CN104143177A - Method for eliminating image interference of line-scan digital camera - Google Patents

Abstract

The invention relates to a method for eliminating image interference of a line-scan digital camera. According to the method, in order to eliminate noise interference found in the application process of the line-scan digital camera in the industrial occasion, a specific signal-to-noise coefficient method is adopted for controlling filtering intensity, which is a very important link in the image preprocessing stage, the noise interference problem, encountered in the industrial occasion, of the line-scan digital camera is solved, in combination with multiple filtering modes, longitudinal multiplicative noise interference and impulse noise in an image can be completely eliminated, and through calculating the signal-to-noise ratio of the image, a multiplicative noise filtering coefficient and an impulse noise filtering coefficient suitable for image processing are selected. Because a longitudinal multiplicative noise signal and an impulse noise signal are filtered out, a longitudinal noise signal generated in the image of the line-scan digital camera is thoroughly eliminated, influences on a source image signal are small, the noise filtering effect is high, and a more accurate image source is provided for further processing the image at the later period. The method is high in flexibility and simple in adjustable coefficient, software is easy and convenient to obtain, meanwhile, the calculation amount is small according to the algorithm, the requirement for a processing unit is low, and the method particularly has a very strong advantage in calculation time when a high-speed motion object is shot.

Description

A kind of method of eliminating linear array camera image interference

Technical field:

The present invention relates to a kind of line-scan digital camera of eliminating and use the noise of finding to eliminate problem in industrial occasions, be specially adapted to utilize noise recently to control the method for noise filtering in different fine sorting image processing procedure.

Background technology:

Gathering high-speed and continuous moving body state mainly relies on line-scan digital camera to complete, in real work occasion, line-scan digital camera camera lens can affect actual image acquisition state owing to using for a long time dust stratification or speckling with foreign material, the longitudinal noise causing due to dust or other impurity reason is called multiplicative noise, in image processing process, can show as longitudinal stripe, in the process of binary conversion treatment because longitudinal noise effect causes longitudinal noise signal to be processed to binary image, destruction binary image is processed, and affects last real image and demarcates effect.The spotted noise signal being caused by ccd sensor electromagnetic interference (EMI) is called saturated or approaches saturated impulse noise, in image acquisition process, image there will be some white or black pixels that differ in size according to ccd sensor quality or external circuit impact, destruction causes bianry image to be degenerated, and affects last real image and demarcates effect.

It is longitudinally to be added up to be formed by stacking by the unicursal graph picture collecting at every turn that linear CCD sensor gathers image, so when dust or foreign material can be always during in ccd sensor position by signal accumulative total to the image collecting, therefore cause occurring in linear array camera image longitudinal texture, this texture badly influences exposure image quality, and isochrone array ccd sensor also can have larger impact to picture quality because the interference of external electrical signal can produce impact noise.

Summary of the invention

The object of the invention is to eliminate a kind of noise elimination problem of finding of using in industrial occasions for line-scan digital camera, be specially adapted to different fine sorting image processing and utilize noise recently to control the method for noise filtering.

The method that elimination linear array camera image provided by the invention disturbs, comprising:

1, image filtering intensity is adjusted, size by snr value in computed image can obtain the image noise intensity value quantizing, take noise intensity as standard regulation image filtering intensity, when noise intensity is large, improve image filtering coefficient to strengthen image filtering effect, when noise intensity is low, reduce image filtering coefficient and reduce image filtering effect.

2, for signal to noise ratio (S/N ratio) coefficient this method, adopt the method for computed image pixel signal intensity and noise signal strength respectively to determine signal noise ratio (snr) of image size, in order accurately to determine image intensity signal and noise intensity, adopt the method for calculating signal quadratic sum, determine signal intensity size.

3, by the snr value calculating, determine the size of square mean filter in smothing filtering process.

4, the wave filter of multiplicative noise is set.

5, adopt smothing filtering to process impulse noise remaining in image.

