CN101807298A - Method for determining intensity of speckle noise in images - Google Patents

Abstract

The invention relates to a method for determining the intensity of speckle noise in images, which comprises the six steps of: selecting an area of which the gray scale is relatively uniform; calculating the grayscale mean value of pixels in the relatively-uniform area and different orders of Gaussian-Hermite moments of all pixels; constructing feature vectors; calculating and detecting noise intensity characteristic values of the images; converting the noise intensity characteristic values; and determining intensity values of speckle noise in the images. In the method, on the basis of a multiplicative noise module, the intensity of the noise is determined under the condition of no any priori knowledge, and thus the method has the characteristics of high precision, high speed and strong practical applicability and generality, and can be used for determining the intensity of the noise of the images containing the speckle noise, such as visible light images with grayscale uniform areas, synthetic aperture radar images, medical ultrasound images and the like.

Description

Determine the method for intensity of speckle noise in the image

Technical field

The invention belongs to technical field of image processing, be specifically related in conjunction with the method for determining intensity of speckle noise in the image based on the noise analysis model of Gauss's hermitian square (Gaussian-Hermite).

Background technology

Image Acquisition and transmission course make most of digital pictures have noise in various degree, not only influence the visual effect of image, and hinder the various processing such as target detection, feature extraction and parameter measurement of postorder, directly influence the image interpretation quality.

According to the stacked system of noise and image, picture noise comprises multiplicative noise and additive noise two classes.Research emphasis concentrates on the Gaussian noise in the additive noise at present, existing noise parameter estimates that representative achievement comprises: nineteen ninety, people such as Peter Meer are research object with the Gaussian noise of zero-mean, come the variance of the noisy image of blind estimation by image pyramid, be characterized in that the average error rate is 0.06, be O (log (N)) (N is the image size) estimated time.1999, people such as K Rank are research object with the image that is polluted by Gaussian noise, go on foot the estimation of finishing noise by three, at first suppress the suffered influence of original image by level and vertical operation, second step was calculated local signal variance histogram, the last resulting estimation variance value of statistical computation histogram.2005, people such as DH Shin introduced a kind of noise estimation method fast, and this method by the image that additive white Gaussian noise pollutes, is based on the noise estimation method of piecemeal with the gaussian filtering estimation, can be used for commercial graphic and vedio noise reduction.Widespread use be wavelet field noise criteria variance estimation formulas σ=MAD/0.6745 that Donoho and Johnstone propose, wherein MAD is the intermediate value of diagonal subband wavelet coefficient amplitude.This method at noise hour, estimating noise can be bigger than normal, so people improve this method in engineering is used, using maximum at present is overall variance and local variance.

Speckle noise is common a kind of multiplicative noise, and for example in diameter radar image that plays a significant role day by day and medical ultrasonic image, because image-forming mechanism, speckle noise becomes intrinsic noise.Reasonable analysis and to suppress speckle noise be precondition and the committed step that various speckle noise Flame Image Process etc. are used.Be different from Gaussian noise,, be difficult to mode direct estimation noise details by noisy image and noise suppression image subtraction because speckle noise is to be added in the image in the mode that multiplies each other.Mostly the method for handling speckle noise at present is it computing of taking the logarithm is converted into the Gaussian noise model of additivity, analyzes again.So far, the method that does not also exist a kind of general intensity of speckle noise to determine.

Summary of the invention

Technical matters to be solved by this invention is to overcome the deficiency that present noise intensity is determined method, and a kind of method of determining intensity of speckle noise in the image fast and accurately is provided.

Solving the problems of the technologies described above the technical scheme that is adopted comprises the steps:

1, chooses the relative homogeneous area of gray scale

In noisy image, choose gray scale zone relatively uniformly.

2, calculate the relative homogeneous area pixel of gray scale gray average and should the zone in the not same order Gauss hermitian square of pixel;

Above-mentioned gray average is the average gray of each picture element in the relative homogeneous area of gray scale.

The not same order Gauss hermitian square of the relative homogeneous area pixel of gray scale is calculated as follows:

$M_{p,q}(x,y,I(x,y))=\underset{t=-1}{\overset{1}{\mathrm{\Σ}}}\underset{v=-1}{\overset{1}{\mathrm{\Σ}}}G(t,v,\mathrm{\σ})H_{P,q}(t/\mathrm{\σ},v/\mathrm{\σ})I(x+t,y+v)---\left(1\right)$

In the formula (1), (t, v σ) are two-dimensional Gaussian function, and have G

$G(t,v,\mathrm{\&sigma;})=\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}\exp (-\frac{({\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HSA00000006267300021.png"\; state="\{\{state\}\}">t</figure-callout>}}^2+v^2)}{2{\mathrm{\&sigma;}}^2})---\left(2\right)$

H _{P, q}(t/ σ, v/ σ) is two dimension (p, q) rank hermitian polynomial expression, and having

H _P，q(t/σ，v/σ)＝H _p(t/σ)H _q(v/σ) (3)

H _n(t)＝(-1) ⁿexp(t ²)(d ⁿ/dt ⁿ)exp(-t ²) (4)

T is-1 or 0 or 1 in formula (1)～(4), and v is-1 or 0 or 1, and σ is 0.7.

