CN103996177A - Snow noise removing algorithm free of reference detection - Google Patents

Abstract

Provided is a snow noise removing algorithm free of reference detection. First, pixel points adjacent in position and similar in gray scale (color) are classified into a block, and isolated blocks with the area smaller than a threshold value are judged to be noise points; the noise points are filtered, and non-noise points are directly output; when the number of signal points within a 7*7 neighborhood region around the noise points is zero, the noise points are not processed, and otherwise, a gray scale self-adaptation weighting average value of the signal points in the neighborhood region serves as a new gray scale value of the pixel points. The deviation between the filtering result and the original image pixel gray scale is subjected to statistics, snow noise is detected, when the deviation is larger than a certain value, it is thought that snow noise exists, the filtering result is output, otherwise, it is thought that no snow noise exists, and an original image is output.

Description

A kind of without eliminating snow noise algorithm with reference to detecting

Affiliated technical field

The invention belongs to technical field of image processing, be specially a kind of algorithm that detects and eliminate snow noise in video.

Background technology

Snow noise is to have compared with the isolated point of high-gray level poor (color difference) with background around.Video camera is adopted to such an extent that image has five kinds containing the main cause of snow noise: (1) circuit noise is larger, (2) power frequency is disturbed, (3) high frequency interference, (4) enlargement factor is larger, and (5) magnetic head needs to clean.

Circuit noise is made up of thermonoise and shot noise.Thermonoise is mainly to be produced by the random thermal motion of free electron of electric conductor inside.Shot noise is to be produced by the semiconductor in the active devices such as amplifier.When shooting end and watch-dog end are simultaneously when ground connection, due to the existence of earth resistance and cable sheath resistance, between two places, the each phase load imbalance of electric system or earthing mode difference cause potential difference (PD), disturb thereby produce power frequency.

Disturbing for snow is due to due to signal attenuation on transmission line and the high frequency interference that has been coupled, this interference ratio is easier to eliminate, between video camera and gating matrix, rational position increases a video amplifier, the signal to noise ratio (S/N ratio) of signal is improved, or high frequency interference source is avoided in the path that changes vision cable, and the problem of high frequency interference can be resolved substantially.The another kind of snow noise method of eliminating is the method with software.In conventional elimination snow noise software approach, there is medium filtering, the methods such as average value filtering.But these methods can not detect snow noise and often can not effectively eliminate snow noise.

Because salt-pepper noise is the one of snow noise, the removing method of salt-pepper noise has reference value to the elimination of snow noise very much.For high-intensity salt-pepper noise, the median filtering algorithm of standard is suppressing there is contradiction aspect picture noise and protection details two.For the salt-pepper noise in image, a lot of articles have proposed filtering method.(the Li Xiaohong such as Li Xiaohong, Jiang Jianguo, Wu Congzhong, Zhan Shu. image goes the research [J] of salt-pepper noise wave filter. Journal of Engineering Graphics, 2009,6) propose and design a kind of multiwindow adaptive median filter (MWMF wave filter) that cross window and fork-shaped window are organically combined, can, according to the shape of image in window self, select adaptively cross window and fork-shaped window.After the relative merits of many window adaptive median filters and these two kinds of wave filters of shape filtering, also propose and designed nonlinear combination wave filter.But nonlinear combination wave filter is the simple addition of two kinds of methods.Wen Shengqiang etc. (Wen Shengqiang, Fu Minglan, Chen Lvqiang. based on the greatly relevant alternative salt-pepper noise filtering algorithm [J] of pixel value. scientific and technical information, 2010,19) first noise is differentiated, and centered by this pixel, 3 × 3 neighborhood territory pixels are sorted, replace this pixel value with intermediate value.Then intermediate value and surrounding pixel are compared, obtain and a pixel value of its absolute difference minimum, finally replace intermediate value with this grey scale pixel value.But the pixel of this kind of method improvement is limited.(the Liu Mingkun such as Liu Mingkun; Long Yi. the salt-pepper noise self-adaptation filtering algorithm [J] based on relevance predication. software guide; 2010.6; 9 (6)) analyzing on the basis of existing details protection filtering algorithm, a kind of image salt-pepper noise self-adaptation filtering algorithm based on relevance predication has been proposed.For signal pixels, keep gray-scale value constant.For noise suspicion pixel, utilize the neighborhood gray scale related function of definition as the tolerance of signal neighborhood relevance, and coefficient using this tolerance as predictive filtering algorithm.But in article, the definition of the degree of correlation is too simple.(the Kong Fanzhen such as Kong Fanzhen, Li Zhaoyuan. a kind of improvement [J] of switching median filter device. Taiyuan science and technology, 2009,7) by judging the self-adaptive processing of extreme point, corrected impulse walkaway parameter r and the threshold value to r, a kind of adaptive threshold switching median filter device based on extreme value has been proposed.But article is not considered noise and is positioned at the situation of corner point.

