CN105225244A - Based on the noise detection method that minimum local mean square deviation calculates - Google Patents

Abstract

The present invention relates to image processing field, carry out local according to local mean square deviation and the difference size of local mean square deviation removing itself in neighborhood of pixel points and detect for providing a kind of and judge that whether this point is the method for noise spot.The method can reduce the probability that non-noise point is mistaken for noise spot effectively.For this reason, the technical scheme that the present invention takes is, based on the noise detection method that minimum local mean square deviation calculates, adopt the pixel in certain neighborhood of pixel points, choose small neighbourhood respectively, OK, row, whether its local mean square deviation and local go the absolute value of the difference of heart mean square deviation and certain threshold value to compare certain pixel by several groups in two oblique five groups is that noise spot judges.The present invention is mainly used in image procossing.

Description

Based on the noise detection method that minimum local mean square deviation calculates

Technical field

The present invention relates to image processing field, particularly relating to when carrying out image denoising process, the differentiation for marginal point and noise judges to change problem.Specifically, the noise detection method calculated based on minimum local mean square deviation is related to.

Background technology

When carrying out images steganalysis with tracking, the image that video camera gathers, in imaging, digitizing and transmitting procedure, can be subject to the interference of various noise unavoidably, often there is unsatisfactory degeneration in the quality of image, have impact on the visual effect of image.Usually these noise make image degradation, and show as image blurring, feature is flooded, and it is unfavorable that this can produce graphical analysis, makes obtained picture quality lower.The recognition and tracking such image directly being carried out to target is more difficult.Suppress to make the various undesired signals of image degradation, the useful signal strengthened in image, and the different images observed is carried out under same constraint condition correction process and just seem extremely important.So process can be carried out by a series of modes such as level and smooth, filtering to the noise in image can improve picture quality.

The image de-noising method comparing main flow in recent years has: gaussian filtering, is applicable to filtering white Gaussian noise, has been widely used in the pretreatment stage of image procossing; Mean filter is also referred to as linear filtering, and the method can smoothed image, and speed is fast, and algorithm is simple.But cannot noise be removed, can only be faint weaken it; Median filtering method is a kind of nonlinear smoothing technology, in image procossing, medium filtering is commonly used to Protect edge information information, it is the method for classical smooth noise, the method is very effective to elimination spiced salt noise, in the phase analysis disposal route of optical measurement stripe pattern, have special role, but effect is little in fringe center analytical approach; Bilateral filtering a kind ofly can remove the wave filter of noise by preserving edge; The advantage of anisotropy parameter is that it while removal noise, can retain the marginal information even strengthened in image.This shows, for gaussian filtering, the filtering method of this kind of smoothed image of average, it is for noise, marginal point and other pixels to process be the same, the marginal information of this sampled images is just cut down; And for medium filtering, bilateral filtering and anisotropic filtering, they can retain the marginal information even strengthened in image, and simultaneously some noises equally also can strengthen by they, and under high intensity noise environment, noise removal capability declines to a great extent.

So for the differentiation of noise spot and marginal point and image denoising committed step when carrying out different disposal to it.

Summary of the invention

For overcoming the deficiencies in the prior art, providing a kind of and carrying out local according to local mean square deviation and the difference size of local mean square deviation removing itself in neighborhood of pixel points and detect and judge that whether this point is the method for noise spot.The method can reduce the probability that non-noise point is mistaken for noise spot effectively.For this reason, the technical scheme that the present invention takes is, based on the noise detection method that minimum local mean square deviation calculates, adopt the pixel in certain neighborhood of pixel points, choose small neighbourhood respectively, OK, row, whether its local mean square deviation and local go the absolute value of the difference of heart mean square deviation and certain threshold value to compare certain pixel to be that noise spot judges, specifically, certain pixel to be calculated to the local mean square deviation S of in this vertex neighborhood five groups by several groups in two oblique five groups _i' and neighborhood in remove this point after local remove heart mean square deviation S _i, the absolute value of both calculating | S _i'-S _i|, and get value S=min|S minimum in five groups of results _i'-S _i| judged, S=min|S _i'-S _i| large this point of expression has considerable influence to neighborhood gray scale, thinks noise spot, and adopt minimum value can reduce probability marginal point being mistaken for noise spot, at different conditions, user also according to circumstances can make free burial ground for the destitute by oneself and select several groups of data to carry out choosing of S.

Determine that its threshold step is according to the partial statistics characteristic in different neighborhood is adaptive, when the S value in local neighborhood is greater than this threshold value, judge that it is noise spot, when being less than this threshold value, being judged as non-noise point, being expressed as:

S＝min|S _i'-S _i|(1)

$S_0=\frac{{10}^{\frac{254*\ln S}{\stackrel{\‾}{S_i}}}}{2}---\left(2\right)$

S ₀threshold value selected by this method; Work as S _i' >S ₀time, this pixel is judged as noise spot; Work as S _i'≤S ₀time, this pixel is judged as non-noise point; Known according to formula (2), when time very large, the mean square deviation that namely in this neighborhood, five groups are not removed the heart is very large, and this point is in unsmooth region, and so need more strict to the judgement of this point, selected threshold value is less than the threshold value at smooth region, i.e. S ₀value is little.

Feature of the present invention and beneficial effect are:

By the absolute value choosing mean square deviation difference minimum in five groups in neighborhood, the present invention judges whether certain point is noise spot, and reduce the probability only with point several in certain neighborhood, marginal point being mistaken for noise spot, it is more accurate to make the judgement of noise spot.

