CN102222326A - Method and device for deblurring images based on single low resolution - Google Patents

Abstract

The invention provides a method and a device for deblurring an image based on single low resolution, relating to the technical field of image processing. The method for deblurring the image based on single low resolution comprises the following steps of: filtering an image to be processed to obtain a filter image; convoluting the filtered image by means of a Laplace operator to obtain a gradient image; correcting the gradient image based on the gradient difference value of a neighbourhood; and overlapping the image to be processed with the corrected gradient image to obtain a sharpened image. By adopting the method and the device for deblurring the image based on single low resolution of the invention, the noise of a smooth region in the gradient image is reduced and the profile and detail losses of an object are lessened during the sharpening process of the single low-resolution image, so that the visual effect of the image is greatly improved.

Description

A kind of image deblurring method and device based on single low-resolution

Technical field

The present invention relates to technical field of image processing, relate in particular to a kind of image deblurring method and device based on single low-resolution.

Background technology

High-definition picture has purposes very widely in actual life.For example,, can improve spectators' visual experience if Digital Television (DTV) signal can be converted into high-definition television (HDTV) signal at digital TV field; In the safety detection field, though generally all camera can be installed, the image that obtains is very fuzzy, and high-definition picture can either be assisted usual safety management, can help the police to handle a case when merit takes place again.

Occur various raising image resolution ratios in recent years, improved the method for picture quality, can be divided into two classes: a class is based on many low-resolution images, at first by estimating that the anglec of rotation and offset distance proofread and correct image, utilize again modes such as interpolation, reconstruction or study obtain a width of cloth resolution higher, comprise the more image of information.Though these class methods can obtain result preferably, they limit to some extent for the kinematic parameter between the image; The another kind of single low-resolution image that is based on carries out interpolation to it, because it only needs low-resolution image itself, so applicability is wider.

The interpolation method that extensively adopts comprises neighbor interpolation method, bilinear interpolation and bicubic interpolation method in practice, and wherein the amplification effect of bicubic interpolation method is best, but these interpolation methods all can cause the fuzzy of image.Fuzzy problem at image, generally need carry out sharpening to image, the most frequently used is adopts Sobel operator and Laplace operator to carry out joint operation, strengthens the details of having been blured, and for example notification number is that the Chinese patent of CN2626003Y discloses a kind of image deblurring scheme.As shown in Figure 1, this scheme specifically may further comprise the steps:

S1: (x y) carries out second-order differential, i.e. Laplace conversion obtains the gradient image Laplace (F (x shown in Fig. 2 (a) to input picture F, y)) and to input picture F (x, y) carry out the single order differential, i.e. Sobel conversion obtains the (F (x, y)) of the gradient image Sobel shown in Fig. 2 (b);

S2: to the image behind the single order differential, be gradient image Sobel (F (x, y)) smoothly obtain the image A verage (F (x shown in Fig. 2 (c) with Weighted Mean Filter, y)), with described image A verage (image after F (x, y)) and the second-order differential conversion, i.e. gradient image Laplace (F (x, y)) multiply each other and obtain (the F (x, y)) of the mask image Coverage shown in Fig. 2 (d);

Therefore second-order differential has stronger response to isolated point, and comparing the single order differential, can to produce the noise of more noise, particularly smooth region very obvious.Because the single order differential is strong at the response ratio second-order differential of gray ramp or ladder, but a little less than the response ratio second-order differential to noise and little details, particularly by can further reducing noise behind the mean filter.

S3: with described mask image Coverage (F (x, y)) and original image F (and x, y) superpose just can obtain after the deblurring shown in the output map 2 (e) also sharpening image G (x, y).

Adopt above-mentioned image deblurring scheme both to keep details in the strong zone of grey scale change, again in the area decreases of grey scale change relatively flat noise.But such result is that cost obtains with the partial contour and the details of loss object.

Summary of the invention

(1) technical matters that will solve

The technical problem to be solved in the present invention is, at above-mentioned defective, how a kind of image deblurring method and device based on single low-resolution of deblurring better effects if are provided, can be in the sharpening process of single low-resolution image, both reduced the noise of smooth region in the gradient image, make the profile and the loss of detail of object less again, thereby the visual effect of image is greatly improved.