Usefulness of the present invention is: for industrial line-scan digital camera, in industrial occasions, run into practical problems, it adopts special signal to noise ratio (S/N ratio) coefficient method to control filtering strength, belong to the very important ring of image pretreatment stage, overcome the noise problem that line-scan digital camera runs in industrial occasions, in conjunction with multiple filtering mode, longitudinal multiplicative noise interference and impulse noise in image can be removed completely, by computed image signal to noise ratio (S/N ratio) size, select to be applicable to multiplicative noise filter factor and the impact noise filter factor that image is processed, by longitudinal multiplicative noise signal and impulse noise target signal filter, thoroughly solved the longitudinal noise signal producing in linear array camera image, little to source images effect of signals, noise filtering effect is strong, for later image is further processed image source is more accurately provided.This method is used flexibility ratio high, and adjustability coefficients is simple, and software is realized easy, and algorithm calculated amount is little simultaneously, low for processing unit requirement, especially for high-speed moving object, takes and has very strong advantage computing time.The inventive method is simple in addition, and the feature that this technology also has is simple to operate, Real-time Feedback is adjusted computing method, is applicable to various industry spot, and implementation platform is various, can in embedded vision system, realize also and can in industrial control computer, realize.There is very strong practicality.Can heavy industrialization application.

Embodiment

The invention provides a kind of method that linear array camera image disturbs of eliminating, concrete grammar is as follows:

If image size is x * y, suppose that the quadratic sum of whole noises in picture signal is expressed as:

$N=\underset{(x,y)}{\mathrm{\Σ}}{n}^2(x,y)$

In same hypothesis picture signal, all the quadratic sum of signal is expressed as:

$F=\underset{(x,y)}{\mathrm{\Σ}}{f}^2(x,y)$

The ratio (take the quantitative criteria of constant μ as image noise intensity) that obtains signal to noise ratio (S/N ratio) as:

$\mathrm{\μ}=\frac{N}{F}$

Filtered picture signal function is K (x, y), and the source images signal that line-scan digital camera collects is F (x, y), and noise filter pulsed transfer function is H (x, y), and filtered picture signal can be shown by formula table:

$K(x,y)=F(x,y)\⊗H(x,y)$

Because line-scan digital camera gathers image, be all two dimensional image, so picture signal is processed, all adopt two-dimentional Fourier transform, the transport function of source images conversion can represent with following formula:

${\▿}^{2}H(x,y)=\frac{1}{2\mathrm{\π}{\mathrm{\δ}}^4}(\frac{x^2+y^2}{{\mathrm{\δ}}^2}-2)e^{-\left(\frac{x^2+y^2}{{\mathrm{\δ}}^2}\right)}$

By changing us, obtained the pixel filter effect under the single pixel coordinate of image.If source images size is M * N, by expansion, by Q (μ, c) expansion and by f (x, y) substitution formula, the filtering that can obtain complete image signal is changed transport function formula into as follows:

$H(\mathrm{\α},\mathrm{\β})=\frac{1}{\mathrm{MN}}\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}f(m,n)\exp [-2\mathrm{\πi}(\frac{\mathrm{m\α}}{M}+\frac{\mathrm{n\β}}{N})]$

α=0 wherein, 1 ..., M-1, β=0,1 ..., N-1

After obtaining filtering conversion transport function, only need complete simple convolutional calculation can obtain the image information after filtering completes.In image processing process, we are equivalent to filter transfer function in the wave filter of m * n.The wave filter that is m * n for size, at least fills m-1 capable 0 in original image top and bottom, in left side and right side, at least fill n-1 row 0.Whole like this each point of pending image can be complete filtered device calculate.If p=(m-1)/2, q=(n-1)/2, in order to represent that conveniently we suppose that p and q are odd-integral numbers.Be formulated as follows:

$u(x,y)*g(x,y)=\underset{\mathrm{\α}=-p}{\overset{p}{\mathrm{\Σ}}}\underset{\mathrm{\β}=-q}{\overset{q}{\mathrm{\Σ}}}u(\mathrm{\α},\mathrm{\β})g(x-\mathrm{\α},y-\mathrm{\β})$

By the snr value calculating, determine the size of square mean filter in smothing filtering process.If square mean filter size is r, can be expressed as formula:

r＝μ*k*m

Wherein k is called filter and amplification coefficient, and m is electric-wave filter matrix maximal value, and general electric-wave filter matrix maximal value is no more than 9.