Divide from method of discrimination, the evaluation of picture quality can be divided into subjective method of discrimination and objective method of discrimination.Subjective method is wasted time and energy, and is not suitable for the occasion that multiple video cameras detect.Whether traditional objective image quality evaluating method can be divided into three kinds according to the existence of the original image for reference.First method is " full reference " (full-reference), second method is called as " without reference " (non-reference) or blind (blind) image quality evaluation, and the third method is called as the evaluation method of " reducing reference " (reduced-reference).Snowflake in the present invention detects differentiates the objective method of discrimination belonging to without reference.

Summary of the invention

In order to detect and eliminate snow noise, the present invention first position is the adjacent and close pixel of gray scale (color) is classified as one, judge noise spot by decision block size, non-noise spot initial value output, and the point self-adapted average of the number of winning the confidence is replaced snow noise point, and relatively filtering result and former figure gray-scale deviation detect snow noise.Concrete steps are:

1. input picture f, and image is divided into sub-block;

2. by the screening of sub-block is formed to the alternative collection of noise piece;

3. alternative concentrated each of pair noise piece is taked auto adapted filtering, generates the estimated image of a width picture rich in detail;

4. judged whether snow noise by the deviation of statistical estimate image and original image, if judgement contains snow noise, output filtering image, otherwise export former figure.

In step 1, the method that image is divided into sub-block is: traversing graph picture, the neighbours territory of each pixel is analyzed, if the absolute value of the difference of two pixel values is less than T ₁think that two pixels are close, and be divided in a piece, form a block of pixels.

In step 1, the method that forms the alternative collection of noise piece by the screening to sub-block is:

1) first analyze the block of pixels that number of pixels is less than 6, if the sum of all pixels of this class block of pixels is greater than T divided by the ratio of total number of image pixels ₂, the piece of dividing this type is the alternative collection of noise piece, otherwise to be less than 11 piece be the alternative collection of noise piece to number of pixels in divided block;

2) two pieces that locus in alternative collection are connected merge, form new alternative collection, if step 2) in alternative collection be less than 6 piece by number of pixels and form, cast out newly concentrating number of pixels to be greater than 6 piece, otherwise cast out newly concentrating number of pixels to be greater than 11 piece;

3) using step 2) in the new collection that obtains as the final alternative collection of noise piece, prepare for subsequent step does data.

In step 2, take the concrete grammar of auto adapted filtering to be to the alternative concentrated sub-block of noise piece: first to define the alternative pixel in addition that integrates of noise piece as signaling point, then each pixel 3 × 3 neighborhoods around in decision block successively, in 5 × 5 neighborhoods and 7 × 7 neighborhoods, whether the number of signaling point is 0, for example, in if block, in pixel 3 × 3 neighborhoods, the number of signaling point is 0, judge the number of signaling point in 5 × 5 neighborhoods, the like, until judge that the number of signaling point in 7 × 7 neighborhoods is still 0, detect next piece, otherwise, if in the process detecting, signaling point number is not 0, get the weighted mean of all signaling point gray scales in current neighborhood as the new gray-scale value of this pixel, generate the estimated image of picture rich in detail concrete formula is as follows:

$\hat{f}(x_0,y_0)=\underset{(x,y)\∈S}{\mathrm{\Σ}}f(x,y)w_{(x,y)\→(x_0,y_0)}---\left(1\right)$

Wherein S is signaling point set in 3 × 3 or 5 × 5 or 7 × 7 neighborhoods, the estimation of picture rich in detail, for point (x ₀, y ₀) to the weights of point (x, y), computing method are as follows:

$w_{(x,y)\→(x_0,y_0)}=\frac{d_{(x,y)\→(x_0,y_0)}}{\underset{(p,q)\∈S}{\mathrm{\Σ}}{d}_{(p,q)\→(x_0,y_0)}}---\left(2\right)$

Wherein that point (p, q) is to point (x ₀, y ₀) distance:

$d_{(p,q)\→(x_0,y_0)}=\frac{1}{\sqrt{{(p-x_0)}^2+{(q-y_0)}^2}}.---\left(3\right)$

In step 3, judge whether that by the deviation of statistical estimate image and original image the concrete grammar of snow noise is, obtain the deviation of two images by formula (4):

$\mathrm{diff}=\frac{\underset{(x,y)\∈N}{\mathrm{\Σ}}|f(x,y)-\hat{f}(x,y)|}{H\×W}---\left(4\right)$

Wherein N is the alternative set of concentrating all pixels of noise piece, and H is picture altitude, and W is picture traverse, if deviation diff is greater than T ₃, judge that original image contains snow noise, the estimation of output picture rich in detail , otherwise think there is no snow noise, export former figure f.

