|
发布时间: 2017-06-16 |
图像处理和编码 |
|
|
收稿日期: 2016-10-11; 修回日期: 2017-03-01
基金项目: 国家自然科学基金项目(61203330);山东省科技重大专项(新兴产业)(2015ZDXX0801A01);山东大学基本科研业务费资助项目(2015QY001);山东省自然科学基金项目(ZR2014HL093)
第一作者简介: 齐现英(1974-), 女, 副教授, 山东大学控制科学与工程学院博士研究生, 主要研究方向为数字图像处理、数字影像设备、计算机应用等。E-mail:qxy9228@163.com
中图法分类号: TP391.41
文献标识码: A
文章编号: 1006-8961(2017)06-0754-13
|
摘要
目的 随机噪声的噪声阈值具有不确定性和敏感性,寻找一个鲁棒的阀值是非常困难的,这严重影响了噪声的提取效率。为提高噪声判断的准确性,提出一种基于方向特性与中智不确定性融合的双端脉冲检测算法;另外,为加强优良像素在滤波过程中的权重,构建了一种基于像素中智不确定性和ROAD(rank-ordered absolute differences)统计量的新型双边滤波函数。 方法 在噪声检测阶段,首先根据ROLD(rank-ordered logarithmic difference)与噪声阈值T的关系,将污染图像的像素分为超限域像素(ROLD≥ T)、邻限域像素(0.8T ≤ ROLD<T)和安全域像素(ROLD < 0.8T),并利用开关机制完成一次噪声检测。在此基础上,为提高超限域和邻限域像素噪声检测的准确性,采用不同策略对其进行二次噪声排查:对超限域像素,利用新型25像素和9像素4方向模板计算像素基于排序的方向对数差统计量,由该统计量与T的大小关系决定当前像素的噪声真伪;对邻限域像素,则结合当前像素中智不确定性在滤波窗内的排序信息来进一步确定其噪声特性。在滤波阶段,利用像素中智不确定性和ROAD统计量构建新型双边滤波函数,以加强低不确定性和高相似性像素在图像恢复中的权重。 结果 针对实验图像,双端脉冲检测算法的边缘像素提取率最高可达67%、邻限域像素的噪声剔除率最高可达91%,大大降低了阈值对噪声提取的敏感性,从而提高了噪声判断的正确率。在10%~80%噪声范围内,本文算法的主观性能和峰值信噪比都优于其他7种算法。 结论 本文基于双端检测和新型双边滤波函数的新算法,在噪声检测和去噪过程中均充分考虑了图像本身的方向性和噪声的不确定性,因此提高了噪声提取及像素滤波权重的准确性,从而有效地保护了图像的边缘和细节信息。
关键词
随机脉冲噪声; 阈值敏感性; 双端噪声检测; 方向模板; 中智不确定性; 新双边滤波
Abstract
Objective The inherent uncertainty and randomness of a random-valued impulse result in the indeterminacy and sensibility of the noise threshold. Obtaining a robust noise threshold during noise reduction is therefore difficult because of these poor features. Moreover, the noise extraction efficiencies of algorithms used for random-valued impulse reduction are seriously affected. To improve the accuracy of noise detection, this study proposes an efficient double-side impulse noise detection algorithm based on the directional characteristic and neutrosophic indeterminacy information. Given that an accurate and effective weight function is important to the performance of a filter, a new bilateral filter based on neutrosophic indeterminacy and rank-ordered absolute difference (ROAD) statistic is constructed to strengthen the weights of pixels that are similar to the current pixel and restrain the influence of pixels with high indeterminacy. Method For noise detection, an adaptive function of noise threshold T is designed according to the relation between the performance of the peak signal-to-noise ratio (PSNR) and threshold T. The value of T can be automatically adjusted in this function according to the density of noise. After the values of the rank-ordered logarithmic difference (ROLD) and neutrosophic indeterminacy of each pixel are calculated, the pixels in the polluted image are divided into three types according to the relationship between the ROLD value and the threshold. The first type comprises exceeding-threshold pixels whose ROLD values are greater than or equal to threshold T. The second category comprises