融合提升小波阈值与多方向边缘检测的矿区遥感图像去噪
Denoising algorithm based on the fusion of lifting wavelet thresholding and multidirectional edge detection of remote sensing image of mining area
责任编辑: 张 仙
收稿日期: 2020-03-11 修回日期: 2020-06-3 网络出版日期: 2020-12-15
Received: 2020-03-11 Revised: 2020-06-3 Online: 2020-12-15
作者简介 About authors
王小兵(1988-),男,硕士,工程师,主要研究方向为数字图像处理、矿山测量和数字矿山等。Email:
遥感图像在矿区生态修复、地质灾害监测与防治等方面发挥了重要作用,但由于矿区环境复杂,导致获取的遥感图像存在噪声,在很大程度上影响了后续的图像解译与分析。为此,融合图像边缘检测与噪声抑制思路,提出了一种基于提升小波阈值(lifting wavelet thresholding,LWT)与改进Prewitt算子边缘检测(improved edge dectction of Prewitt operator,IEDPO)的矿区遥感图像去噪算法(LWT-IEDPO)。首先,对原始遥感图像进行提升小波变换,在保留低频小波子带不作处理的情况下,设计了一种双参数阈值函数模型对高频子带进行自适应噪声抑制,经过小波逆变换获得初步去噪后的遥感图像; 其次,为有效增强滤波后图像的细节信息,将经典Prewitt算子的检测模板扩展到0°,30°,60°,90°,120°和150°这6个方向,并设计了相应的检测结果融合规则,提出了改进的Prewitt算子来提取图像边缘轮廓,获得轮廓图像和非轮廓图像; 然后,为了进一步改善视觉效果,针对非轮廓图像采用改进的Pal-King模糊算法提升对比度; 最后,将增强后的非轮廓图像和轮廓图像进行叠加,实现对遥感图像的高清晰度复原。在MATLAB平台上,将所提出的遥感图像处理方法与经典硬阈值、软阈值模型以及2种已有的改进小波阈值算法进行对比,并引入峰值信噪比(peak signal to noise ratio, PSNR)和边缘保持指数(edge protection index, EPI)对各算法的噪声抑制性能进行定量分析和比较。研究表明所提方法能够有效实现遥感图像去噪,其总体性能优于其余4种算法。
关键词:
Remote sensing image plays an important role in ecological restoration, geological disaster monitoring and prevention of mining area; nevertheless, due to the complex environment of mining area, the obtained remote sensing images of the mining area contains different kinds of intensity noise, which affects the subsequent image interpretation and analysis to a great extent. In this paper, the study ideas of image edge detection and noise suppression are effectively fused together, and the improved denoising method (LWT-IEDPO) of remote sensing image in mining area based on the fusion of lifting wavelet thresholding (LWT) and improved edge dectction of Prewitt operator (IEDPO) is proposed. According to the basic principal of the new method proposed in this paper, firstly, lifting wavelet transform is done for the original remote sensing image; under the condition that the low-frequency wavelet sub-band is left untreated, a two-parameters thresholding function model is designed for adaptive noise suppression of high-frequency sub-bands, and the remote sensing image after initial denoising is obtained by the operation of inverse lifting wavelet transform. Secondly, for the purpose of effectively enhancing the details of the filtered remote sensing image, the detection template of the classical Prewitt operator is extended to 6 directions of 0°, 30°, 60°, 90°, 120° and 150°, and the corresponding detection results fusion rules are proposed. The improved Prewitt operator is put forward to extract the image edge information of the filtered image, and the edge and non-edge image are obtained. Then, the visual effect non-edge image is further improved by adopting the improved Pal-King fuzzy algorithm. Finally, the goal of high definition restoration of the original remote sensing image is realized by the superimposition of the enhanced non-edge image and edge image. Based on MATLAB language, the proposed remote sensing image processing method is compared with the classical hard thresholding model, soft thresholding model and two existing improved wavelet thresholding algorithms; in addition, the two indices of peak signal to noise ratio (PSNR) and edge protection index (EPI) are used to conduct quantitative analysis and comparison of the performance of the above algorithms. The study results show that the goal of effectively filtering of noise remote sensing image can be realized effectively, and the overall performance of the proposed algorithm is better than that of the other four algorithms.
Keywords:
本文引用格式
王小兵.
WANG Xiaobing.
0 引言
随着矿山生态日益受到重视,矿山遥感图像在矿区土壤监测、开采沉陷监测、地质灾害调查等方面得到了广泛应用,为矿区生态环境监测与治理提供了大量有价值的关键数据[1,2]。遥感传感器在采集矿区地物信息的过程中,极易受到矿区复杂成像环境(尤其是露天矿区空气中高浓度粉尘)的影响,导致获取的图像含有一定的噪声,视觉效果不佳。并且,在地物信息采集、传输等环节中,也不可避免地会混入一定的噪声,影响了后续对矿区遥感图像的准确判读。因而,有必要对遥感图像进行高质量去噪处理。对于图像去噪方面的研究,现阶段主要采用空间域滤波和变换域滤波2类思路进行图像噪声抑制,前者主要采用中值滤波、非局部均值滤波和人工神经网络等算法[3,4,5,6]对图像进行降噪处理; 后者主要通过对图像进行多尺度变换,在变换域对部分分解子带根据噪声分布特征,设定相应的噪声抑制模型进行处理,通过逆变换来复原图像,代表性的方法有小波变换[7,8,9]、复小波变换[10]和轮廓波变换[11,12]等。近年来,不少学者将上述2种思路进行有机结合,实现对遥感图像的去噪,如袁明月等[13]将中值滤波算法与小波变换相结合,利用高斯噪声来模拟遥感图像中含有的噪声,通过对经过中值滤波预处理后的图像进行二级小波变换,对于高频子带,通过逐点计算灰色关联度,并将其与经典小波阈值进行对比,来实现各高频子带的滤波,通过小波逆变换得到去噪后的遥感图像。该算法对于去除遥感图像噪声有一定的成效,但是在逐点计算灰色关联度等环节中,计算量较大,并且该项研究中小波阈值选取仍沿用传统方法,灵活性有所不足。
在结合已有成果[14,15,16,17,18]的基础上,引入提升小波变换[19,20]对遥感图像进行处理。本研究融合图像边缘检测与噪声抑制思路,提出了一种基于提升小波阈值(lifting wavelet thresholding,LWT)与改进Prewitt算子边缘检测(improved edge dectction of Prewitt operator,IEDPO)的矿区遥感图像去噪算法(LWT-IEDPO)。以遥感图像中常见的加性噪声为例,通过对遥感图像添加不同方差的高斯白噪声来模拟不同失真程度的遥感噪声图像,将图像进行多尺度提升小波分解,在小波域,针对传统硬阈值、软阈值函数模型的不足,设计了一种双参数改进型小波阈值函数模型处理高频子带中的噪声; 在实现遥感图像去噪后,为进一步提升图像质量,采用了一种改进的Prewitt算子进行6个方向边缘检测; 对于非边缘图像,采用改进型的Pal-King模糊算法进行增强处理,通过增强后的非边缘图像和边缘图像进行叠加,获取复原后的遥感图像。
1 算法原理
1.1 LWT函数模型
式中:
式中:
式中median(·)为取中值运算函数。
式(1)—(4)构成了本研究双阈值改进型阈值函数模型。该模型借鉴了经典硬阈值、软阈值去噪的思路,通过将传统硬阈值、软阈值模型进行有机结合,通过设置双阈值,将幅值较大的小波子带直接予以保留,最大限度地保留图像的细节信息。模型的阈值中,
1.2 基于改进Prewitt算子的图像边缘检测
本研究将经典的Prewitt算子的检测模板由0°和90°方向扩展到6个方向: 0°,30°,60°,90°,120°和150°方向,如图1所示。采用此6个方向模板对图像进行边缘信息检测后,需要对各个方向的检测信息进行有机融合,方可获得最终的检测结果。以图像中任一
图1
图1
扩展后的6方向Prewitt算子检测模板
Fig.1
Expanded detection template of the Prewitt operator with six directions
针对集合
1.3 改进Pal-King模糊增强算法
Pal-King算法用于图像增强处理的基本思路是: 首先构建模糊隶属映射函数将图像由空间域变换至模糊域,然后在模糊域中采用非线性函数进行增强处理,最后通过逆变换得到增强后的图像[13]。边缘轮廓信息是遥感图像中最为重要的一类信息,对于后续图像判读与分析至关重要。许健才[25]采用Pal-King算法对经过非下采样非轮廓波阈值去噪后的水果图像进行增强处理,取得了较好效果。但在该项研究中,Pal-King算法运算对象是整幅图像,对于小尺寸图像效果较好,但对于大尺寸遥感图像来说,运算耗时较长,并且在构建模糊变换函数中,未能有效顾及图像细节特征。这是因为,本研究是针对经过提升小波去噪后的遥感图像进行增强处理,经过去噪后的图像中,细节信息基本为图像中的有用信息,需要保留,否则会影响后续图像准确判读。为此,本研究对该算法进行改进: 一方面将增强运算范围限定在
首先将去噪后的遥感图像由空间域变换至模糊域,对变换后的图像进行
式中:
式中:
最后,对经过模糊增强后的图像进行逆变换,可得到最终增强后的图像,即
式中:
1.4 本文算法步骤
本文提出的融合提升小波阈值与多方向边缘检测的矿区遥感图像去噪算法(LWT-IEDPO)步骤为: ①对遥感噪声图像进行3层(式(1)中,
2 算法实验分析
2.1 实验一
图2
图3
通过分析图2和图3可知,小波硬阈值、软阈值对于2景遥感图像噪声的抑制效果均不理想,特别是软阈值去噪后图像中出现了很大程度的模糊(图2(d)和图3(d)),分析是由该函数模型通过小波幅值减去恒定的数值所致; 文献[22]和文献[7]分别提出的改进小波阈值方法对于失真程度较小的图像1(噪声方差为0.10)去噪效果较理想(图2(e)和图2(f)),但是图像2中噪声方差达到0.20时,两者的去噪性能有了大幅度下降(图3(e)和图3(f)); 本文方法(LWT-IEDPO算法)对于噪声方差为0.10的图像1,去噪效果(图2(g))与文献[7]和文献[22]所提算法较为接近,当图像2中的噪声方差提升至0.20时,去噪后图像的清晰度(图3(g))明显优于其余4种方法。
此外,本研究采用峰值信噪比(peak signal to noise ratio,PSNR)[26]以及边缘保持指数(edge protection index,EPI)[27]2个指标对上述5种方法的去噪性能进行定量评估,结果见表1。对于文中5种方法的PSNR和EPI指标值分析可知,当噪声方差逐步增加的过程中,5种方法的PSNR和EPI值均有所减小,反映出随着图像失真程度的逐步提升,各类方法对于噪声的抑制能力也有所下降,相对而言,LWT-IEDPO算法对应的PSNR和EPI指标值下降幅度最小; 对于方差为0.10,0.15以及0.20噪声图像,该算法对应指标值均优于其余4种方法,表明在5种方法中采用该算法复原后的图像质量与原始图像最为接近。
表1 算法去噪性能评价指标取值
Tab.1
遥感图像 | 高斯噪 声方差 | PSNR/dB | EPI | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
小波硬 阈值 | 小波软 阈值 | 文献[22] 方法 | 文献[7] 方法 | LWT- IEDPO算法 | 小波硬 阈值 | 小波软 阈值 | 文献[22] 方法 | 文献[7] 方法 | LWT- IEDPO算法 | ||
图像1 | 0.10 | 23.345 | 24.480 | 26.603 | 28.135 | 28.305 | 0.678 8 | 0.703 4 | 0.808 6 | 0.828 7 | 0.863 2 |
0.15 | 21.056 | 22.305 | 25.559 | 26.406 | 27.002 | 0.599 0 | 0.609 8 | 0.775 4 | 0.769 3 | 0.810 5 | |
0.20 | 19.807 | 20.209 | 20.698 | 23.790 | 25.329 | 0.506 7 | 0.598 8 | 0.620 1 | 0.670 4 | 0.755 6 | |
图像2 | 0.10 | 24.478 | 25.387 | 26.668 | 28.090 | 28.669 | 0.700 9 | 0.675 9 | 0.795 6 | 0.821 4 | 0.864 7 |
0.15 | 22.006 | 23.076 | 24.490 | 25.308 | 26.365 | 0.574 3 | 0.620 3 | 0.603 2 | 0.706 5 | 0.795 9 | |