3, structural attitude vector

The not same order Gauss hermitian square that calculates according to formula (1), by following formula structural attitude vector:

$\left[\begin{array}{c}{M}_u(x,y)\\ M_v(x,y)\end{array}\right]=\left[\begin{array}{c}\mathrm{\&lambda;}{M}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{3,0}}(x,y,I(x,y))\\ \mathrm{\&lambda;}{M}_{\mathrm{0,1}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{0,3}}(x,y,I(x,y))\end{array}\right]---\left(5\right)$

In the formula (5), λ is the associating weight coefficient of Gauss's hermitian square of not same order, 0＜λ＜1.(x, y), through type (5) can obtain a proper vector [M to each point of selected homogeneous area in the image _u, M _v] ^T

4, calculate the noise intensity eigenwert of detected image

Press formula (6) calculating noise characteristic strength value:

$M_{\mathrm{uv}}=\frac{1}{2N}\mathrm{\&Sigma;}\left(\right|M_u|+|M_v\left|\right)---\left(6\right)$

M in the formula _UvBe the noise intensity eigenwert, N is the number of proper vector, and N is a positive integer.

5, conversion noise characteristic strength value

With the average gray of each picture element in the relative homogeneous area of the gray scale of step 2 and the detected image noise intensity eigenwert substitution formula (7) of step 4:

The conversion noise characteristic strength value.

6, determine the intensity level of speckle noise in the image

Determine the intensity level of speckle noise in the image by following formula:

y＝99.1468x ⁴-32.0896x ³+51.8903x ²-0.1269x+0.0013 (8)

X is the noise intensity eigenwert after step 5 conversion in the formula.

The present invention has set up the polynomial function of determining based on the noise intensity of Gauss's hermitian square, and this function can utilize the zonule of input picture to determine the strength information of speckle noise fast; Different with the processing mode of common " multiplicative noise is converted to additive noise through log-transformation ", the present invention is directly based on the multiplicative noise model, under situation without any priori, carry out determining of noise intensity, have that precision height, speed are fast, the characteristics of practicality and highly versatile, visible images, diameter radar image and the medical ultrasonic images etc. that can be used for having the uniform gray level district contain the determining of picture noise intensity of speckle noise.

Description of drawings

Fig. 1 determines the process flow diagram of method for intensity of speckle noise of the present invention.

Fig. 2 is the intensity of speckle noise of determining in the emulating image.

Fig. 3 is the intensity of speckle noise of determining in the visible images.

Fig. 4 is the intensity of speckle noise of determining in the diameter radar image of shore line.

Embodiment

The present invention is described in more detail below in conjunction with drawings and Examples, but the invention is not restricted to these embodiment.

Embodiment 1

Intensity of speckle noise with definite emulating image (containing intensity of speckle noise is 0.02) is an example, and its method step is as follows:

1, chooses the relative homogeneous area of gray scale

Containing intensity of speckle noise is to choose a gray areas 1 relatively uniformly in 0.02 the emulating image, and the gray areas of choosing 1 is seen Fig. 2 for the zone in the rectangular box.

2, calculate institute's favored area pixel gray average and should the zone in the not same order Gauss hermitian square of pixel

Above-mentioned gray average is the average gray of each picture element in the relative homogeneous area of gray scale, and the gray average that gets gray areas 1 pixel is 191.

The not same order Gauss hermitian square of each pixel calculates as follows in the gray areas 1:

$M_{p,q}(x,y,I(x,y))=\underset{t=-1}{\overset{1}{\mathrm{\&Sigma;}}}\underset{v=-1}{\overset{1}{\mathrm{\&Sigma;}}}G(t,v,\mathrm{\&sigma;})H_{P,q}(t/\mathrm{\&sigma;},v/\mathrm{\&sigma;})I(x+t,y+v)---\left(1\right)$

(t, v are two-dimensional Gaussian function σ), and have G in the formula

$G(t,v,\mathrm{\&sigma;})=\frac{1}{2\mathrm{\&pi;}{\mathrm{\&sigma;}}^2}\exp (-\frac{({\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="HSA00000006267300021.png"\; state="\{\{state\}\}">t</figure-callout>}}^2+v^2)}{2{\mathrm{\&sigma;}}^2})---\left(2\right)$

H _{P, q}(t/ σ, v/ σ) is two dimension (p, q) rank hermitian polynomial expression, and having

H _P，q(t/σ，v/σ)＝H _p(t/σ)H _q(v/σ) (3)

H _n(t)＝(-l) ⁿexp(t ²)(d ⁿ/dt ⁿ)exp(-t ²) (4)

T is-1 or 0 or 1 in formula (1)～(4), and v is-1 or 0 or 1, and σ is 0.7.