Owing to considering that filter operator setting may cause undercuring or cross and process, and solves this problem so we increase filter operator control coefrficient:

$K\×\left[\begin{array}{ccc}2& 1& 2\\ 4& 1& 4\\ 2& 1& 2\end{array}\right]$

Filter operator control coefrficient size determines to be K=μ * k by signal to noise ratio (S/N ratio) ratio, and the substitution of signal to noise ratio (S/N ratio) ratio can be obtained

$(\mathrm{\μ}*k)\×\left[\begin{array}{ccc}2& 1& 2\\ 4& 1& 4\\ 2& 1& 2\end{array}\right]$

The wave filter of multiplicative noise has just represented like this.

After multiplicative noise is filtered, adopt smothing filtering to process impulse noise remaining in image,

If p=(m-1)/2, q=(n-1)/2, in order to represent that conveniently we suppose that p and q are odd-integral numbers.The filtering of an image process smoothing filter can be provided by following formula:

$g(x,y)=\frac{{\mathrm{\Σ}}_{\mathrm{\α}=-p}^p{\mathrm{\Σ}}_{\mathrm{\β}=-q}^{q}u(\mathrm{\α},\mathrm{\β})g(x+\mathrm{\α},y+\mathrm{\β})}{{\mathrm{\Σ}}_{\mathrm{\α}=-p}^p{\mathrm{\Σ}}_{\mathrm{\β}=-q}^{q}u(\mathrm{\α},\mathrm{\β})}$

Employing is established 3 * 3 image arrays as the method for expressing of image list pixel, and wherein h0 is the image list pixel of required calculating, and h1-h8 is that required calculating pixel is selected image slices vegetarian refreshments around, can use following matrix representation:

$\left[\begin{array}{ccc}h4& h3& h2\\ h5& h0& h1\\ h6& h7& h8\end{array}\right]$

The large I of smoothing filter is changed according to the ratio μ value size of signal to noise ratio (S/N ratio), when the larger interval scale noise in image of μ value signal is stronger, can by 3 * 3, change smoothing filter size into 4 * 4 until N * N.Implementation method needs all pixel brightness within the scope of wave filter to average.When signal noise ratio (snr) of image μ value becomes large or when image impulse noise is too much, can adjust matrix coefficient expands filter range and can by 3 * 3, change filtering matrix size into 4 * 4 or n * n, image false code is corresponding to be adjusted according to coefficient, as be 4 * 4 matrixes, only middle 2 * 2 pixels need be averaged as central point, other computing method are identical with 3 * 3 matrix computations methods, can calculate N * N matrix by that analogy.

Claims (4)

1. for line-scan digital camera, in industrial occasions, use the noise of finding to eliminate a problem, specifically refer to a kind of method of utilizing noise recently to control noise filtering.

2. according to the method described in right 1, it is characterized in that, described method adopts the method for adding signal to noise ratio (S/N ratio) coefficient to come control chart as the intensity of filtering signal, guarantees the least possible change of source images signal data.For signal to noise ratio (S/N ratio) coefficient this method, adopt the method for computed image pixel signal intensity and noise signal strength respectively to determine signal noise ratio (snr) of image size, in order accurately to determine image intensity signal and noise intensity, adopt the method for calculating signal quadratic sum to determine signal intensity size, filter multiplicative noise.

3. according to the method described in right 1, it is characterized in that, after described multiplicative noise is filtered, adopt smothing filtering to process impulse noise remaining in image, the method that solves impulse noise mainly contains two kinds, a kind of is to adopt mean filter to filter signal, and another adopts pixel size judgement in conjunction with the average method of pixel, and determination methods can be adjusted according to user's needs or real image noise situations.

4. according to the method described in right 1, it is characterized in that, described method judges that by pixel intensity wave filter carries out filtering processing to picture point, pixel judgement wave filter mainly judges the difference size of surrounding pixel and central pixel point, due to actual acquisition to picture signal generally can there is not unexpected conversion, brightness transition is also stepping, so cross bright or cross when dark and substantially can judge that this pixel is noise pixel point when single pixel intensity, can reduce the impulse noise signal in image by furthering with the method for surrounding pixel point brightness.