(2) technical scheme

For solving the problems of the technologies described above, the invention provides a kind of image deblurring method based on single low-resolution, comprising:

Step C, pending image is carried out filtering obtain filtering image;

Step D, employing Laplace operator carry out convolution to described filtering image and obtain gradient image;

Step e, according to the described gradient image of gradient difference correction of neighborhood;

Step F, described pending image and the stack of revised described gradient image are obtained the sharpening image.

Wherein, before described step C, described image deblurring method further comprises:

Steps A, the original image of rgb format is converted to the pending image of LAB form;

After described step F, described image deblurring method further comprises:

Step G, export after the sharpening image transitions that obtains become rgb format.

Wherein, before step C, described image deblurring method further comprises:

Step B, described pending image is carried out interpolation.

Wherein, before described step B, described image deblurring method further comprises:

Steps A, the original image of rgb format is converted to the pending image of LAB form;

After described step F, described image deblurring method further comprises:

Step G, export after the sharpening image transitions that obtains become rgb format.

Wherein, described step B specifically comprises: adopt the bicubic interpolation method that described pending image is carried out interpolation.

Wherein, described step e comprises:

Step e 1, Grad is made as 0 less than the Grad of the point of predetermined threshold;

Step e 2, compute gradient value are not the gradient difference of the neighborhood point of 0 point and its preset range;

Step e 3, compute gradient difference are not the average of 0 point;

Step e 4, compute gradient value are not the weight of 0 point;

Step e 5, compute gradient value are not 0 the revised Grad of point.

Wherein, the mode of filtering described in the step C is a bilateral filtering.

The present invention also provides a kind of image deblurring device based on single low-resolution, comprising:

The image filtering unit is used for that pending image is carried out filtering and obtains filtering image;

The gradient image generation unit, the filtering image that adopts the Laplace operator that described image filtering unit is generated carries out convolution and obtains gradient image;

The gradient image amending unit is used for the gradient image that generates according to the described gradient image generation unit of the gradient difference correction of neighborhood;

The image overlay unit is used for described pending image and the revised gradient image stack of described gradient image amending unit are obtained the sharpening image.

Wherein, described image deblurring device based on single low-resolution further comprises:

Interpolating unit is used for described pending image is carried out offering described image filtering unit after the interpolation.

Wherein, described image deblurring device based on single low-resolution further comprises:

Image format conversion unit is used for offering described image filtering unit or described interpolating unit behind the pending image that original image with rgb format is converted to the LAB form;

The image output unit is exported after being used for the sharpening image transitions that described image overlay unit obtains become rgb format.

Wherein, described gradient image amending unit comprises:

The Grad discriminating module is used for Grad is made as 0 less than the Grad of the point of predetermined threshold;

Neighborhood gradient difference calculating module, the Grad that is used to calculate the setting of described Grad discriminating module are not the gradient difference of the neighborhood point of 0 point and its preset range;

The mean value computation module is used to calculate gradient difference that described neighborhood gradient difference calculating module obtains and is not the average of 0 point;

Weight computation module, the Grad that is used to calculate the setting of described Grad discriminating module are not the weight of 0 point;

The Grad computing module, the Grad that is used to calculate the setting of described Grad discriminating module is not 0 the revised Grad of point.

(3) beneficial effect

The invention discloses a kind of image deblurring method and device based on single low-resolution.According to the present invention, RGB image transitions that will input reaches the purpose that keeps try one's best broad colour gamut and abundant colors for the LAB form after, the bicubic interpolation method of employing standard is carried out interpolation, utilize bilateral filtering to remove noise and sharpen edges, according to Laplace operator compute gradient, the weight correction Grad that obtains according to the gradient difference value of neighborhood strengthens the profile of object in the image and reduces the noise of smooth region, revised gradient image and interpolation image stack are obtained the sharpening result, change back at last rgb format output, can reduce the noise of smooth region in the gradient image, make the profile and the loss of detail of object in the gradient image less again, reached and removed image blurring purpose after the interpolation.

