CN102096909A - Improved unsharp masking image reinforcing method based on logarithm image processing model - Google Patents

Abstract

The invention provides an improved unsharp masking image reinforcing method based on a logarithm image processing model, belonging to the filed of image processing. The method comprises the following steps: transforming a gray level-based input image under a conventional model into a gray tone-based image under the logarithm image processing model; carrying out the improved unsharp masking reinforcement on the image; and transforming the image subjected the unsharp masking reinforcement into the gray level-based image under the conventional model. After the unsharp masking image reinforcing method based on the logarithm image processing model is used, the defect of the 'overflowing' of the gray level in the unsharp method under the conversional model is overcome; due to the isomorphic relationship between the operation of the logarithm image processing model and the operation of the conventional process model, the isomorphic transform and the non-isomorphic transform, the method is simpler and more efficient; and the edge region and smooth region of the image are reinforced to different extents by utilizing the gradient information of the image, so that the defect that the whole image is reinforced by the constant factor at the same intensity based on the conversional method is overcome, and the obtained result is better.

Description

Improved unsharp masking image enchancing method based on the logarithmic image transaction module

Technical field

The invention belongs to image processing field, particularly a kind of improved unsharp masking image enchancing method based on the logarithmic image transaction module.

Background technology

(Unsharp Masking is the image edge enhancement method that proposes in order to adapt to the characteristics of image of handling different size UM) to the unsharp masking method, and this method is to be applied in the camera work the earliest, to strengthen edge of image and details; Optic method of operating is that positive that will focus on and the negative film that defocuses superpose on egative film, the result has strengthened positive high frequency composition, thereby strengthened profile, the negative film that defocuses is equivalent to " bluring " template (mask), it is just in time opposite with the effect of sharpening, therefore, this method is called as the unsharp masking method.

The principle of classical linear unsharp masking method as shown in Figure 1.At first a passivation blurred picture will be produced after the original image low-pass filtering, original image and this blurred picture are subtracted each other the image that obtains the reserved high-frequency composition, again high frequency imaging is amplified back and original image stack with parameter K, this just produces an image that strengthens the edge, and the conventional model formula is as follows:

I _um＝I+K·(I-I _us)

Wherein, I is an original image, I _UsBe image after the low-pass filtering, get the image behind the mean filter here, K is positive constant;

Although this method is simple, it is also better relatively to strengthen effect, and it has 2 outstanding shortcomings:

(1) system is very responsive to noise, owing to adopted linear Hi-pass filter, the details and the noise of image are enhanced simultaneously, especially at the flat site of image, even very little noise is also very obvious;

(2) overshoot (overshoot) phenomenon, because the high details area of image is bigger with respect to other regional enhancing, the image after the processing can present tangible artificial treatment vestige;

In the disclosed at home and abroad document, in order to overcome the shortcoming of linear unsharp masking method, especially to the susceptibility of noise, the researcher in image field has proposed various methods, these method major parts are to replace linear Hi-pass filter with nonlinear filter, proposed nonlinear operator based on the Teager algorithm to reducing noise and strengthening details consideration: the S.K.Mitra that compromises, this nonlinear operator can be approximately local mean value weight Hi-pass filter; G.Ramponi has proposed a cube unsharp masking method, and the essence of this method is to multiply by Laplace operator with a square filtering device operator to edge sensitive, only strengthens the image detail in local brightness variation zone, less relatively noise; Y.H.Lee has proposed the operator based on sequence statistics Laplace method, and the difference of the output of this operator and local mean value and local intermediate value is proportional, and it can remove white Gaussian noise effectively; A.Polesel has proposed self-adaptation unsharp masking method, and this method adopts sef-adapting filter that the details of image is strengthened degree greatly, and the flat site of image is strengthened hardly, thereby can reduce the noise of flat site; But, though method above-mentioned has reduced noise with respect to linear unsharp masking method, but at flat site, noise is still apparent in view, and, strengthen effect preferably for details area in the image is reached, the high details area of image often strengthens excessive, causes the appearance of overshoot phenomenon.

Summary of the invention

For overcoming the defective of said method, the present invention proposes a kind of improved unsharp masking image enchancing method based on the logarithmic image transaction module.