3, structural attitude vector

The not same order Gauss hermitian square that calculates according to formula (1), by following formula structural attitude vector:

$\left[\begin{array}{c}{M}_u(x,y)\\ M_v(x,y)\end{array}\right]=\left[\begin{array}{c}\mathrm{\&lambda;}{M}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{3,0}}(x,y,I(x,y))\\ \mathrm{\&lambda;}{M}_{\mathrm{0,1}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{\mathrm{0,3}}(x,y,I(x,y))\end{array}\right]---\left(5\right)$

In the formula (5), λ is the associating weight coefficient of Gauss's hermitian square of not same order, 0＜λ＜1.(x, y), through type (5) can obtain a proper vector [M to each point of selected homogeneous area in the image _u, M _v] ^T

4, calculate the noise intensity eigenwert of detected image

Press formula (6) calculating noise characteristic strength value:

$M_{\mathrm{uv}}=\frac{1}{2N}\mathrm{\&Sigma;}\left(\right|M_u|+|M_v\left|\right)---\left(6\right)$

M in the formula _UvBe the noise intensity eigenwert, N is the number of proper vector, and N is 5664, gets noise intensity eigenwert M _UvFor: 0.0771.

5, conversion noise characteristic strength value

With the average gray of each picture element and the detected image noise intensity eigenwert substitution formula (7) of step 4 in the gray areas 1 of step 2:

The conversion noise characteristic strength value, the gray scale constant is 50 in the formula, must change back noise intensity eigenwert is 0.0202.

6, determine the intensity level of speckle noise in the image

Determine the intensity level of speckle noise in the image by following formula:

y＝99.1468x ⁴-32.0896x ³+51.8903x ²-0.1269x+0.0013 (8)

X is the noise intensity eigenwert after step 5 conversion in the formula, in the gray areas 1 intensity level of speckle noise be 0.0196, be 0.02 to compare with the intensity of speckle noise that contains in the image, error rate is 2%, be 0.26 second working time.

Embodiment 2

Intensity of speckle noise with definite visible images " coin " (containing intensity of speckle noise is 0.1) is an example, and its method step is as follows:

In choosing the relative homogeneous area of gray scale step 1, be to choose a gray areas 2 relatively uniformly in 0.1 the visible images " coin " containing intensity of speckle noise, the gray areas of choosing 2 is seen Fig. 3 for the zone in the rectangular box.

In the gray average that calculates institute's favored area pixel and this zone in the not same order Gauss hermitian square step 2 of pixel, the gray average of gray areas 2 pixels is 226, the used computing formula of the not same order Gauss hermitian square of each pixel is identical with embodiment 1 in the gray areas 2, and t, the v in formula (1)～(4), the value of σ are identical with embodiment 1.

Structural attitude vector step 3 is identical with embodiment 1.

In calculating the noise intensity eigenwert step 4 of detected image, the used computing formula of calculating noise characteristic strength value is identical with embodiment 1, M in the formula _UvBe the noise intensity eigenwert, N is the number of proper vector, and N is 3234, gets noise intensity eigenwert M _UvBe 0.2072.

In conversion noise characteristic strength value step 5, the used computing formula of conversion noise characteristic strength value is identical with embodiment 1, and the gray scale constant is 50 in the formula, and must change back noise intensity eigenwert is 0.0457.

In determining image in the intensity level step 6 of speckle noise, determine that the used computing formula of the intensity level of speckle noise in the image is identical with embodiment 1, the intensity level that gets speckle noise in the gray areas 2 is 0.1014, with the intensity of speckle noise that contains in the image is 0.1 to compare, error rate is 1.4%, and be 0.18 second working time.

Embodiment 3

Intensity of speckle noise with definite synthetic-aperture radar shore line image is an example, and its method step is as follows:

In choosing the relative homogeneous area of gray scale step 1, in the image of synthetic-aperture radar shore line, choose a gray areas 3 relatively uniformly, the gray areas of choosing 3 is seen Fig. 4 for the zone in the rectangular box.