Description of drawings

Fig. 1 is the process flow diagram of the image deblurring method based on single low-resolution of the prior art;

Fig. 2 (a)-Fig. 2 (e) is based on the effect synoptic diagram of the image deblurring method of single low-resolution described in Fig. 1;

Fig. 3 is the process flow diagram of the described image deblurring method based on single low-resolution of the embodiment of the invention;

Fig. 4 (a)-Fig. 4 (f) is the effect synoptic diagram based on the image deblurring method of single low-resolution described in Fig. 3;

Fig. 5 is the structural representation of the described image deblurring device based on single low-resolution of the embodiment of the invention.

Embodiment

Below in conjunction with drawings and Examples, the specific embodiment of the present invention is described in further detail.Following examples are used to illustrate the present invention, but are not used for limiting the scope of the invention.

Core concept of the present invention is: at first the RGB image transitions with input is the LAB form, keeps try one's best broad colour gamut and abundant colors; The bicubic interpolation method of employing standard is carried out interpolation and is obtained the initial high-definition picture of a width of cloth; The weight correction Grad of removing noise and sharpen edges by interpolation image being carried out bilateral filtering then, utilizing the Grad of image after the Laplace operator calculation of filtered to obtain marginal information, obtain according to the gradient difference value of neighborhood, strengthen the profile of object in the image and reduce the noise of smooth region, revised gradient image and interpolation image stack are obtained the sharpening result; Change back at last rgb format output, remove the fuzzy purpose of interpolation image thereby reach.

As shown in Figure 3, the image deblurring method based on single low-resolution of the present invention comprises:

Steps A, the original image I of rgb format that will be shown in Fig. 4 (a) _InBe converted to the pending image I of LAB form _Lab

The Lab colour model is by illumination L (being equivalent to brightness) and a relevant with color, and three key elements of b are formed.It has the equipment of not relying on and the broad advantage of colour gamut, has remedied the deficiency of rgb color model COLOR COMPOSITION THROUGH DISTRIBUTION inequality, and is helpful aspect the adjustment image definition.Work under the Lab colour model in addition, speed and RGB are almost fast.Therefore in order to keep try one's best broad colour gamut and abundant colors, the RGB image with input among the present invention is converted to the LAB form earlier, carries out Flame Image Process under the Lab colour model, converts RGB after Flame Image Process is finished again to and shows.

Step B, adopt the pending image I of bicubic interpolation method to the LAB form _LabCarry out the pending image I after interpolation obtains interpolation shown in Fig. 4 (b) _Bcu

Shown in (1):

$k\left(x\right)\left\{\begin{array}{cc}1-2{\left|x\right|}^2+{\left|x\right|}^3& \left|x\right|<1\\ 4-8\left|x\right|+5{\left|x\right|}^2-{\left|x\right|}^3& 1\&le;\left|x\right|<2\\ 0& \mathrm{otherwise}\end{array}\right.---\left(1\right)$

The bicubic interpolation method is according to the weight coefficient of cube form the pixel value of 16 points around the sampled point to be carried out interpolation, weight coefficient k gets the segmentation cube form of side-play amount x, the original image of input among Fig. 3 (a), after being converted to the LAB form, be interpolated to 2*2 times of former figure with the bicubic interpolation method, and change back the RGB image shown in Fig. 3 (b).Can find that the image after the interpolation has thickened, therefore needs adopt the method for image sharpening to recover the details of losing.

Because bilateral filtering can make edge of image sharper keen, has kept edge of image in filtering image noise as far as possible, therefore the result who asks gradient behind the interpolation image bilateral filtering again is better than directly interpolation image being asked the result of gradient.Specifically comprise:

Step C, to the pending image I after the interpolation _BcuCarry out bilateral filtering and obtain filtering image I shown in Fig. 4 (c) _Bf

Bilateral filtering is the form that the weight coefficient of traditional Gauss wave filter is optimized to the monochrome information product of Gauss function and image, and the weight coefficient after the optimization is made convolution algorithm with image again.Coordinate position is at (x, the central point I that y) locates _Bcu(x, y) pixel value that carries out behind the bilateral filtering is I _Bf(x y), is calculated by following formula:

$I_{\mathrm{bf}}(x,y)=\underset{\mathrm{\&Delta;y}=-j}{\overset{j}{\mathrm{\&Sigma;}}}\underset{\mathrm{\&Delta;x}=-j}{\overset{j}{\mathrm{\&Sigma;}}}c(\mathrm{\&Delta;x},\mathrm{\&Delta;y})\&times;s(\mathrm{\&Delta;x},\mathrm{\&Delta;y})\&times;I_{\mathrm{bcu}}(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})/\mathrm{Nor}---\left(2\right)$

$\mathrm{Nor}=\underset{\mathrm{\&Delta;y}=-j}{\overset{j}{\mathrm{\&Sigma;}}}\underset{\mathrm{\&Delta;x}=-j}{\overset{j}{\mathrm{\&Sigma;}}}c(\mathrm{\&Delta;x},\mathrm{\&Delta;y})\&times;s(\mathrm{\&Delta;x},\mathrm{\&Delta;y})---\left(3\right)$

$c(\mathrm{\&Delta;x},\mathrm{\&Delta;y})=e^{-\frac{1}{2}\left(\frac{\mathrm{\&Delta;}{x}^2+\mathrm{\&Delta;}{y}^2}{{\mathrm{\&sigma;}}_d^2}\right)}---\left(4\right)$

$s(\mathrm{\&Delta;x},\mathrm{\&Delta;y})=e^{-\frac{1}{2}\left(\frac{{\left|\right|I_{\mathrm{bcu}}(x,y)-I_{\mathrm{bcu}}(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})\left|\right|}^2}{{\mathrm{\&sigma;}}_s^2}\right)---\left(5\right)}$

$\left|\right|I_{\mathrm{bcu}}(x,y)-I_{\mathrm{bcu}}(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})\left|\right|=\sqrt{{\left(\mathrm{\&Delta;L}\right)}^2+{\left(\mathrm{\&Delta;A}\right)}^2+{\left(\mathrm{\&Delta;B}\right)}^2}---\left(6\right)$

Wherein j is the neighborhood scope of carrying out filtering, and Nor is a normalized factor, and c (Δ x, Δ y), s (Δ x, Δ y) have described some I respectively _Bcu(x, y) with the some I _BcuThe influence of the distance difference between (x+ Δ x, y+ Δ y) and the influence of value differences, σ _dBe that criterion distance is poor, σ _sBe the standard deviation of pixel value, Δ L, Δ A, Δ B are respectively L, the A of pixel value, the difference of B characteristic component.

The variable that influences the bilateral filtering result comprises j, σ _dAnd σ _s, j is big more, and then to participate in the point of filtering many more, σ _dThe rate of decay of big more then distance affects is slow more, and the image that obtains is fuzzy more, but σ _sCan be to σ _dVariation make compensation, to value differences greater than σ _sIts influence of pixel reduce rapidly, greatly kept the characteristic information of edge in the image.

Get parameter j=3, σ _d=5, σ _s=5 pairs of images 4 (b) carry out bilateral filtering, change back the rgb format result displayed shown in Fig. 4 (c), can find that the image border that obtains is sharper keen, and image is more level and smooth.

Step D, employing Laplace operator are to described filtering image I _BfCarry out convolution and obtain gradient image I _Grad

Because second-order differential has stronger response to for example details such as fine rule and isolated point, it is better therefore to form the ability that strengthens details, and when gray-value variation was similar in image, point was stronger than the response of line, and line is stronger than the response of ladder.Therefore concerning strengthening image detail, second-order differential is better than single order differential.

Therefore adopt the Laplace operator among the present invention to filtering image I _BfCarry out convolution and promptly obtain gradient image I by following formula _Grad:

I _grad＝f _laplace*I _bf (7)

Be to adopt the Laplace template to calculate in the practical application, step is that the Laplace template is roamed in image, and Laplace template center overlapped with certain pixel among the figure, this pixel multiply by value corresponding in the Laplace template respectively with the gray-scale value of the pixel of its neighborhood, and is that addition obtains and be our desired value promptly.Its expression formula is shown below:

$R=\left[\begin{array}{ccc}f1& f2& f3\\ f4& f5& f6\\ f7& f8& \mathrm{<figure-callout\; id="9"\; label="f"\; filenames="HDA0000071921700000021.png"\; state="\{\{state\}\}">f</figure-callout>}9\end{array}\right]*\left[\begin{array}{ccc}I1& I2& I3\\ I4& I5& I6\\ I7& I8& I9\end{array}\right]=f1I1+f2I2+...+f9I9=\underset{i=1}{\overset{9}{\mathrm{\&Sigma;}}}{f}_i{I}_i---\left(8\right)$

