miss 32566 20857 23683 13868 11690 50% false-hit 998 10288 2895 12693 10349 miss 45365 26169 32712 23793 12424 60% false-hit 2651 15778 7644 17573 14789 random-valued impulse noise. 6:298–311. [4] R. [9] T. 2005.-Y. [2] R.-J. 2001. C. pages 125–128. L. Raymond H. IEEE Transactions on Image Processing. 5 Conclusions In this paper. Abreu.

- 2016 Rsd List
- Goldmine grading guidelines.pdf
- Lo-Fi lista.txt
- Kurt Cobain Top 50 Albums of all time
- Goldmine Grading Guidelines
- German Punk
- Punktur01
- hate_recs
- Music on vinyl catalogue
- Belgian Punk
- Originali (2)
- Originali (2)
- Originali (2) 3
- Originali (2)
- Rsd 2011 Releases
- Belgian Punk
- Document
- Cad 3
- Divider
- Readme Sdl
- Under Beograd Agm 020603
- Singer 518

- Recently, stream data mining applications has drawn vital attention from several research communities. Stream data is continuous form of data which is distinguished by its online nature. Traditionally, machine learning area has been developing learning algorithms that have certain assumptions on underlying distribution of data such as data should have predetermined distribution. Such constraints on the problem domain lead the way for development of smart learning algorithms performance is theoretically verifiable.by Lewis Torres

- Use of Kalman Filters in Time and Frequency Analysis
- 4 Box Plots
- Formulae Sheet 2013 Regression
- Automated Filter Parameter Selection Using Measures of Noiseness
- Biometry Face Recognition
- Quantitative Applications in Management
- Yangyang
- Discriminant Ana
- Background Subtraction Algorithm
- Predictability of Nonstationary Time Series Using Wavelet and EMD Based ARMA Models
- 36.Combining Neural Network Model With Seasonal Time Series ARIMA Model
- Recently, stream data mining applications has drawn vital attention from several research communities. Stream data is continuous form of data which is distinguished by its online nature. Traditionally, machine learning area has been developing learning algorithms that have certain assumptions on underlying distribution of data such as data should have predetermined distribution. Such constraints on the problem domain lead the way for development of smart learning algorithms performance is theoretically verifiable.
- Kalman Filters
- 1_Measures of Location Dispersion
- Uncertainties
- Data Scrub
- zeng-wei2013+
- kimki colortexturelpr
- R Z423333342w
- Composite Failure Criteria
- How to Do Call Volume Forecasting for Service Desks
- Taddy
- R Package - Timereg - Analyzing Competing Risk Data 2011 JSS
- Assignment
- fulltext
- 1-s2.0-S003991400500192X-main
- Improved Kalman Filter
- Multimodal Nested Sampling- An Efficient and Robust Alternative to MCMC Methods for Astronomical Data Analysis
- 9fcfd50a35516eabe2
- Six Sigma Green and Black Belt Question Paper From Iibm

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd