CN104143177A - A Method of Eliminating Interference in Line Scan Camera Image - Google Patents

本发明涉及一种针对线阵相机在工业场合运用发现的噪声干扰消除问题，其采用专门的信噪比系数方法控制滤波强度，属于图像预处理阶段很重要的一环，克服了线阵相机在工业场合中遇到的噪声干扰问题，结合多种滤波方式可将图像中的纵向乘性噪声干扰及冲激噪声完全去除，通过计算图像信噪比大小选择适合图像处理的乘性噪声滤波系数及冲击噪声滤波系数，将纵向乘性噪声信号及冲激噪声信号滤除，彻底解决了线阵相机图像中产生的纵向噪声信号，对源图像信号影响小，噪声过滤效果强，为后期图像进一步处理提供更准确的图像源。本方法使用灵活度高，可调系数简单，软件实现简便，同时算法计算量小，对于处理单元要求低，尤其对于高速运动物体拍摄有很强的计算时间优势。The invention relates to a problem of eliminating noise interference found in the use of line array cameras in industrial occasions. It adopts a special signal-to-noise ratio coefficient method to control the filtering strength, which belongs to a very important part of the image preprocessing stage, and overcomes the problems of line array cameras in the industrial field. The problem of noise interference encountered in industrial occasions, combined with a variety of filtering methods can completely remove the longitudinal multiplicative noise interference and impulse noise in the image, and select the multiplicative noise filter coefficient suitable for image processing by calculating the image signal-to-noise ratio The impact noise filter coefficient filters out the longitudinal multiplicative noise signal and the impulse noise signal, completely solves the longitudinal noise signal generated in the image of the line array camera, has little influence on the source image signal, and has a strong noise filtering effect, which can be further processed for the later image Provide a more accurate image source. The method has high flexibility in use, simple adjustable coefficients, easy software implementation, and at the same time, the calculation amount of the algorithm is small, and the requirements for the processing unit are low, and it has a strong calculation time advantage especially for shooting high-speed moving objects.

一种消除线阵相机图像干扰的方法A Method of Eliminating Interference in Line Scan Camera Image

技术领域： Technical field:

本发明涉及一种消除线阵相机在工业场合运用发现的噪声干扰消除问题，特别适用于异纤分拣图象处理过程中利用信噪比来控制噪声滤波的方法。  The invention relates to a method for eliminating noise interference found in the use of line array cameras in industrial occasions, and is especially suitable for controlling noise filtering by using signal-to-noise ratio in the image processing process of foreign fiber sorting. the

背景技术： Background technique:

采集高速连续运动物体状态主要依靠线阵相机完成，在实际工作场合中线阵相机镜头由于长时间使用积灰或沾有杂物会影响实际图像采集状态，由于灰尘或其它杂质原因导致的纵向噪声干扰称为乘性噪声，在图像处理过程中会表现为纵向条纹，在二值化处理的过程中由于纵向噪声影响导致将纵向噪声信号处理至二值化图像中，破坏二值化图像处理，影响最后实际图像标定效果。由CCD传感器电磁干扰引起的点状噪声信号称为饱和或接近饱和的冲激噪声，在图像采集过程中图像根据CCD传感器质量或外部电路影响会出现一些大小不等白的或黑的像素点，破坏导致二值图像退化，影响最后实际图像标定效果。  Acquisition of the state of high-speed continuous moving objects is mainly completed by the line array camera. In the actual workplace, the actual image acquisition status will be affected by the long-term use of the line array camera lens due to dust or debris. Longitudinal noise interference caused by dust or other impurities It is called multiplicative noise, which will appear as vertical stripes in the process of image processing. In the process of binarization processing, due to the influence of longitudinal noise, the longitudinal noise signal will be processed into the binarized image, which will destroy the binarized image processing and affect The final actual image calibration effect. The point-like noise signal caused by the electromagnetic interference of the CCD sensor is called saturated or near-saturated impulse noise. During the image acquisition process, some white or black pixels of different sizes will appear in the image according to the quality of the CCD sensor or the influence of the external circuit. The destruction leads to the degradation of the binary image, which affects the final actual image calibration effect. the

线阵CCD传感器采集图像是由每次采集到的单行图像纵向累计叠加而成，所以当灰尘或杂物处于CCD传感器位置时会一直将信号累计至采集到的图像中，因此导致线阵相机图像中出现纵向纹理，这种纹理严重影响到曝光图像质量，同时线阵CCD传感器由于外部电气信号干扰会产生冲击噪声也会对图像质量有较大的影响。  The image collected by the line array CCD sensor is accumulated and superimposed by the single line of images collected each time, so when dust or debris is at the position of the CCD sensor, the signal will always be accumulated into the collected image, thus causing the line array camera image Longitudinal texture appears in the center, which seriously affects the quality of the exposed image. At the same time, the linear array CCD sensor will produce impact noise due to external electrical signal interference, which will also have a greater impact on the image quality. the

发明内容 Contents of the invention

本发明的目的是消除一种针对线阵相机在工业场合运用发现的噪声干扰消除问题，特别适用于异纤分拣图象处理利用信噪比来控制噪声滤波的方法。  The purpose of the present invention is to eliminate a noise interference elimination problem found in the use of line array cameras in industrial occasions, and is especially suitable for the method of controlling noise filtering by using signal-to-noise ratio in image processing of foreign fiber sorting. the

本发明提供的消除线阵相机图像干扰的方法，包括：  The method for eliminating line array camera image interference provided by the present invention includes:

1、对图像滤波强度进行调整，通过计算图像中信噪比值的大小可以得到量化的图像噪声强度值，以噪声强度为标准规定图像滤波强度，当噪声强度大的时候提高图像滤波系数以增强图像滤波效果，当噪声强度低的时候降低图像滤波系数减小图像滤波效果。  1. Adjust the image filtering intensity. By calculating the signal-to-noise ratio value in the image, the quantized image noise intensity value can be obtained. The image filtering intensity is specified based on the noise intensity. When the noise intensity is large, the image filtering coefficient is increased to enhance Image filtering effect, when the noise intensity is low, reduce the image filtering coefficient to reduce the image filtering effect. the

2、对于信噪比系数本方法采用分别计算图像像素点信号强度及噪声信号强度的方法来确定图像信噪比大小，为了准确确定图像信号强度及噪声强度，采用计算信号平方和的方法来确定信号强度大小。  2. For the signal-to-noise ratio coefficient, this method adopts the method of separately calculating the image pixel signal strength and noise signal strength to determine the image signal-to-noise ratio. In order to accurately determine the image signal strength and noise strength, the method of calculating the sum of signal squares is used to determine The size of the signal strength. the

3、通过计算出的信噪比值来确定平滑滤波过程中方形均值滤波器的大小。  3. The size of the square mean filter in the smoothing filtering process is determined by the calculated signal-to-noise ratio. the

4、设置乘性噪声的滤波器。  4. Set the filter for multiplicative noise. the

5、采用平滑滤波来处理图像中剩下的冲激噪声。  5. Use smoothing filter to deal with the remaining impulse noise in the image. the

本发明的有益之处在于：针对工业线阵相机在工业场合遇到实际问题而来，其采用专门的信噪比系数方法控制滤波强度，属于图像预处理阶段很重要的一环，克服了线阵相机在工业场合中遇到的噪声干扰问题，结合多种滤波方式可将图像中的纵向乘性噪声干扰及冲激噪声完全去除，通过计算图像信噪比大小选择适合图像处理的乘性噪声滤波系数及冲击噪声滤波系数，将纵向乘性噪声信号及冲激噪声信号滤除，彻底解决了线阵相机图像中产生的纵向噪声信号，对源图像信号影响小，噪声过滤效果强，为后期图像进一步处理提供更准确的图像源。本方法使用灵活度高，可调系数简单，软件实现简便，同时算法计算量小，对于处理单元要求低，尤其对于高速运动物体拍摄有很强的计算时间优势。另外本发明方法简单，该技术还具有操作简单、实时反馈调整计算方法的特点，适用于各种工业现场，实现平台多样，即可在嵌入式视觉系统中实现也可在工业控制计算机中实现。具有很强的实用性。可以大规模工业化应用。  The benefit of the present invention lies in that: in view of the practical problems encountered by industrial line scan cameras in industrial occasions, it adopts a special signal-to-noise ratio coefficient method to control the filtering strength, which belongs to a very important part of the image preprocessing stage and overcomes the linear The noise interference problem encountered by array cameras in industrial occasions, combined with a variety of filtering methods can completely remove the longitudinal multiplicative noise interference and impulse noise in the image, and select the multiplicative noise suitable for image processing by calculating the image signal-to-noise ratio The filter coefficient and the impact noise filter coefficient filter out the longitudinal multiplicative noise signal and the impulse noise signal, which completely solves the longitudinal noise signal generated in the image of the line array camera, has little influence on the source image signal, and has a strong noise filtering effect, which is the best choice for the later stage. Image further processing provides a more accurate image source. The method has high flexibility in use, simple adjustable coefficients, easy software implementation, and at the same time, the algorithm has a small amount of calculation, and has low requirements for processing units, and has a strong calculation time advantage especially for shooting high-speed moving objects. In addition, the method of the present invention is simple, and the technology also has the characteristics of simple operation and real-time feedback adjustment calculation method, is applicable to various industrial sites, and realizes various platforms, which can be implemented in embedded vision systems or in industrial control computers. Has a strong practicality. It can be applied in large-scale industrialization. the

具体实施方式 Detailed ways

本发明提供一种消除线阵相机图像干扰的方法，具体方法如下：  The present invention provides a method for eliminating image interference of a line array camera, the specific method is as follows:

设图像大小为x×y，假设图像信号中全部噪声的平方和表示为：  Let the image size be x×y, and assume that the sum of squares of all noise in the image signal is expressed as:

NN == ΣΣ (( xx ,, ythe y )) nno 22 (( xx ,, ythe y ))

同样假设图像信号中全部信号的平方和表示为：  Also assume that the sum of squares of all signals in the image signal is expressed as:

Ff == ΣΣ (( xx ,, ythe y )) ff 22 (( xx ,, ythe y ))

则得到信噪比的比值(以常数μ作为图像噪声强度的量化标准)为：  Then the ratio of signal-to-noise ratio (using the constant μ as the quantification standard of image noise intensity) is:

μμ == NN Ff

滤波后的图像信号函数为K(x，y)，线阵相机采集到的源图像信号为F(x，y)，噪声滤波器脉冲传递函数为H(x，y)，则滤波后的图像信号可以通过公式表示为：  The filtered image signal function is K(x, y), the source image signal collected by the line array camera is F(x, y), and the noise filter pulse transfer function is H(x, y), then the filtered image The signal can be expressed by the formula as:

KK (( xx ,, ythe y )) == Ff (( xx ,, ythe y )) ⊗⊗ Hh (( xx ,, ythe y ))

由于线阵相机采集图像都为二维图像，所以图像信号处理都采用二维傅里叶转换，源图像转换的传递函数可以用如下公式表示：  Since the images collected by the line array camera are all two-dimensional images, the image signal processing adopts two-dimensional Fourier transform, and the transfer function of the source image conversion can be expressed by the following formula:

▿▿ 22 Hh (( xx ,, ythe y )) == 11 22 ππ δδ 44 (( xx 22 ++ ythe y 22 δδ 22 -- 22 )) ee -- (( xx 22 ++ ythe y 22 δδ 22 ))

通过转换我们得到了图像单一像素坐标下的像素滤波效果。设源图像大小为M×N，通过扩展，将Q(μ，c)展开并将f(x，y)代入公式，则可得到完整图像信号的滤波转换传递函数公式为如下所示：  Through the conversion, we get the pixel filtering effect under the single pixel coordinate of the image. Suppose the size of the source image is M×N, expand Q(μ, c) and substitute f(x, y) into the formula through expansion, then the filter conversion transfer function formula of the complete image signal can be obtained as follows:

Hh (( αα ,, ββ )) == 11 MNMN ΣΣ mm == 00 Mm -- 11 ΣΣ nno == 00 NN -- 11 ff (( mm ,, nno )) expexp [[ -- 22 πiπi (( mαmα Mm ++ nβnβ NN )) ]]

其中α＝0，1，......，M-1，β＝0，1，......，N-1  where α=0, 1,..., M-1, β=0, 1,..., N-1

得到滤波转换传递函数后只需完成简单的卷积计算即可得到滤波完成后的图像信息。在图像处理过程中我们将滤波器传递函数等效为m×n的滤波器。对于大小为m×n的滤波器，在原图像顶部和底部至少填充m-1行0，在左侧和右侧至少填充n-1列0。这样整张待处理图像每个点都可以完整的被滤波器计算到。设p＝(m-1)/2，q＝(n-1)/2，为了表示方便，我们假设p和q都是奇整数。用公式表示如下所示：  After obtaining the filtering conversion transfer function, it only needs to complete simple convolution calculation to obtain the image information after filtering. In the process of image processing, we regard the filter transfer function as equivalent to an m×n filter. For a filter of size m×n, pad at least m-1 rows of 0s on the top and bottom of the original image, and at least n-1 columns of 0s on the left and right. In this way, each point of the entire image to be processed can be completely calculated by the filter. Let p=(m-1)/2, q=(n-1)/2, for the convenience of expression, we assume that both p and q are odd integers. The formula is expressed as follows:

uu (( xx ,, ythe y )) ** gg (( xx ,, ythe y )) == ΣΣ αα == -- pp pp ΣΣ ββ == -- qq qq uu (( αα ,, ββ )) gg (( xx -- αα ,, ythe y -- ββ ))

通过计算出的信噪比值来确定平滑滤波过程中方形均值滤波器的大小。设方形均值滤波器大小为r，则可以表示为如下公式：  The size of the square mean filter in the smoothing process is determined by the calculated SNR value. Assuming that the size of the square mean filter is r, it can be expressed as the following formula:

r＝μ*k*m  r＝μ*k*m

其中k称为滤波放大系数，m为滤波器矩阵最大值，一般滤波器矩阵最大值不超过9。  Among them, k is called the filter amplification factor, and m is the maximum value of the filter matrix. Generally, the maximum value of the filter matrix does not exceed 9. the

由于考虑到滤波算子设置可能会引起欠处理或者过处理，所以我们增加滤波算子控制系数来解决此问题：  Considering that the filter operator setting may cause under-processing or over-processing, we increase the filter operator control coefficient to solve this problem:

KK ×× 22 11 22 44 11 44 22 11 22

滤波算子控制系数大小由信噪比比值决定即K＝μ*k，将信噪比比值代入可得  The size of the filter operator control coefficient is determined by the signal-to-noise ratio, that is, K=μ*k, and the signal-to-noise ratio can be substituted into

(( μμ ** kk )) ×× 22 11 22 44 11 44 22 11 22

这样乘性噪声的滤波器就表示完成。  In this way, the filter of multiplicative noise is completed. the

乘性噪声被过滤后，采用平滑滤波来处理图像中剩下的冲激噪声，  After the multiplicative noise is filtered, the smoothing filter is used to deal with the remaining impulse noise in the image,

设p＝(m-1)/2，q＝(n-1)/2，为了表示方便，我们假设p和q都是奇整数。一张图像经过平滑滤波器的滤波过程可以由下式给出：  Let p=(m-1)/2, q=(n-1)/2, for the convenience of expression, we assume that both p and q are odd integers. The filtering process of an image through a smoothing filter can be given by the following formula:

gg (( xx ,, ythe y )) == ΣΣ αα == -- pp pp ΣΣ ββ == -- qq qq uu (( αα ,, ββ )) gg (( xx ++ αα ,, ythe y ++ ββ )) ΣΣ αα == -- pp pp ΣΣ ββ == -- qq qq uu (( αα ,, ββ ))

采用设3×3图像矩阵作为图像单像素点的表示方法，其中h0即为所需计算的图像单像素点，h1-h8为所需计算像素点周围的图像像素点，可以用如下矩阵表示：  A 3×3 image matrix is used as the representation method of a single pixel of the image, where h0 is the single pixel of the image to be calculated, and h1-h8 are the image pixels around the pixel to be calculated, which can be represented by the following matrix:

hh 44 hh 33 hh 22 hh 55 hh 00 hh 11 hh 66 hh 77 hh 88

平滑滤波器大小可根据信噪比的比值μ值大小进行更改，当μ值较大时代表图像中噪声信号较强，可以将平滑滤波器大小由3×3改为4×4直至N×N。实现方法则需将滤波器范围内所有像素点亮度求平均值即可。当图像信噪比μ值变大时或图像冲激噪声过多时，可以调整矩阵系数来扩大滤波范围可以将滤波矩阵大小由3×3改为4×4或者n×n，图像伪代码相应根据系数进行调整，如为4×4矩阵，只需将中间2×2个像素点求平均值作为中心点，其它计算方法与3×3矩阵计算方法相同，以此类推可以计算N×N矩阵。  The size of the smoothing filter can be changed according to the value of the signal-to-noise ratio μ. When the value of μ is larger, it means that the noise signal in the image is stronger. You can change the size of the smoothing filter from 3×3 to 4×4 until N×N . The implementation method needs to average the brightness of all pixels within the filter range. When the image signal-to-noise ratio μ value becomes larger or the image has too much impulse noise, the matrix coefficients can be adjusted to expand the filtering range. The size of the filter matrix can be changed from 3×3 to 4×4 or n×n, and the image pseudo code is correspondingly based on Adjust the coefficients. For example, if it is a 4×4 matrix, you only need to take the average value of the middle 2×2 pixels as the center point. The other calculation methods are the same as those for the 3×3 matrix, and N×N matrix can be calculated by analogy. the