The Laplace template is:

$L=\left|\begin{array}{ccc}-1& -1& -1\\ -1& 8& -1\\ -1& -1& -1\end{array}\right|---\left(9\right)$

Because it is second-order differential has stronger response to isolated point, therefore very obvious at the noise of smooth region.The Laplace gradient image that directly interpolation image is calculated in the prior art shown in the Laplace gradient image that the present invention shown in the comparison diagram 4 (d) obtains and Fig. 2 (a), Fig. 4 (d) at the noise of smooth region still less, and profile is more clear, but still has noise spot in the part of left and right sides cat head and body of the tower.

Step e, according to the described gradient image I of gradient difference correction of neighborhood _Grad

In order to remove the noise of smooth region, consider the difference of neighborhood gradient among the present invention, give gradient point different weights, revise Grad.This step specifically comprises:

Step e 1, Grad is made as 0 less than the Grad of the point of predetermined threshold;

In this step, ignore the influence of the minimum point of Grad, can be with Grad I _Grad(x, absolute value y) is made as 0 less than the gradient of the point of predetermined threshold T, that is:

$I_{\mathrm{grad}}(x,y)\left\{\begin{array}{cc}0& \left|I_{\mathrm{grad}}(x,y)\right|<T\\ I_{\mathrm{grad}}(x,y)& \left|I_{\mathrm{grad}}(x,y)\right|\&GreaterEqual;T\end{array}\right.---\left(10\right)$

Step e 2, compute gradient be not 0 point (x, y) with its preset range for example ± the gradient difference dI of the neighborhood point of k (x, y):

$\mathrm{dI}(x,y)=\frac{\underset{\mathrm{\&Delta;y}=-k}{\overset{k}{\mathrm{\&Sigma;}}}\underset{\mathrm{\&Delta;x}=-k}{\overset{k}{\mathrm{\&Sigma;}}}|I_{\mathrm{grad}}(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})-I_{\mathrm{grad}}(x,y)|}{k^2-1}---\left(11\right)$

Step e 3, compute gradient difference dI (x y) is not the average m of 0 point:

$m=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{dI}}_i}{n}---\left(12\right)$

Wherein n is that gradient difference dI is not the number of 0 point in the image.

Step e 4, compute gradient value be not 0 point (x, weight w y) (x, y):

$w(x,y)=e^{1-\frac{m}{\mathrm{dI}(x,y)+\mathrm{\&delta;}}}---\left(13\right)$

Wherein δ is a minimum positive number, prevents that denominator from being 0.

If the gradient difference of this point is less than the average m of gradient difference, then weight is less than 1, and the gradient difference is more little, and the weight decay is fast more; If the gradient difference of this point is greater than the average m of gradient difference, then weight is greater than 1, and weight slowly rises until e with the increase of gradient difference.

Step e 5, compute gradient value are not 0 point (x, y) revised Grad

${\tilde{I}}_{\mathrm{grad}}(x,y)=I_{\mathrm{grad}}(x,y)\&times;w(x,y)---\left(14\right)$

Get parameter T=0.1, k=3, correction result is carried out shown in Fig. 4 (e) in δ=0.001 couple Fig. 4 (d), can find that the noise of smooth region has obtained effective inhibition.Adopt Sobel operator and Laplace operator that the bicubic interpolation image is carried out Fig. 2 (d) that the associating sharpening obtains simultaneously in the comparison diagram 2, can find that the edge of window among Fig. 4 (e) is sharper keen clear.

Step F, with the described pending image I after the interpolation _BcuObtain the sharpening image I with revised described gradient image stack _Sharp:

$I_{\mathrm{sharp}}(x,y)=I_{\mathrm{bcu}}(x,y)+{\tilde{I}}_{\mathrm{grad}}(x,y)---\left(15\right)$

Step G, with the sharpening image I that obtains _SharpExport after converting rgb format to.