adjacent-threshold pixels whose ROLD values are equal to or greater than 0.8 T but less than T. The last category comprises security pixels that have ROLD values less than 0.8 T. All pixels belonging to these three types are roughly divided into noise and noise-free sets by using a switching strategy. Pixels belonging to the first type are temporarily classified as corrupted pixels because their ROLD values are greater than T.All pixels belonging to the second and third types are temporarily regarded as uncorrupted pixels because their ROLD values are less than T. To improve the accuracy of noise evaluation in the exceeding-and adjacent-threshold regions, different strategies are adopted to conduct a second noise examination. Pixels in the exceeding-threshold region are further detected by exploiting the directional ROLD statistics. Before the directional statistics are calculated, 45°, 90°, 135°and 180° directional templates are designed, and different templates are applied under different noise densities. Templates with 9 pixels are used in the low-noise environment, and templates with 25 pixels are used in the high-noise condition to limit the harmful impact of noise. The smallest among the four directional ROLD values is selected as a pixel's final directional ROLD. Whether a pixel is a real noise is decided by the relationship between directional ROLD and threshold T. When the selected directional statistic is less than the preset T, the present pixel is firmly regarded as false noise and deleted from the noise set. Pixels in the adjacent-threshold region are re-detected by applying the statistical information of rank-ordered neutrosophic indeterminacy. When the ranking order of a pixel's neutrosophic indeterminacy is in the first three places of the filtering window, the current pixel is reclassified as noise and removed from the noise-free set. During filtering, to enhance the weights of excellent pixels that possess low indeterminacy and high similarity, a new bilateral function is built based on the ROAD statistic and neutrosophic indeterminacy. The weight of a specific pixel is jointly determined by the value of ROAD and the degree of indeterminacy information. All pixels extracted as noises are restored by applying the new bilateral function, and an iterative filtering mechanism is used for denoising to achieve a satisfactory result. Result Experiments show that in the exceeding-threshold region, a number of edge pixels that have been incorrectly evaluated as noises in the first checking can be extracted, and the extraction ratio of these pixels is as high as 67%. In the adjacent-threshold region, about 91% of the pixels whose statistical values are less than but close to the noise threshold can be re-extracted from noise-free set. The