The present invention adopt adaptive threshold can according to the partial statistics characteristic in neighborhood adaptive adjustment threshold size, to non-smooth areas adopt less threshold value, in smooth region point adopt larger threshold value, can judge noise spot more accurately like this.

Accompanying drawing illustrates:

Fig. 1 local neighborhood mean square deviation judges noise schematic diagram.

Embodiment

Technical solution of the present invention as shown in Figure 1, the present invention adopts the pixel in certain neighborhood of pixel points, choose small neighbourhood respectively, OK, row, whether its local mean square deviation and local go the absolute value of the difference of heart mean square deviation and certain threshold value to compare certain pixel by several groups in two oblique five groups is that noise spot judges.This is because consider that noise is the random comparatively independently point produced, all comparatively large in the grey scale change of all directions, remove the neighborhood region after this noise spot then comparatively level and smooth, grey scale change is milder; Meanwhile the edge of image is generally one section of continuous curve, remove on edge line a bit little on the grey scale change impact on certain direction of this neighborhood in neighborhood region, and this direction of curve the unknown, therefore these multiple directions are chosen, can judge that certain point is noise spot or marginal point more accurately.Thus, the present invention calculates the local mean square deviation S of in this vertex neighborhood five groups to certain pixel _i' and neighborhood in remove this point after local remove heart mean square deviation S _i, the absolute value of both calculating | S _i'-S _i|, and get value S=min|S minimum in five groups of results _i'-S _i| judged, S=min|S _i'-S _i| large this point of expression has considerable influence to neighborhood gray scale, thinks noise spot, and adopt minimum value can reduce probability marginal point being mistaken for noise spot, at different conditions, user also according to circumstances can make free burial ground for the destitute by oneself and select several groups of data to carry out choosing of S.

This noise detection method calculated based on Minimum Mean Square Error in local neighborhood can judge that certain pixel is pixel in noise spot or smooth region and marginal point comparatively accurately, in filtering afterwards also can according to the judgement of pixel to its select different smoothing methods with reach remove noise while the object of preserving edge information.

The present invention adaptively according to the partial statistics characteristic in different neighborhood can also determine its threshold value, when the S value in local neighborhood is greater than this threshold value, judges that it is noise spot, and when being less than this threshold value, be judged as non-noise point, method of the present invention can be expressed as:

S＝min|S _i'-S _i|(1)

$S_0=\frac{{10}^{\frac{254*\ln S}{\stackrel{\‾}{S_i}}}}{2}---\left(2\right)$

S ₀threshold value selected by this method.Work as S _i' >S ₀time, this pixel is judged as noise spot; Work as S _i'≤S ₀time, this pixel is judged as non-noise point.Known according to formula (2), when time very large, the mean square deviation that namely in this neighborhood, five groups are not removed the heart is very large, and this point is in unsmooth region, and so need more strict to the judgement of this point, selected threshold value is less than the threshold value at smooth region, i.e. S ₀value is little.

The present invention is further described below in conjunction with the drawings and specific embodiments.

As shown in Figure 1, what provide be 9 × 9 neighborhoods is example, in different situations, user can carry out the neighborhood that the size of image of noise spot detection and the difference of contrast choose different size according to actual needs, picture size larger in order to save noise spot judge the time can select similar 5 × 5 comparatively small neighbourhood, and the accuracy in order to judge, increase neighborhood size that again can be appropriate, such as 9 × 9 neighborhoods, the same selection for Ji Zu local data also can be carried out self-defined choosing according to different characteristics of image and adapt to different images environmental demand, reach and judge more accurately.

Claims (2)

1. the noise detection method calculated based on minimum local mean square deviation, it is characterized in that, adopt the pixel in certain neighborhood of pixel points, choose small neighbourhood respectively, OK, row, whether its local mean square deviation and local go the absolute value of the difference of heart mean square deviation and certain threshold value to compare certain pixel by several groups in two oblique five groups is that noise spot judges, specifically, certain pixel is calculated to the local mean square deviation S of in this vertex neighborhood five groups _i' and neighborhood in remove this point after local remove heart mean square deviation S _i, the absolute value of both calculating | S _i'-S _i|, and get value S=min|S minimum in five groups of results _i'-S _i| judged, S=min|S _i'-S _i| large this point of expression has considerable influence to neighborhood gray scale, thinks noise spot, and adopt minimum value can reduce probability marginal point being mistaken for noise spot, at different conditions, user also according to circumstances can make free burial ground for the destitute by oneself and select several groups of data to carry out choosing of S.

2. as claimed in claim 1 based on the noise detection method that minimum local mean square deviation calculates, it is characterized in that, determine that its threshold step is according to the partial statistics characteristic in different neighborhood is adaptive, when the S value in local neighborhood is greater than this threshold value, judge that it is noise spot, when being less than this threshold value, being judged as non-noise point, being expressed as:

S＝min|S _i'-S _i|(1)

$S_0=\frac{{10}^{\frac{254*\ln S}{\stackrel{\‾}{S_i}}}}{2}---\left(2\right)$

S ₀threshold value selected by this method; Work as S _i' >S ₀time, this pixel is judged as noise spot; Work as S _i'≤S ₀time, this pixel is judged as non-noise point; Known according to formula (2), when time very large, the mean square deviation that namely in this neighborhood, five groups are not removed the heart is very large, and this point is in unsmooth region, and so need more strict to the judgement of this point, selected threshold value is less than the threshold value at smooth region, i.e. S ₀value is little.