Fig. 4 (f) is sharpening output result of the present invention, adopts the sharpening of Sobel operator and Laplace operator to export Fig. 2 (e) as a result in the comparison diagram 2, and Fig. 2 (e) is in too sharpening of Ta Yanchu, and the yellowing variable color must be compared severity, and Fig. 4 (f) more presses close to original input picture.

By above step, in the RGB image transitions with input is the LAB form, after reaching the purpose that keeps try one's best broad colour gamut and abundant colors, the bicubic interpolation method of employing standard is carried out interpolation, the weight correction Grad that utilizes bilateral filtering to remove noise and sharpen edges, obtain according to Laplace operator compute gradient, according to the gradient difference value of neighborhood, strengthen the profile of object in the image and reduce the noise of smooth region, revised gradient image and interpolation image stack are obtained the sharpening result; Change back at last rgb format output, can reduce the noise of smooth region in the gradient image, make the profile of object in the gradient image and loss of detail less again, reached and removed image blurring purpose after the interpolation.

Said method is applicable to the fuzzy problem after the solution gray level image interpolation too, gets step B-step F and gets final product, and at this moment, the difference of LAB form pixel value becomes the difference of yuv format brightness value in the formula (6):

$\left|\right|I_{\mathrm{bcu}}(x,y)-I_{\mathrm{bcu}}(x+\mathrm{\&Delta;x},y+\mathrm{\&Delta;y})\left|\right|=\sqrt{{\left(\mathrm{\&Delta;Y}\right)}^2}---\left(16\right)$

Wherein, Δ Y represents the difference of the luminance component Y in the YUV component.

More generally, said method also can be directly used in the original colour and the fuzzy problem of gray level image, strengthens fuzzy details, and coloured image is got steps A, step C-step G, gray level image is only got step C-step F get final product.

If no special instructions, the order in the above-mentioned various step combination is all carried out according to the order of English alphabet.

Corresponding with the image deblurring method based on single low-resolution of the present invention, the present invention also provides a kind of image deblurring device based on single low-resolution.

As shown in Figure 5, the image deblurring device based on single low-resolution of the present invention comprises: the image filtering unit is used for that pending image is carried out bilateral filtering and obtains filtering image;

The gradient image generation unit, the filtering image that adopts the Laplace operator that described image filtering unit is generated carries out convolution and obtains gradient image;

The gradient image amending unit is used for the gradient image that generates according to the described gradient image generation unit of the gradient difference correction of neighborhood;

The image overlay unit is used for described pending image and the revised gradient image stack of described gradient image amending unit are obtained the sharpening image.

Therefore, the weight correction Grad that the present invention removes noise and sharpen edges by pending image being carried out bilateral filtering, utilize the Grad of image after the Laplace operator calculation of filtered to obtain marginal information, obtain according to the gradient difference value of neighborhood, strengthen the profile of object in the image and reduce the noise of smooth region, revised gradient image and pending image overlay are obtained the sharpening result; Remove pending image blurring purpose thereby reach.

Further, described image deblurring device based on single low-resolution also comprises:

Interpolating unit is used for described pending image is carried out offering described image filtering unit after the interpolation;

Can solve the fuzzy problem after the gray level image interpolation further.

Further, described image deblurring device based on single low-resolution also comprises:

Image format conversion unit is used for offering described image filtering unit or described interpolating unit behind the pending image that original image with rgb format is converted to the LAB form;

The image output unit is exported after being used for the sharpening image transitions that described image overlay unit obtains become rgb format.

Can further solve the fuzzy problem of the coloured image after original or the interpolation.

Described gradient image amending unit comprises:

The Grad discriminating module is used for Grad is made as 0 less than the Grad of the point of predetermined threshold;

Neighborhood gradient difference calculating module, the Grad that is used to calculate the setting of described Grad discriminating module are not the gradient difference of the neighborhood point of 0 point and its preset range;

The mean value computation module is used to calculate gradient difference that described neighborhood gradient difference calculating module obtains and is not the average of 0 point;

Weight computation module, the Grad that is used to calculate the setting of described Grad discriminating module are not the weight of 0 point;

The Grad computing module, the Grad that is used to calculate the setting of described Grad discriminating module is not 0 the revised Grad of point.