proposed double-side impulse noise detection algorithm not only reduces the sensitivity of the impulse noise threshold but also improves the validity of noise judgment. Under noise ratios from 10% to 80%, the proposed algorithm is compared with seven other algorithms. The new method is the best among the algorithms in terms of PSNR, mean differences in PSNR and subjective quality. Conclusion The double-side impulse noise detection algorithm maximizes the use of the image directional characteristic and neutrosophic indeterminacy information during noise checking. It efficiently reduces the false rate in the exceeding-threshold region and the missed rate in the adjacent-threshold region. Therefore, many edge pixels are not damaged. The double-side detection algorithm significantly decreases the sensitivity of the noise threshold and enhances the correctness of noise detection. Furthermore, the new bilateral filter improves the accuracy of a pixel's weight by using two features to measure pixel similarity. In short, the proposed filter based on double-side impulse noise detection and a novel bilateral function fully considers uncertainty and directional information. The edges and details of the image are well protected and efficiently recovered.
Key words
random impulse noise; sensitivity of noise threshold; double-side noise detection; directional template; neutrosophic indeterminacy; new bilateral filter
0 引言
图像在获取、传输等过程中产生的噪声,严重影响了特征提取、目标识别等环节的准确性,因此图像滤波是图像分析处理的首要任务[1]。脉冲噪声是图像常见噪声之一,可分为椒盐噪声和随机噪声两种,前者灰度值固定,检测相对简单,而后者因噪声的随机性和不确定性,其检测效率相对较低。国内外学者就此提出了大量相关算法,但针对这种随机性和不确定性,目前还没有比较理想的、高效的滤波算法。因此,本文主要基于此问题进行研究。
中值滤波MF (median filter)[2]是目前最常用的滤波算法,但由于该算法对图像中所有像素进行滤波处理,致使高污染图像的模糊化现象较严重。解决此问题的方法是利用具有图像细节保护功能的开关策略[3],即只对噪声像素进行处理,其他像素保持不变。对于开关滤波器,噪声特征及其阈值的选取是影响噪声准确判断的关键因素。在方向加权中值滤波DWM (directional weighted median)[4]中,中心像素在45°、135°、180°、90°等4方向上的灰度绝对差加权和被作为噪声判断依据,当四方向加权和的最小值大于噪声阈值时,中心像素将被判为噪声。在低噪声情况下,该算法的检测效率较高;高噪声情况下,由于绝对差加权和受噪声影响较大,算法效率严重下降。Garnett[5]提出了基于ROAD的三边滤波器,该滤波器对具有较大ROAD的像素赋予较小的滤波权重,反之则分配较大的权重。在图像平坦区域,该算法对噪声的抑制能力较强,但在细节区域,由于边缘像素也可能具有较大的ROAD值,降低其滤波权重,必然会导致边缘信息受损。为提高边缘像素和噪声的区分度,Dong[6]将ROAD(rank-ordered absolute differences)作对数变换,提出了ROLD(rank-ordered logarithmic difference)算法,从而有效地保护了图像的边缘信息。但ROLD计算过程中忽略了图像本身的方向特性,对具有较大ROLD值的边缘像素存在误判现象。同源像素滤波HAB (homogeneous amount based)[7]算法将同源像素的个数作为噪声判断特征,并与TVI(total variation inpainting)模型相结合,提高了对随机脉冲的抑制能力。上述几种算法几乎都存在噪声阈值的选取问题,由于随机脉冲噪声本身的不确定性,阈值的大小也具有一定的不确定性,而这种不确定性将会影响噪声提取的准确性,即噪声提取的准确性对阈值具有敏感性,上述算法均忽略了此敏感性的存在。为此,模糊判断准则被引入滤波过程[8-9]。在模糊权重非局部滤波FWNL(fuzzy weighted non-local)[8]算法中,采用ROAD特征的两个阈值
众所周知,随机噪声的最大特点就是其随机性和不确定性,如果在滤波过程把噪声的不确定性考虑进去,必将对算法的滤波效果有所提高。中智学(neutrosopy)专门从对立统一的角度来分析和探索自然科学和社会科学中的不确定性,它的基本观点就是任何命题、事物等都具有真实性、不确定性和虚假性[12]。假如存在一命题 < A>,则 < A>的非记为 < Anti-A>,既不是 < A>又不是 < Anti-A>的不可确定性记为 < Neut-A〉,并用3个中智元素