At first; (comprise traditional add operation and scalar take advantage of operation) can not well meet the physical laws of image overlay and amplification because classical Flame Image Process model; and for gray-scale value digital picture within limits; traditional add operation usually can produce spillover; so shortcoming and deficiency in order to improve classical model, Jourlin and Pinoli have proposed the logarithmic image transaction module in the eighties middle period in last century:

In the logarithmic image transaction module, image is represented with the form of mapping, is called gray tone function (gray tone functions) and represents that with alphabetical f logarithmic image transaction module formula is as follows:

In the formula, alphabetical S represents the vector space at gray tone function place, [0, the M) span of expression gray tone function, g represents the gradient factor,

Be respectively addition, subtraction and the multiply operation under the logarithmic image transaction module;

Technical scheme of the present invention is achieved in that the improved unsharp masking image enchancing method based on the logarithmic image transaction module, may further comprise the steps:

Step 1: will be under conventional model convert under the logarithmic image transaction module based on the image of gray tone based on the input picture of gray scale, formula is:

$I=M-\stackrel{\‾}{I}$

In the formula, I represents half tone picture, and M represents the maximum gradation value matrix,

The expression gray level image;

Step 2: image is carried out improved unsharp masking strengthen: utilize the gradient factor to replace traditional constant enhancer, to treat the reinforcing coefficient of smooth region and fringe region with a certain discrimination, image after step 1 is handled, while execution in step 2-1 and step 2-2, method is as follows:

Step 2-1: the gradient factor g of computed image, at first utilize Gauss's smoothing operator to carrying out filtering based on image gray under the conventional model, formula is:

I _s＝I*G _σ

In the formula, I _sFor the input picture based on gray scale under the conventional model is carried out the later image of gaussian filtering, G _σFor variance is the Gaussian filter of σ;

Then, utilize the gradient information Gray of difference method computed image;

At last, the gradient factor g of computed image, formula is:

$g=\frac{1}{1+\mathrm{Gray}/m}$

Image is carried out unsharp masking strengthen, formula is:

I _um＝I+g×(I-I _us)

In the formula, I _UsThe image of expression after low-pass filtering, formula is:

$I_{\mathrm{us}}(x,y)=\frac{1}{M\×N}\×\underset{i=x-(M-1)/2}{\overset{x+(M-1)/2}{\mathrm{\Σ}}}\underset{j=y-(N-1)/2}{\overset{y+(N-1)/2}{\mathrm{\Σ}}}I(i,j)$

In the formula, N, M are the height and width of image, and (x, y represent the current horizontal ordinate of handling image pixel point respectively to I for i, j) expression original image;

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula,

Expression is carried out image after the anti-isomorphism conversion, execution in step 3 to half tone picture;

Step 2-2: image is carried out improved unsharp masking strengthen: at first, handle according to isomorphism transfer pair half tone picture, transformation for mula is:

$\stackrel{\‾}{I}=\mathrm{\φ}\left(I\right)=-M\ln (1-\frac{I}{M})$

In the formula, φ (I) expression is carried out image after the isomorphism conversion to half tone picture;

Then, utilize mean filter that the input picture based on gray scale under the conventional model is carried out low-pass filtering treatment, formula is:

$I_{\mathrm{us}}(x,y)=\frac{1}{M\×N}\×\underset{i=x-(M-1)/2}{\overset{x+(M-1)/2}{\mathrm{\Σ}}}\underset{j=y-(N-1)/2}{\overset{y+(N-1)/2}{\mathrm{\Σ}}}I(i,j)$

In the formula, N, M are the height and width of image, I (i, j) expression original image;

Afterwards, utilize marginal information as weight factor original image to be carried out the unsharp masking enhancement process, the operation of this moment is traditional linear operation, and its formula is:

I _um＝I+g×(I-I _us)

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula,

Expression is carried out image after the anti-isomorphism conversion to half tone picture;

Step 3: the unsharp masking that step 2 is obtained strengthen image transitions be under the conventional model based on image gray, formula is:

${\stackrel{\‾}{I}}_{\mathrm{um}}=M-I_{\mathrm{um}}$

In the formula,

Expression strengthens result's gray level image, and M represents the maximum gradation value matrix, and big or small identical and each the value of this matrix and image array all is the maximum gradation value of image, I _UmExpression strengthens result's half tone picture;

Advantage of the present invention: the present invention improves the effect and the efficient of unsharp masking image enchancing method by three gordian techniquies, the one, under the logarithmic image transaction module, realize the unsharp masking image enchancing method, overcome that the deficiency that gray scale " is overflowed " appears in anti-sharpening method under the conventional model, and the logarithmic image transaction module meets the human visual light sensitivity principles, and the enhanced results visual effect is better; The 2nd, utilize the isomorphic relations of the operation room of the operation of logarithmic image transaction module and conventional process model, by isomorphism conversion and anti-isomorphism conversion, realize simpler, efficient; The 3rd, utilize the gradient information definition edge detection operator of image to improve the factor that anti-sharpening strengthens, to territory, edge of image regional peace skating area enforcement enhancing in various degree, remedied and utilized constant factor that entire image is carried out the deficiency that same intensity strengthens in the classic method, the gained result is better.

Description of drawings

Fig. 1 is the schematic diagram of linear unsharp masking method that the present invention is based on the improved unsharp masking image enchancing method classics of logarithmic image transaction module;

Fig. 2 is the improved unsharp masking image enchancing method general flow chart that the present invention is based on the logarithmic image transaction module;

Fig. 3 is that the improved unsharp masking image enchancing method that the present invention is based on the logarithmic image transaction module carries out improved unsharp masking enhancing process flow diagram to image;

Fig. 4 (a) is for the present invention is based on the improved unsharp masking image enchancing method original image synoptic diagram of logarithmic image transaction module;

Fig. 4 (b) is for the present invention is based on the improved unsharp masking image enchancing method speckle noise image of logarithmic image transaction module;

Fig. 4 (c) is for the present invention is based on the improved unsharp masking image enchancing method Gaussian noise image of logarithmic image transaction module;

Fig. 4 (d) is for the present invention is based on the improved unsharp masking image enchancing method poisson noise image of logarithmic image transaction module;

Fig. 4 (e) is for the present invention is based on the enhancing result schematic diagram of the improved unsharp masking image enchancing method traditional algorithm of logarithmic image transaction module to the speckle noise image;

Fig. 4 (f) is for the present invention is based on the enhancing result schematic diagram of the improved unsharp masking image enchancing method traditional algorithm of logarithmic image transaction module to the Gaussian noise image;

Fig. 4 (g) is for the present invention is based on the enhancing result schematic diagram of the improved unsharp masking image enchancing method traditional algorithm of logarithmic image transaction module to the poisson noise image;

Fig. 4 (h) improves the enhancing result schematic diagram of algorithm to the speckle noise image for the improved unsharp masking image enchancing method that the present invention is based on the logarithmic image transaction module;

Fig. 4 (i) improves the enhancing result schematic diagram of algorithm to the Gaussian noise image for the improved unsharp masking image enchancing method that the present invention is based on the logarithmic image transaction module;

Fig. 4 (j) improves the enhancing result schematic diagram of algorithm to the poisson noise image for the improved unsharp masking image enchancing method that the present invention is based on the logarithmic image transaction module;

Fig. 5 (a) is for the present invention is based on the improved unsharp masking image enchancing method medical science original image synoptic diagram of logarithmic image transaction module;

Fig. 5 (b) strengthens result schematic diagram for the improved unsharp masking image enchancing method tradition unsharp masking that the present invention is based on the logarithmic image transaction module;

Fig. 5 (c) strengthens the result for the improved unsharp masking of the improved unsharp masking image enchancing method that the present invention is based on the logarithmic image transaction module.

Embodiment

Below in conjunction with drawings and Examples the present invention is described in further detail.

Present embodiment adopts the improved unsharp masking image enchancing method based on the logarithmic image transaction module, and flow process such as Fig. 2, shown in Figure 3 may further comprise the steps:

Step 1: will be under conventional model convert under the logarithmic image transaction module image to based on gray tone based on the input picture of gray scale, at first, import under the width of cloth conventional model based on image gray, shown in Fig. 4 (a), the part of this image is expressed as with matrix form:

$\left[\begin{array}{cccc}144& 176& 168& 148\\ 188& 141& 169& 145\\ 158& 158& 155& 171\\ 169& 154& 163& 160\end{array}\right]$

Utilize formula

Above-mentioned image transitions is become under the logarithm transaction module based on the image of gray tone, is converted to the following matrix representation of image after the gray tone:

$\left[\begin{array}{cccc}112& 80& 88& 108\\ 68& 115& 87& 111\\ 98& 98& 101& 85\\ 87& 102& 93& 96\end{array}\right]$

In the formula,

The expression gray level image, I represents half tone picture, M represents the maximum gradation value matrix;

Step 2: image is carried out improved unsharp masking strengthen: utilize the gradient factor to replace traditional constant enhancer, the image after step 1 is handled, while execution in step 2-1 and step 2-2, method is as follows:

Step 2-1: the gradient factor g of computed image, at first utilize Gauss's smoothing operator to carrying out filtering based on image gray under the conventional model, formula is:

I _s＝I*G _σ

In the formula, I _sFor the input picture based on gray scale under the conventional model is carried out the later image of gaussian filtering, G _σFor variance is the Gaussian filter of σ;

Then, utilize the gradient information Gray of difference method computed image;

At last, the gradient factor g of computed image, formula is:

$g=\frac{1}{1+\mathrm{Gray}/m}$

In the formula, get m=50, the gradient factor based on image gray under the calculation procedure 1 described conventional model is as follows:

$\left[\begin{array}{cccc}0.4& 0.5& 0.7& 0.9\\ 0.4& 0.6& 0.8& 0.9\\ 0.5& 0.7& 0.8& 1\\ 0.6& 0.7& 0.8& 0.9\end{array}\right]$

Image is carried out unsharp masking strengthen, formula is:

I _um＝I+g×(I-I _us)

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula,

Expression is carried out image after the anti-isomorphism conversion, execution in step 3 to half tone picture;

Step 2-2: image is carried out improved unsharp masking strengthen: at first, handle according to isomorphism transfer pair half tone picture, transformation for mula is:

$\stackrel{\‾}{I}=\mathrm{\φ}\left(I\right)=-M\ln (1-\frac{I}{M})$

In the formula, φ (I) expression is carried out image after the isomorphism conversion to half tone picture;

Then, utilize mean filter that the input picture based on gray scale under the conventional model is carried out low-pass filtering treatment, formula is:

$I_{\mathrm{us}}(x,y)=\frac{1}{M\×N}\×\underset{i=x-(M-1)/2}{\overset{x+(M-1)/2}{\mathrm{\Σ}}}\underset{j=y-(N-1)/2}{\overset{y+(N-1)/2}{\mathrm{\Σ}}}I(i,j)$

In the formula, N, M are the height and width of image, and I (i, j) expression is an original image;

In the formula, the input picture based on gray scale under the conventional model through the matrix after the low-pass filtering is:

$\left[\begin{array}{cccc}138& 154& 156& 156\\ 153& 161& 159& 159\\ 166& 161& 157& 160\\ 171& 169& 170& 166\end{array}\right]$

Afterwards, utilize marginal information as weight factor original image to be carried out the unsharp masking enhancement process, the operation of this moment is traditional linear operation, and its formula is:

I _um＝I+g×(I-I _us)

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula,

Expression is carried out image after the anti-isomorphism conversion to half tone picture;

Step 3: the unsharp masking that step 2 is obtained strengthen image transitions be under the conventional model based on image gray, its image array form is expressed as:

$\left[\begin{array}{cccc}145& 180& 171& 146\\ 193& 137& 172& 141\\ 157& 157& 154& 174\\ 169& 151& 161& 158\end{array}\right]$

Fig. 4 (a)～4 (j), wherein Fig. 4 (a) is an original image, Fig. 4 (b), Fig. 4 (c) and Fig. 4 (d) are for having added the image (having added speckle noise, Gaussian noise, poisson noise successively) of noise, Fig. 4 (e), Fig. 4 (f) and Fig. 4 (g) are the enhancing result of traditional unsharp masking algorithm, by among the figure we integral image is all brighter as can be seen, this is because classic method is directly carried out linear, additive with the high fdrequency component and the original image of image, do like this and can cause the gray scale spillover, entire image is brighter; Fig. 4 (h), Fig. 4 (i) and Fig. 4 (j) by we can see that the entire image contrast is fine among the figure, do not occur general very bright result for improving the enhancing result of unsharp masking algorithm, have especially obviously obtained the enhancing effect in edge.

Because human vision is difficult to characterize with mathematical method to the evaluation of picture quality, therefore various method for adaptive image enhancement also are difficult to compare to the performance of Flame Image Process, usually, select some criterions as various Enhancement Method benchmark relatively, in the process that the Enhancement Method performance is assessed, present embodiment will use Y-PSNR (PSNR), and computing formula is:

$\mathrm{PSNR}=10\·\lg \left(\frac{{255}^2}{\frac{1}{M\×N}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{(I_{\mathrm{um}}(i,j)-I(i,j))}^2}\right)$

In the formula, N, M are the height and width of image, and I (i, j), I _Um(i j) is the original image and the gray scale of image after treatment respectively;

Table 1 be improved unsharp masking Enhancement Method with classical inverse sharpening mould Enhancement Method to different noise peak signal to noise ratio (S/N ratio) contrast tables, as shown in Table 1: the Y-PSNR of the image after handling through improving one's methods is big by 20% than classic method all, illustrates that this paper method is better than classic method.