Brief description of the drawings

Fig. 1 is the process flow diagram that a kind of snow noise detects elimination algorithm.

Fig. 2 is band noise image.

Fig. 3 is noise spot distribution plan.

Fig. 4 is filtering result.

Embodiment

As shown in Figure 1, in order to detect and eliminate snow noise, the present invention first position is the adjacent and close pixel of gray scale (color) is classified as one, judge noise spot by decision block size, non-noise spot initial value output, and the point self-adapted average of the number of winning the confidence is replaced snow noise point, and relatively filtering result and former figure gray-scale deviation detect snow noise.Taking Fig. 2 as example, containing noisy image, apply the detection elimination algorithm of a kind of nothing with reference to snow noise for a width, it comprises step:

5. input picture f, and image is divided into sub-block;

6. by the screening of sub-block is formed to the alternative collection of noise piece;

7. alternative concentrated each of pair noise piece is taked auto adapted filtering, generates the estimated image of a width picture rich in detail;

8. judged whether snow noise by the deviation of statistical estimate image and original image, if judgement contains snow noise, output filtering image, otherwise export former figure.

In step 1, the method that image is divided into sub-block is: traversing graph picture, the neighbours territory of each pixel is analyzed, if the absolute value of the difference of two pixel values is less than T ₁think that two pixels are close, and be divided in a piece, form a block of pixels.

In step 1, the method that forms the alternative collection of noise piece by the screening to sub-block is:

4) first analyze the block of pixels that number of pixels is less than 6, if the sum of all pixels of this class block of pixels is greater than T divided by the ratio of total number of image pixels ₂, the piece of dividing this type is the alternative collection of noise piece, otherwise to be less than 11 piece be the alternative collection of noise piece to number of pixels in divided block;

5) two pieces that locus in alternative collection are connected merge, form new alternative collection, if step 2) in alternative collection be less than 6 piece by number of pixels and form, cast out newly concentrating number of pixels to be greater than 6 piece, otherwise cast out newly concentrating number of pixels to be greater than 11 piece;

6) using step 2) in the new collection that obtains as the final alternative collection of noise piece, prepare for subsequent step does data.

In step 2, take the concrete grammar of auto adapted filtering to be to the alternative concentrated sub-block of noise piece: first to define the alternative pixel in addition that integrates of noise piece as signaling point, then each pixel 3 × 3 neighborhoods around in decision block successively, in 5 × 5 neighborhoods and 7 × 7 neighborhoods, whether the number of signaling point is 0, for example, in if block, in pixel 3 × 3 neighborhoods, the number of signaling point is 0, judge the number of signaling point in 5 × 5 neighborhoods, the like, until judge that the number of signaling point in 7 × 7 neighborhoods is still 0, detect next piece, otherwise, if in the process detecting, signaling point number is not 0, get the weighted mean of all signaling point gray scales in current neighborhood as the new gray-scale value of this pixel, generate the estimated image of picture rich in detail concrete formula is as follows:

$\hat{f}(x_0,y_0)=\underset{(x,y)\∈S}{\mathrm{\Σ}}f(x,y)w_{(x,y)\→(x_0,y_0)}---\left(5\right)$

Wherein S is signaling point set in 3 × 3 or 5 × 5 or 7 × 7 neighborhoods, the estimation of picture rich in detail, for point (x ₀, y ₀) to the weights of point (x, y), computing method are as follows:

$w_{(x,y)\→(x_0,y_0)}=\frac{d_{(x,y)\→(x_0,y_0)}}{\underset{(p,q)\∈S}{\mathrm{\Σ}}{d}_{(p,q)\→(x_0,y_0)}}---\left(6\right)$

Wherein that point (p, q) is to point (x ₀, y ₀) distance:

$d_{(p,q)\→(x_0,y_0)}=\frac{1}{\sqrt{{(p-x_0)}^2+{(q-y_0)}^2}}.---\left(3\right)$

In step 3, judge whether that by the deviation of statistical estimate image and original image the concrete grammar of snow noise is, obtain the deviation of two images by formula (4):

$\mathrm{diff}=\frac{\underset{(x,y)\∈N}{\mathrm{\Σ}}|f(x,y)-\hat{f}(x,y)|}{H\×W}---\left(8\right)$

Wherein N is the alternative set of concentrating all pixels of noise piece, and H is picture altitude, and W is picture traverse, if deviation diff is greater than T ₃, judge that original image contains snow noise, the estimation of output picture rich in detail , otherwise think there is no snow noise, export former figure f.

As shown in Figure 3, filtering result as shown in Figure 4 for the snow noise distribution plan that this method detects.