综上所述,为克服噪声阈值的不确性和敏感性,研究新的噪声特征并将多个特征进行融合是提高随机脉冲噪声滤除的一个有效途径。基于此,本文首先提出一种基于方向信息的新RODLD(rank-ordered directional logarithmic difference)统计量,在将其与NI相融合的基础上提出一种双端噪声检测算法;另外,为加强优良像素的滤波权重,本文提出用ROAD和NI两个特征构建新型双边滤波函数,以加强高相似性和低不确定性像素对图像的恢复作用。
1 方向模板的建立及RODLD统计量
1.1 传统ROAD和ROLD特征
设中心像素
$\begin{align} & {{d}_{(k,l)}}\left( i,j \right)=|u\left( k,l \right)-u\left( i,j \right)|,\text{且} \\ & \quad \quad \quad \quad \left( k,l \right)\ne \left( i,j \right) \\ \end{align}$ | (1) |
$R_{m}^{\text{ROAD}}\left( i,j \right)=\sum\limits_{k=1}^{m}{{{R}_{n}}\left( i,j \right)}$ | (2) |
式中,
在图像平坦区域,ROAD特征具有较强的噪声判断能力,但在边缘及细节区域,部分像素有时具有与噪声相似的ROAD值,容易出现误判现象。为增加噪声与原始像素的区分度,文献[6]在ROAD基础上提出了ROLD统计量,即对式(1) 中的绝对灰度差进行对数计算。ROLD的计算为
$\begin{align} & {{D}_{\left( k,l \right)}}\left( i,j \right)={{\log }_{2}}\left| u\left( k,l \right)-u\left( i,j \right) \right|, \\ & \quad \quad \quad \quad \left( k,l \right)\ne \left( i,j \right) \\ \end{align}$ | (3) |
$R_{m}^{\text{ROAD}}\left( i,j \right)=\sum\limits_{k=1}^{m}{{{\varphi }_{n}}\left( i,j \right)}$ | (4) |
式中,
虽然ROLD较ROAD具有更高的噪声检出特性,但两者都忽略了图像的方向特性对该特征值的影响。其实在
1.2 4方向模板的建立
基于上述原因,为降低噪声误判率,提出基于排序的方向对数差统计量RODLD。
在文献[6]中,进行ROLD计算时采用了5×5和3×3方形模板。要计算像素的RODLD,必须先设计方向模板。为保证与5×5或3×3模板噪声阈值的一致性,方向模板既要考虑图像的方向性,又要保证模板像素数量为25或9。简单起见,只沿45°、135°、180°、90°等4个主方向计算RODLD值。
首先分析25像素模板的设计思路:设
在低噪声环境下,为减小大尺寸模板造成的图像模糊,本文又设计了如图 2(a)—(d)所示的9像素模板。4邻域像素与中心像素的相似性最强,所以
1.3 RODLD统计量
假如
${\boldsymbol{A}_{1}}=\left[ \begin{matrix} {4.95} & {4.35} & {3.27} & {2.13} & {6.94} \\ {2.53} & {3.54} & {3.39} & {2.19} & {4.31} \\ {2.19} & {6.06} & \boldsymbol{3.70} & {4.54} & {4.63} \\ {7.06} & {3.13} & {4.39} & {6.72} & {1.20} \\ {1.87} & {7.89} & {3.44} & {3.96} & {5.27} \\ \end{matrix} \right]\\ {\boldsymbol{A}_{2}}=\left[ \begin{matrix} {4.95} & {4.35} & {3.27} & {2.13} & {6.94} \\ {2.53} & {2.63} & {3.39} & {2.19} & {4.31} \\ {2.19} & {6.06} & \boldsymbol{2.10} & {4.54} & {4.63} \\ {7.06} & {3.13} & {4.39} & {6.72} & {1.20} \\ {1.87} & {7.89} & {2.20} & {3.96} & {5.27} \\ \end{matrix} \right]\\ {\boldsymbol{B}_{1}}=\left[ \begin{matrix} {1.41} & {3.80} & {2.58} & {3.65} & {3.34} \\ {0.64} & {5.78} & {2.53} & {1.58} & {1.22} \\ {1.46} & {4.76} & \boldsymbol{3.59} & {4.77} & {1.60} \\ {2.95} & {3.04} & {2.24} & {2.38} & {3.29} \\ {8.99} & {4.21} & {3.95} & {3.91} & {2.80} \\ \end{matrix} \right]\\ {\boldsymbol{B}_{2}}=\left[ \begin{matrix} {1.41} & {3.80} & {2.58} & {3.65} & {3.34} \\ {0.64} & {5.78} & {2.53} & {1.58} & {1.22} \\ {1.46} & {4.76} & \boldsymbol{2.66} & {4.77} & {1.60} \\ {2.95} & {3.04} & {2.24} & {2.38} & {3.29} \\ {8.99} & {4.21} & {3.95} & {3.91} & {2.80} \\ \end{matrix} \right]$ |
根据实验统计,噪声密度在40 % 60 %时,图像的ROLD噪声阈值
对A点来说,在
1.4 RONI统计量
设
$I\left( i,j \right)=\frac{\delta \left( i,j \right)-{{\delta }_{\min }}}{{{\delta }_{\max }}-{{\delta }_{\min }}}$ | (5) |