0.20 | 18.565 | 19.690 | 21.212 | 22.210 | 24.497 | 0.489 6 | 0.556 9 | 0.573 3 | 0.601 2 | 0.703 1 |
2.2 实验二
图4
图4
某露天采场图像滤波效果对比
Fig.4
Comparison of the filtering results of an open-pit stope image
3 结论
实现对矿山遥感图像的高质量复原,有助于提升遥感图像信息判读与分析的准确性。针对矿山遥感图像在获取、传输等过程中易被噪声干扰的情形,融合提升小波阈值去噪与多方向边缘检测思路,提出了一种遥感图像改进去噪算法(LWT-IEDPO)。通过算法实验,该算法总体性能不仅优于传统的小波硬阈值、软阈值函数模型,相对于已有的2种改进型小波去噪算法,也有一定的优势。但是,算法在图像处理中还存在一定的过增强现象,后续工作还需要对算法中的模糊增强环节进行不断优化,确保在高效去除图像噪声的同时,进一步改善图像的视觉效果。
参考文献
矿山遥感图像自适应加权改进中值滤波算法
[J].
Adaptive weighted improved median filtering algorithm of mine remote sensing image
[J].
基于双边滤波与小波阈值法的矿区遥感图像处理
[J].
Remote sensing image processing method of mining area based on bilateral filtering algorithm and wavelet thresholding method
[J].
结合BM3D和多级非线性加权平均中值滤波的遥感影像混合噪声去噪方法
[J].
Mixed noise denoising method for remote sensing images combining BM3D and multi-level nonlinear weighted average median filtering
[J].
基于变差系数的SAR图像非局部均值滤波算法
[J].
Non-local mean filtering algorithm for SAR image based on coefficient of variation
[J].
基于小波变换与卷积神经网络的图像去噪算法
[J].
Image denoising algorithm based on wavelet transform and convolutional neural network
[J].
基于卷积神经网络的遥感图像去噪算法
[J].
Remote sensing image denoising based on convolutional neural network
[J].
小波阈值改进算法的遥感图像去噪
[J].
Remote sensing image denoising based on improved wavelet threshold algorithm
[J].
基于双重离散小波变换的遥感图像去噪算法
[J].
Remote sensing image de-noising algorithm based on double discrete wavelet transform
[J].
联合双边滤波器和小波阈值收缩去噪算法研究
[J].
Research on image denoising algorithm of joint bilateral filter and wavelet threshold shrinkage
[J].
结合NAPCA和复小波变换的高光谱遥感图像去噪
[J].
Denoising of hyperspectral remote sensing imagery using NAPCA and complex wavelet transform
[J].
基于NSCT和自适应模糊阈值遥感图像去噪算法
[J].
Remote sensing image denoising algorithm based on NSCT and adaptive fuzzy threshold
[J].