Finally it should be noted that: the foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, although the present invention is had been described in detail according to previous embodiment, for a person skilled in the art, the technical scheme that still can record previous embodiment is modified, or part technical characterictic is wherein equal to replacement.Within the spirit and principles in the present invention all, any amendment of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Ultimate principle of the present invention and principal character and advantage of the present invention have more than been described.Industry technician should understand; the present invention is not restricted to the described embodiments; that in above-described embodiment and instructions, describes just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.

Claims (5)

1. nothing, with reference to a detection elimination algorithm for snow noise, is characterized in that, said method comprising the steps of:

(1) input picture f, and image is divided into sub-block;

(2) by the screening of sub-block is formed to the alternative collection of noise piece;

(3) alternative concentrated each of noise piece is taked to auto adapted filtering, generate the estimated image of a width picture rich in detail;

(4) judged whether snow noise by the deviation of statistical estimate image and original image, if judgement contains snow noise, output filtering image, otherwise export former figure.

2. the detection elimination algorithm of snow noise according to claim 1, it is characterized in that in step (1), the method that image is divided into sub-block is: traversing graph picture, the neighbours territory of each pixel is analyzed, if the absolute value of the difference of two pixel values is less than T ₁think that two pixels are close, and be divided in a piece, form a block of pixels.

3. the detection elimination algorithm of snow noise according to claim 1, is characterized in that in step (1), and the method that forms the alternative collection of noise piece by the screening to sub-block is:

1) first analyze the block of pixels that number of pixels is less than 6, if the sum of all pixels of this class block of pixels is greater than T divided by the ratio of total number of image pixels ₂, the piece of dividing this type is the alternative collection of noise piece, otherwise to be less than 11 piece be the alternative collection of noise piece to number of pixels in divided block;

2) two pieces that locus in alternative collection are connected merge, form new alternative collection, if step 2) in alternative collection be less than 6 piece by number of pixels and form, cast out newly concentrating number of pixels to be greater than 6 piece, otherwise cast out newly concentrating number of pixels to be greater than 11 piece;

3) using step 2) in the new collection that obtains as the final alternative collection of noise piece, prepare for subsequent step does data.

4. the detection elimination algorithm of snow noise according to claim 1, it is characterized in that in step (2), take the concrete grammar of auto adapted filtering to be to the alternative concentrated sub-block of noise piece: first to define the alternative pixel in addition that integrates of noise piece as signaling point, then each pixel 3 × 3 neighborhoods around in decision block successively, in 5 × 5 neighborhoods and 7 × 7 neighborhoods, whether the number of signaling point is 0, for example, in if block, in pixel 3 × 3 neighborhoods, the number of signaling point is 0, judge the number of signaling point in 5 × 5 neighborhoods, the like, until judge that the number of signaling point in 7 × 7 neighborhoods is still 0, detect next piece, otherwise, if in the process detecting, signaling point number is not 0, get the weighted mean of all signaling point gray scales in current neighborhood as the new gray-scale value of this pixel, generate the estimated image of picture rich in detail concrete formula is as follows:

$\hat{f}(x_0,y_0)=\underset{(x,y)\∈S}{\mathrm{\Σ}}f(x,y)w_{(x,y)\→(x_0,y_0)}---\left(1\right)$

Wherein S is signaling point set in 3 × 3 or 5 × 5 or 7 × 7 neighborhoods, the estimation of picture rich in detail, for point (x ₀, y ₀) to the weights of point (x, y), computing method are as follows:

$w_{(x,y)\→(x_0,y_0)}=\frac{d_{(x,y)\→(x_0,y_0)}}{\underset{(p,q)\∈S}{\mathrm{\Σ}}{d}_{(p,q)\→(x_0,y_0)}}---\left(2\right)$

Wherein that point (p, q) is to point (x ₀, y ₀) distance:

$d_{(p,q)\→(x_0,y_0)}=\frac{1}{\sqrt{{(p-x_0)}^2+{(q-y_0)}^2}}.---\left(3\right)$

5. the detection elimination algorithm of snow noise according to claim 1, it is characterized in that in step (3), the concrete grammar that judges whether snow noise by the deviation of statistical estimate image and original image is to obtain the deviation of two images by formula (4):

$\mathrm{diff}=\frac{\underset{(x,y)\∈N}{\mathrm{\Σ}}|f(x,y)-\hat{f}(x,y)|}{H\×W}---\left(4\right)$

Wherein N is the alternative set of concentrating all pixels of noise piece, and H is picture altitude, and W is picture traverse, if deviation diff is greater than T ₃, judge that original image contains snow noise, the estimation of output picture rich in detail , otherwise think there is no snow noise, export former figure f.