$\delta \left( i,j \right)=\text{abs}\left( u\left( i,j \right)-z\left( i,j \right) \right)$ | (6) |
式中,
根据传统开关机制,如果像素的ROLD小于噪声阈值
2 本文算法
2.1 双端噪声检测
如图 4所示,首先根据ROLD与噪声阈值
所谓双端噪声检测,是指为提高噪声提取的准确性,利用新的相似性特征分别对超限域像素和邻限域像素进行二次噪声检测。针对超限域像素,用RODLD替换ROLD,然后再次利用开关机制进行判断;针对邻限域像素,则结合RONI信息在降序序列中的位置进行二次判断。
由上述分析可知,虽然本文算法也要预先设定噪声阈值
2.2 新型双边滤波函数
噪声提取的准确性是影响滤波算法的关键因素之一,除此之外,权重函数也是影响滤波性能的重要因素。传统双边滤波函数[21]利用了位置信息和几何信息来定义像素的权重,对高斯噪声很有效,但对脉冲噪声的效果较差。本文提出一种基于像素ROAD和NI的新型双边滤波函数。
像素ROAD值越小,该像素为噪声的可能性越小,对应滤波权重应该被加强;像素的不确定性越低,其为噪声的可能性也越低,对目标点的权重也应被加强。把具有低不确定性和小ROAD值的像素称为优良像素。为充分发挥优良像素在图像恢复中的作用,分别设定如式(7)(8) 所示的NI因子和ROAD因子,并将二者融入权重函数。
$IF=\left( 1-\beta \right)$ | (7) |
$RF=\exp \left( -\frac{\text{ROA}{{\text{D}}^{2}}}{2{{h}^{2}}} \right)$ | (8) |
式中,
$w=IF\times RF$ | (9) |
将
$w=\left( 1-\beta \right)\exp \left( -\frac{\text{ROA}{{\text{D}}^{2}}}{2{{h}^{2}}} \right)$ | (10) |
从式(10) 可见,ROAD和NI共同决定了像素的滤波权重,所以该新型双边滤波函数增强了那些小ROAD、低NI像素的权重,图像的原始信息得到了充分利用。
2.3 算法的实现步骤
1) 计算所有像素的ROAD、ROLD和NI矩阵。
2) 根据ROLD值和阈值
3) 双端检噪声检测:
(1) 超限域检测。利用新设计的4方向模板计算像素的RODLD。当RODLD≤0.8
(2) 邻限域检测。如果像素在滤波窗内的RONI信息位于排序序列的前三位,该像素被重新判定为噪声。
(3) 安全域检测。步骤(1) 中RODLD≤0.8
4) 重复步骤3),完成所有像素检测。
5) 利用式(10) 对所有噪声进行滤波。
6) 迭代滤波,滤波次数在4 7之间都可达到较好的效果,本文迭代次数选为6。
3 实验结果
为验证本文算法的有效性,对近100幅图像进行实验仿真,图 5为4幅代表性图像。实验环境为MATLAB R2013a。
3.1 噪声阈值$T$ 的选取
虽然双端噪声检测算法降低了噪声阈值的敏感性,但如果
$T=2+5(0.4-Q),Q<0.4$ | (11) |
式中,
由式(11) 可见,当
由于
综上所述,阈值
$T=\left\{ \begin{array}{*{35}{l}} 2+5\times \left( 0.4-Q \right) & Q<40% \\ 3.5 & Q\ge 40% \\ \end{array} \right.$ | (12) |
3.2 RODLD的边缘提取性能
3.2.1 RODLD的有效性
首先研究进行RODLD计算的规律。图 7(a)为40 %噪声密度下Lena图像的所有RODLD像素。设图像中噪声位置是已知的,通过RODLD计算可提取的非噪声像素如图 7(b)所示,图 7(c)则为真实噪声像素。图 8(a)—(c)为Fish图像对应的像素图。分析图 7和图 8,可发现RODLD计算的两大规律:1) RODLD像素中很大一部分属于边缘像素而不是真正的噪声;2) RODLD计算主要发生在图像边缘纹理部分而非平坦区域。
3.2.2 RODLD边缘提取性能
通过RODLD计算和双端检测算法,如果能将图 7(b)、图 8(b)中具有大于噪声阈值
设图像中进行RODLD计算的像素数量为
以噪声密度在40 % 60 %范围内的边缘提取为例进行分析。表 1为测试图像在不同噪声下的非噪声率及其提取率。经分析可见,噪声密度在40 % 60 %变化时,RODLD像素的非噪声率
表 1
不同噪声下测试图像的
Table 1
The
参数 | Lena | Pepper | Bug | Fish | ||||||||
40% | 50% | 60% | 40% | 50% | 60% | 40% | 50% | 60% | 40% | 50% | 60% | |
9 442 | 9 945 | 14 117 | 8 467 | 9 235 | 13 805 | 5 520 | 6 857 | 10 170 | 9 210 | 10 148 | 12 917 | |
5 362 | 4 442 | 5 010 | 4 520 | 4 237 | 4 950 | 2920 | 2 915 | 3 442 | 5 518 | 5 209 | 5 849 | |
3 522 | 2 607 | 3 127 | 2 942 | 2 625 | 3 430 | 1 917 | 1710 | 2 252 | 2 204 | 2 229 | 3 039 | |
56.79 | 44.67 | 36.66 | 53.38 | 42.76 | 35.86 | 52.90 | 42.51 | 33.85 | 59.91 | 51.33 | 45.28 | |
65.69 | 58.69 | 62.43 | 65.10 | 61.95 | 67.02 | 65.67 | 58.66 | 65.43 | 39.94 | 42.79 | 51.96 |
3.3 邻限域像素的噪声剔除性能
在二次检测过程中,设邻限域像素被重新判定为噪声的数量为
表 2
测试图像在不同噪声下的
Table 2
The
参数 | Lena | Pepper | Bug | Fish | ||||||||
40% | 50% | 60% | 40% | 50% | 60% | 40% | 50% | 60% | 40% | 50% | 60% | |
8 730 | 10 276 | 11 495 | 8 265 | 10 027 | 10 571 | 8 866 | 11 337 | 12 616 | 24 385 | 24 346 | 16 949 | |