顾及纹理信息的遥感图像NSCT域自适应阈值去噪
[J].
Denoising of remote sensing images using adaptive threshold in NSCT domain by concerning texture information
[J].
一种改进的遥感影像去噪综合模型
[J].
An improved model of remote sensing image denoising
[J].
一种基于数学形态学与小波域增强的滤波算法
[J].
A new filter algorithm based on mathematical morphology and wavelet domain enhancement
[J].
基于小波变换的图像混合噪声自适应滤波算法
[J].
Adaptive filtering algorithm for mixed noise image based on wavelet transform
[J].
一种新型煤矿视频监控图像滤波算法
[J].
A new filtering algorithm for video monitoring image of coal mine
[J].
一种新型高斯噪声滤波算法
[J].
A new Gaussian noise filtering algorithm
[J].
一种高斯噪声组合滤波方法
[J].
A mixed filter method on Gaussian noise
[J].
提升小波变换的部分残差相位模拟与去除
[J].
Part of residual phase simulation and purify based on the lifting wavelet transform method
[J].
Region of interest extraction in remote sensing images by saliency analysis with the normal directional lifting wavelet transform
[J].DOI:10.1016/j.neucom.2015.11.093 URL [本文引用: 1]
基于LoG算子改进的自适应阈值小波去噪算法
[J].
DOI:10.3788/YJYXS20142902.0275
URL
[本文引用: 1]
图像在传输过程中会受到各种噪声干扰,为了实现消除噪声的目的,提出一种基于LoG算子改进的自适应阈值去噪算法。首先,利用LoG算子提取图像的边缘特征信息。接着,根据图像的边缘特征和非边缘特征分别求取改进的阈值函数:对于图像非边缘部分的阈值函数,在软阈值函数的基础上添加一个阈值修正系数,构建新的阈值函数;对于图像边缘部分的阈值函数,将边缘部分小波系数附近的能量和阈值相结合,构建新的阈值函数。最后利用改进的阈值函数对图像R、G、B 3个通道分别处理,保留图像所有的细节信息。实验结果表明,消噪后图像与含噪图像的PSNR值高于传统自适应算法12.09%;MAE值低于传统自适应算法22%。该算法有效保存了图像的边缘信息,综合去噪效果明显提高。
Improved self-adaptive threshold wavelet denoising analysis based on LoG operator
[J].
DOI:10.3788/YJYXS20142902.0275
URL
[本文引用: 1]
The image will be affected by various noise interference in the process.In order to eliminate the noise of the image,an improved self-adaptive threshold wavelet denoising analysis based on LoG operator was proposed.First,the image edge character information was extracted by LoG operator.Then,according to the edge of the image features and the no-edge of the image features,the improved threshold function was calculated respectively:on the basis of soft-thresholding function,a modified threshold coefficient was added for the threshold function in the image no-edge areas;in the no-edge areas, a new threshold function was built combining the energy near the wavelet coefficients with threshold. Finally,dealing with the red,green and blue(R,G,B) channels separately by the improved threshold function,all the details of the image were reserved.Experimental results show that PSNR of the image was above 12.09% compared with the traditional adaptive algorithm;MAE of the image was below 22% compared with the traditional adaptive algorithm.The image of this algorithm has a better quality than others,keeping the edge's information effectively, the integrative denoising is improved obviously.
基于改进阈值函数的自适应图像去噪方法
[J].
Adaptive image denoising based on the improved threshold function
[J].
Quantum image edge extraction based on improved Prewitt operator
[J].
A novel image edge detection algorithm based on Prewitt operator and wavelet transform
[J].
一种非下采样轮廓波变换域水果图像预处理方法
[J].
A pre-processing method for fruit image in non-subsampled contourlet transform domain
[J].
保留边界特征的深度图像增强算法研究
[J].
Depth image enhancement algorithm for preserving boundary
[J].
/
〈 |
|
〉 |