Table 1. tradition and the Y-PSNR contrast table of the anti-sharpening enhancement algorithms of improvement to different noises

Fig. 5 is an another application of the invention: the result that improved unsharp masking method is applied to medical image: by Fig. 5 (a) as can be known, the original image contrast is low, and noise pollution is serious, and bone contours is not obvious; By Fig. 5 (b) as can be known, after traditional unsharp masking strengthened image, contrast was lower, and entire image is brighter, and bone contours does not have clear showing; By Fig. 5 (c) as can be known, after improved unsharp masking strengthened image, the contrast of image increased, and the profile of bone can clearly show.

Claims (2)

1. improved unsharp masking image enchancing method based on the logarithmic image transaction module is characterized in that: may further comprise the steps:

Step 1: will be under conventional model convert under the logarithmic image transaction module based on the image of gray tone based on the input picture of gray scale, formula is:

$I=M-\stackrel{\‾}{I}$

In the formula, I represents half tone picture, and M represents the maximum gradation value matrix,

The expression gray level image;

Step 2: image is carried out improved unsharp masking strengthen;

Step 3: the unsharp masking that step 2 is obtained strengthen image transitions be under the conventional model based on image gray, formula is:

${\stackrel{\‾}{I}}_{\mathrm{um}}=M-I_{\mathrm{um}}$

In the formula,

Expression strengthens result's gray level image, and M represents the maximum gradation value matrix, and big or small identical and each the value of this matrix and image array all is the maximum gradation value of image, I _UmExpression strengthens result's half tone picture.

2. the improved unsharp masking image enchancing method based on the logarithmic image transaction module according to claim 1, it is characterized in that: step 2 is described carries out improved unsharp masking enhancing to image, image after step 1 is handled, while execution in step 2-1 and step 2-2, method is as follows:

Step 2-1: the gradient factor g of computed image, at first utilize Gauss's smoothing operator to carrying out filtering based on image gray under the conventional model, formula is:

I _s＝I*G _σ

In the formula, I _sFor the input picture based on gray scale under the conventional model is carried out the later image of gaussian filtering, G _σFor variance is the Gaussian filter of σ;

Then, utilize the gradient information Gray of difference method computed image;

At last, the gradient factor g of computed image, formula is:

$g=\frac{1}{1+\mathrm{Gray}/m}$

Image is carried out unsharp masking strengthen, formula is:

I _um＝I+g×(I-I _us)

In the formula, I _UsThe image of expression after low-pass filtering, formula is:

$I_{\mathrm{us}}(x,y)=\frac{1}{M\×N}\×\underset{i=x-(M-1)/2}{\overset{x+(M-1)/2}{\mathrm{\Σ}}}\underset{j=y-(N-1)/2}{\overset{y+(N-1)/2}{\mathrm{\Σ}}}I(i,j)$

In the formula, N, M are the height and width of image, and (x, y represent the current horizontal ordinate of handling image pixel point respectively to I for i, j) expression original image;

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula, Expression is carried out image after the anti-isomorphism conversion to half tone picture;

Step 2-2: image is carried out improved unsharp masking strengthen: at first, handle according to isomorphism transfer pair half tone picture, transformation for mula is:

$\stackrel{\‾}{I}=\mathrm{\φ}\left(I\right)=-M\ln (1-\frac{I}{M})$

In the formula, φ (I) expression is carried out image after the isomorphism conversion to half tone picture;

Then, utilize mean filter that the input picture based on gray scale under the conventional model is carried out low-pass filtering treatment, formula is:

$I_{\mathrm{us}}(x,y)=\frac{1}{M\×N}\×\underset{i=x-(M-1)/2}{\overset{x+(M-1)/2}{\mathrm{\Σ}}}\underset{j=y-(N-1)/2}{\overset{y+(N-1)/2}{\mathrm{\Σ}}}I(i,j)$

In the formula, N, M are the height and width of image, I (i, j) expression original image;

Afterwards, utilize marginal information as weight factor original image to be carried out the unsharp masking enhancement process, the operation of this moment is traditional linear operation, and its formula is:

I _um＝I+g×(I-I _us)

At last, utilize the above-mentioned image of anti-isomorphism transfer pair to handle, formula is as follows:

$I={\mathrm{\φ}}^{-1}\left(\stackrel{\‾}{I}\right)=M(1-e^{-\stackrel{\‾}{I}/M})$

In the formula,

Expression is carried out image after the anti-isomorphism conversion to half tone picture.

Images