6 467 | 8 341 | 9 985 | 6 794 | 8 757 | 11 613 | 6 621 | 8 988 | 10 996 | 9 807 | 12 240 | 10 643 | |
74.08 | 81.17 | 86.86 | 82.20 | 87.33 | 91.03 | 74.68 | 79.28 | 87.16 | 40.22 | 50.28 | 62.79 |
3.4 滤波性能比较
3.4.1 算法的主观性能
图 9、图 10是在60 %噪声环境下,各算法对Lena、Fish滤波的局部图像。
由图 9、图 10可见,MF、ADTM均存在模糊且噪声滤除不净的现象;HAB算法对图像细节保持较好,但也存在残余噪声点;ROAD和FWNL算法的过平滑现象较严重;CROLD算法和DWM的滤波效果比前面几种算法要好,但图像细节的清晰度远低于本文算法。图 9(i)中帽子边缘及细纹以及图 10(i)中锯齿状鱼鳍清晰可见,说明新算法对图像的纹理、边缘等细节保护较好,因此其主观效果优于其他算法。
3.4.2 算法的客观性能
表 3给出了10 % 80 %噪声环境下8种算法对Lena和Fish图像的PSNR值。由表 3可见,对于平滑图像Lena来说,在10 % 60 %密度范围内,DWM算法因考虑方向因素而具有较高的PSNR,而新算法不但利用了图像的方向信息,而且利用了噪声的不确定性信息,所以具有比DWM更高的滤波性能。CROLD仅采用了新双边滤波函数,而忽略了图像的方向,导致边缘部分存在误判或漏判,所以并没有显示出新型双边函数的优越性。当噪声增加到60 % 70 %时,DWM算法的性能下降明显,虽然CROLD的性能有所提高,但依然低于本文算法,这说明双端检测算法提高噪声判断准确性的同时,新双边滤波函数又提高了滤波权重的准确性,所以在10 % 70 %噪声范围内,新算法的PSNR明显高于其他所有算法。对于Fish图像,DWM算法较新算法的滤波性能明显下降,这说明对细节丰富图像来说,即便考虑了方向因素,噪声判断的准确性依然需进一步提高。双端检测算法降低了阈值对噪声判断准确性的影响,所以对此类图像仍然具有较高的滤波性能。不可否认的是,无论是对Lena图像还是对Fish图像,当噪声增加到80 %时,新算法滤波性能也有所下降,但其PSNR仍明显高于MF、HAB、ADTM、DWM、CROLD等5种算法,同时仍然稍高于ROAD和FWNL两种算法。
表 3
不同算法在10 % 80 %噪声下的PSNR
Table 3
The PSNR of all algorithms under 10 % 80 % noise density
/dB | |||||||||
图像 | Noise | MF | ROAD | HAB | FWNL | ADTM | DWM | CROLD | 本文 |
Lena | 10% | 31.11 | 31.82 | 34.26 | 32.60 | 31.35 | 35.85 | 35.20 | 35.99 |
20% | 30.21 | 31.33 | 32.46 | 32.27 | 30.27 | 33.59 | 32.84 | 33.72 | |
30% | 28.86 | 30.78 | 31.03 | 31.63 | 28.75 | 31.40 | 30.96 | 31.73 | |
40% | 27.26 | 29.98 | 29.90 | 29.02 | 27.78 | 30.54 | 28.92 | 30.82 | |
50% | 24.53 | 28.47 | 28.23 | 27.93 | 26.08 | 28.88 | 28.10 | 29.67 | |
60% | 21.09 | 26.88 | 26.20 | 26.44 | 23.20 | 26.56 | 26.97 | 28.12 | |
70% | 18.09 | 25.10 | 23.34 | 24.77 | 19.39 | 21.98 | 25.84 | 25.99 | |
80% | 15.66 | 21.31 | 18.76 | 21.21 | 16.53 | 18.01 | 21.02 | 21.40 | |
Fish | 10% | 24.35 | 24.99 | 26.17 | 25.17 | 25.45 | 23.56 | 29.27 | 30.18 |
20% | 23.80 | 24.63 | 25.11 | 24.68 | 24.34 | 22.13 | 26.91 | 27.91 | |
30% | 22.86 | 24.30 | 24.34 | 24.64 | 23.42 | 21.54 | 25.50 | 26.45 | |
40% | 21.61 | 23.71 | 23.35 | 22.80 | 22.59 | 20.49 | 23.42 | 24.39 | |
50% | 19.37 | 22.80 | 22.34 | 21.95 | 21.07 | 18.82 | 22.49 | 23.40 | |
60% | 16.71 | 21.76 | 20.80 | 21.03 | 18.75 | 16.64 | 21.75 | 22.22 | |
70% | 14.43 | 20.21 | 17.81 | 19.56 | 15.84 | 14.48 | 20.12 | 20.46 | |
80% | 12.68 | 16.31 | 13.93 | 16.43 | 13.36 | 12.74 | 16.02 | 16.46 |
为体现新算法的普适性,对100幅图像的PSNR指标进行了均值统计。设新算法对第
${{D}_{i}}={{P}_{i}}-{{P}_{ij}}$ | (13) |
对第
$MFPSN{{R}_{j}}=\frac{1}{100}\sum\limits_{i=1}^{100}{{{D}_{i}}}$ | (14) |
7种比较算法在不同噪声下的MFPSNR如表 4所示。MFPSNR越大,说明对应算法的滤波性能越差。
表 4
7种比较算法在不同噪声情况下的MFPSNR
Table 4
MFPSNRs of 7 methods under different noise
/dB | |||||||
Noise | MF | ROAD | HAB | FWNL | ADTM | DWM | CROLD |
10% | 5.72 | 4.88 | 3.05 | 4.70 | 4.62 | 1.06 | 0.71 |