In the gray average that calculates institute's favored area pixel and this zone in the not same order Gauss hermitian square step 2 of pixel, the gray average of gray areas 3 pixels is 25, the used computing formula of the not same order Gauss hermitian square of gray areas 3 pixels is identical with embodiment 1, and t, the v in formula (1)～(4), the value of σ are identical with embodiment 1.

Structural attitude vector step 3 is identical with embodiment 1.

In calculating the noise intensity eigenwert step 4 of detected image, the used computing formula of calculating noise characteristic strength value is identical with embodiment 1, M in the formula _UvBe the noise intensity eigenwert, N is the number of proper vector, and N is 4020, gets noise intensity eigenwert M _UvBe 0.0193.

In conversion noise characteristic strength value step 5, the used computing formula of conversion noise characteristic strength value is identical with embodiment 1, and the gray scale constant is 50 in the formula, and obtaining changing back noise intensity eigenwert is 0.0388.

In determining image in the intensity level step 6 of speckle noise, determine that the used computing formula of the intensity level of speckle noise in the image is identical with embodiment 1, must gray areas 3 in the intensity level of speckle noise be 0.073, be 0.20 second working time.

Claims (1)

1. the method for intensity of speckle noise in the definite image is characterized in that it comprises the steps:

(1) chooses the relative homogeneous area of gray scale

In noisy image, choose gray scale zone relatively uniformly;

(2) the not same order Gauss hermitian square of pixel in the gray average of the relative homogeneous area pixel of calculating gray scale and this zone;

Above-mentioned gray average is the average gray of each picture element in the relative homogeneous area of gray scale;

The not same order Gauss hermitian square of the relative homogeneous area pixel of gray scale is calculated as follows:

$M_{p,q}(x,y,I(x,y))=\underset{t=-1}{\overset{1}{\mathrm{\&Sigma;}}}\underset{v=-1}{\overset{1}{\mathrm{\&Sigma;}}}G(t,v,\mathrm{\&sigma;})H_{P,q}(t/\mathrm{\&sigma;},v/\mathrm{\&sigma;})I(x+t,y+v)---\left(1\right)$

In the formula (1), (t, v σ) are two-dimensional Gaussian function, and have G

$G(t,v,\mathrm{\&sigma;})=\frac{1}{2{\mathrm{\&pi;\&sigma;}}^2}\exp (-\frac{(t^2+v^2)}{{2\mathrm{\&sigma;}}^2})---\left(2\right)$

H _{P, q}(t/ σ, v/ σ) is two dimension (p, q) rank hermitian polynomial expression, and having

H _P，q(t/σ，v/σ)＝H _p(t/σ)H _q(v/σ) (3)

H _n(t)＝(-1) ⁿexp(t ²)(d ⁿ/dtn)exp(-t ²) (4)

T is-1 or 0 or 1 in formula (1)～(4), and v is-1 or 0 or 1, and σ is 0.7;

(3) structural attitude vector

The not same order Gauss hermitian square that calculates according to formula (1), by following formula structural attitude vector:

$\left[\begin{array}{c}{M}_u(x,y)\\ M_v(x,y)\end{array}\right]=\left[\begin{array}{c}{\mathrm{\&lambda;M}}_{\mathrm{1,0}}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{3.0}(x,y,I(x,y))\\ {\mathrm{\&lambda;M}}_{0.1}(x,y,I(x,y))+(1-\mathrm{\&lambda;})M_{0.3}(x,y,I(x,y))\end{array}\right]---\left(5\right)$

In the formula (5), λ is the associating weight coefficient of Gauss's hermitian square of not same order, 0＜λ＜1, and (x, y), through type (5) can obtain a proper vector [M to each point of selected homogeneous area in the image _u, M _v] ^T

(4) the noise intensity eigenwert of calculating detected image

Press formula (6) calculating noise characteristic strength value:

$M_{\mathrm{uv}}=\frac{1}{2N}\mathrm{\&Sigma;}\left(\right|M_u|+|M_v\left|\right)---\left(6\right)$

M in the formula _UvBe the noise intensity eigenwert, N is the number of proper vector, and N is a positive integer;

(5) conversion noise characteristic strength value

With the average gray of each picture element in the relative homogeneous area of the gray scale of step 2 and the detected image noise intensity eigenwert substitution formula (7) of step 4:

The conversion noise characteristic strength value;

(6) determine the intensity level of speckle noise in the image

Determine the intensity level of speckle noise in the image by following formula:

y＝99.1468x ⁴-32.0896x ³+51.8903x ²-0.1269x+0.0013 (8)

X is the noise intensity eigenwert after step (5) conversion in the formula.

Images