20% | 3.89 | 3.15 | 2.34 | 2.82 | 3.66 | 0.73 | 0.92 |
30% | 2.82 | 1.83 | 1.60 | 1.51 | 2.82 | 0.68 | 0.84 |
40% | 2.70 | 0.86 | 0.96 | 1.78 | 2.01 | 0.57 | 1.27 |
50% | 3.81 | 0.85 | 0.99 | 1.39 | 2.40 | 0.93 | 0.94 |
60% | 5.39 | 0.71 | 1.24 | 1.07 | 3.57 | 1.82 | 0.78 |
70% | 6.20 | 0.56 | 1.88 | 0.68 | 4.76 | 2.98 | 0.47 |
80% | 4.71 | 0.15 | 2.27 | 0..28 | 3.80 | 2.95 | 0.41 |
由表 4可见,在10 % 80 %噪声范围内,MF、HAB、ADTM、DW和CROLD等算法的MFPSNR较大,说明新算法的滤波性能远优于此5种算法。对ROAD和FWNL两种算法来说,在10 % 70 %噪声范围内其MFPSNR较大,本文算法的优越性体现得较明显。当噪声增加到80 %时,ROAD和FWNL两种算法的MFPSNR下降较快,说明本文算法相对这两种算法的优越性有所下降。但从对100幅图像的平均滤波性能来说,因为ROAD和FWNL两算法的MFPSNR仍然大于0,表明它们的降噪能力和普适性仍然低于本文算法。
综上所述,本文算法的主观性能和客观性能均优于相关比较算法。
4 结论
本文设计的四方向滤波模板,充分考虑了图像的方向信息,对应的RODLD统计量比ROLD更能反映像素的噪声特性;在RODLD和RONI基础上提出的双端噪声检测算法,通过利用方向特性和不确定特性对超限域和邻限域像素进行二次噪声检测,在一定程度上降低了噪声阈值的敏感性,提高了噪声检测的准确率;以ROAD和NI为相似性度量的新型滤波函数,提高了高相似性和低不确定性像素的滤波权重,有效保护了图像的边缘纹理信息。在10 % 80 %范围内,本文算法对随机脉冲噪声的滤除和细节保护能力优于相关传统算法。然而,本文算法也有一定的局限性,因只考虑了像素在45°、135°、180°、90°等4个方向的RODLD统计量,在一定程度上限制了算法性能的进一步提高;另一方面,当噪声高于80 %时,算法的滤波性能下降较快,不利于高噪声环境下的噪声去除。后续将利用8方向模板对高密度随机噪声进行研究,以使本文算法在10 % 90 %噪声范围内均可得到较满意的滤波效果。
参考文献
-
[1] Yang X Z, Xu Y, Fang J, et al. New filter based on region segmentation and bilateral filtering[J]. Journal of Image and Graphics, 2012, 17(1): 40–48. [杨学志, 徐勇, 方静, 等. 结合区域分割和双边滤波的图像去噪新算法[J]. 中国图象图形学报, 2012, 17(1): 40–48. ] [DOI:10.11834/jig.20120106]
-
[2] Brownrigg D R K. The weighted median filter[J]. Communications of the ACM, 1984, 27(8): 807–818. [DOI:10.1145/358198.358222]
-
[3] Lin Y M, Li Z Y, Lin Y Y, et al. Distance-weighted approach based on self-adaptive windows to remove salt and pepper noise in images[J]. Journal of Image and Graphics, 2015, 20(8): 1008–1016. [林亚明, 李佐勇, 林叶郁, 等. 自适应滤波窗实现距离加权图像椒盐噪声滤除[J]. 中国图象图形学报, 2015, 20(8): 1008–1016. ] [DOI:10.11834/jig.20150803]
-
[4] Dong Y Q, Xu S F. A new directional weighted median filter for removal of random-valued impulse noise[J]. IEEE Signal Processing Letters, 2007, 14(3): 193–196. [DOI:10.1109/LSP.2006.884014]
-
[5] Garnett R, Huegerich T, Chui C, et al. A universal noise removal algorithm with an impulse detector[J]. IEEE Transactions on Image Processing, 2005, 14(11): 1747–1754. [DOI:10.1109/TIP.2005.857261]
-
[6] Dong Y Q, Chan R H, Xu S F. A detection statistic for random-valued impulse noise[J]. IEEE Transactions on Image Processing, 2007, 16(4): 1112–1120. [DOI:10.1109/TIP.2006.891348]
-
[7] Wu J, Tang C. A new filter for the removal of random-valued impulse noise from highly corrupted images[J]. AEU-International Journal of Electronics and Communications, 2012, 66(10): 847–854. [DOI:10.1016/j.aeue.2012.03.002]
-
[8] Wu J, Tang C. Random-valued impulse noise removal using fuzzy weighted non-local means[J]. Signal, Image and Video Processing, 2014, 8(2): 349–355. [DOI:10.1007/s11760-012-0297-1]
-
[9] Wang G J, Yang Y Q. Algorithm design of fuzzy filter based on salt-and-pepper noise with high probability in image restoration[J]. Acta Electronica Sinica, 2015, 43(1): 24–29. [王贵君, 杨永强. 基于高概率椒盐噪声的模糊滤波器在图像恢复中的算法设计[J]. 电子学报, 2015, 43(1): 24–29. ] [DOI:10.3969/j.issn.0372-2112.2015.01.005]
-
[10] Buades A, Coll B, Morel J M. A review of image denoising algorithms, with a new one[J]. Multiscale Modeling & Simulation, 2005, 4(2): 490–530. [DOI:10.1137/040616024]
-
[11] Gupta V, Chaurasia V, Shandilya M. Random-valued impulse noise removal using adaptive dual threshold median filter[J]. Journal of Visual Communication and Image Representation, 2015, 26: 296–304. [DOI:10.1016/j.jvcir.2014.10.004]
-
[12] Sengur A, Guo Y H. Color texture image segmentation based on neutrosophic set and wavelet transformation[J]. Computer Vision and Image Understanding, 2011, 115(8): 1134–1144. [DOI:10.1016/j.cviu.2011.04.001]
-
[13] Guo Y H, Cheng H D. New neutrosophic approach to image segmentation[J]. Pattern Recognition, 2009, 42(5): 587–595. [DOI:10.1016/j.patcog.2008.10.002]
-
[14] Hanbay K, Talu M F. Segmentation of SAR images using improved artificial bee colony algorithm and neutrosophic set[J]. Applied Soft Computing, 2014, 21: 433–443. [DOI:10.1016/j.asoc.2014.04.008]
-
[15] Guo Y H, Şengür A. A novel image segmentation algorithm based on neutrosophic similarity clustering[J]. Applied Soft Computing, 2014, 25: 391–398. [DOI:10.1016/j.asoc.2014.08.066]
-
[16] Zhang M, Zhang L, Cheng H D. Segmentation of ultrasound breast images based on a neutrosophic method[J]. Optical Engineering, 2010, 49(11): #117001. [DOI:10.1117/1.3505854]
-
[17] Guo Y H, Sengur A. A novel color image segmentation approach based on neutrosophic set and modified fuzzy c-means[J]. Circuits, Systems, and Signal Processing, 2013, 32(4): 1699–1723. [DOI:10.1007/s00034-012-9531-x]
-
[18] Guo Y H, Cheng H D, Zhang Y T. A new neutrosophic approach to image denoising[J]. New Mathematics and Natural Computation, 2009, 5(3): 653–662. [DOI:10.1142/S1793005709001490]
-
[19] Mohan J, Krishnaveni V, Guo Y H. MRI denoising using nonlocal neutrosophic set approach of Wiener filtering[J]. Biomedical Signal Processing and Control, 2013, 8(6): 779–791. [DOI:10.1016/j.bspc.2013.07.005]
-
[20] Qi X Y, Liu B Q, Xu J W. A novel algorithm for removing high-density salt-and-pepper noise based on fusion of indeterminacy information[J]. Acta Electronica Sinica, 2016, 44(4): 878–885. [齐现英, 刘伯强, 徐建伟. 基于不确定性信息融合的高密度椒盐噪声降噪方法[J]. 电子学报, 2016, 44(4): 878–885. ] [DOI:10.3969/j.issn.0372-2112.2016.04.018]
-
[21] Tomasi C, Manduchi R. Bilateral filtering for gray and color images[C]//Proceedings of the 6th International Conference on Computer Vision. Bombay, India, IEEE, 1998:839-846.[DOI:10.1109/ICCV.1998.710815]