In sum, the invention discloses a kind of image deblurring method and device based on single low-resolution, according to the present invention, in the RGB image transitions with input is the LAB form, after reaching the purpose that keeps try one's best broad colour gamut and abundant colors, the bicubic interpolation method of employing standard is carried out interpolation, utilize bilateral filtering to remove noise and sharpen edges, according to Laplace operator compute gradient, the weight correction Grad that obtains according to the gradient difference value of neighborhood, strengthen the profile of object in the image and reduce the noise of smooth region, revised gradient image and interpolation image stack are obtained the sharpening result; Change back at last rgb format output, remove the fuzzy purpose of interpolation image, can reduce the noise of smooth region in the gradient image, make the profile of object in the gradient image and loss of detail less again, reached and removed image blurring purpose after the interpolation thereby reach.

The present invention is applicable to the fuzzy problem of the fuzzy problem that solves original colorful image, original gray level image and the fuzzy problem after the gray level image interpolation too.

Above embodiment only is used to illustrate the present invention; and be not limitation of the present invention; the those of ordinary skill in relevant technologies field; under the situation that does not break away from the spirit and scope of the present invention; can also make various variations and modification; therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (11)

1. the image deblurring method based on single low-resolution is characterized in that, comprising:

Step C, pending image is carried out filtering obtain filtering image;

Step D, employing Laplace operator carry out convolution to described filtering image and obtain gradient image;

Step e, according to the described gradient image of gradient difference correction of neighborhood; With

Step F, described pending image and the stack of revised described gradient image are obtained the sharpening image.

2. the image deblurring method based on single low-resolution according to claim 1 is characterized in that, before described step C, further comprises:

Steps A, the original image of rgb format is converted to the pending image of LAB form;

After described step F, further comprise:

Step G, export after the sharpening image transitions that obtains become rgb format.

3. the image deblurring method based on single low-resolution according to claim 1 is characterized in that, before step C, further comprises:

Step B, described pending image is carried out interpolation.

4. the image deblurring method based on single low-resolution according to claim 3 is characterized in that, before described step B, further comprises:

Steps A, the original image of rgb format is converted to the pending image of LAB form;

After described step F, further comprise:

Step G, export after the sharpening image transitions that obtains become rgb format.

5. according to claim 3 or 4 described image deblurring methods, it is characterized in that described step B specifically comprises: adopt the bicubic interpolation method that described pending image is carried out interpolation based on single low-resolution.

6. according to each described image deblurring method among the claim 1-4, it is characterized in that described step e comprises based on single low-resolution:

Step e 1, Grad is made as 0 less than the Grad of the point of predetermined threshold;

Step e 2, compute gradient value are not the gradient difference of the neighborhood point of 0 point and its preset range;

Step e 3, compute gradient difference are not the average of 0 point;

Step e 4, compute gradient value are not the weight of 0 point; With

Step e 5, compute gradient value are not 0 the revised Grad of point.

7. according to each described image deblurring method among the claim 1-4, it is characterized in that the mode of filtering described in the step C is a bilateral filtering based on single low-resolution.

8. the image deblurring device based on single low-resolution is characterized in that, comprising:

The image filtering unit is used for that pending image is carried out filtering and obtains filtering image;

The gradient image generation unit, the filtering image that adopts the Laplace operator that described image filtering unit is generated carries out convolution and obtains gradient image;

The gradient image amending unit is used for the gradient image that generates according to the described gradient image generation unit of the gradient difference correction of neighborhood; With

The image overlay unit is used for described pending image and the revised gradient image stack of described gradient image amending unit are obtained the sharpening image.

9. the image deblurring device based on single low-resolution according to claim 8 is characterized in that, further comprises:

Interpolating unit is used for described pending image is carried out offering described image filtering unit after the interpolation.

10. according to Claim 8 or 9 described image deblurring devices, it is characterized in that, further comprise based on single low-resolution:

Image format conversion unit is used for offering described image filtering unit or described interpolating unit behind the pending image that original image with rgb format is converted to the LAB form; With

The image output unit is exported after being used for the sharpening image transitions that described image overlay unit obtains become rgb format.

11. according to Claim 8 or 9 described image deblurring devices, it is characterized in that described gradient image amending unit comprises based on single low-resolution:

The Grad discriminating module is used for Grad is made as 0 less than the Grad of the point of predetermined threshold;

Neighborhood gradient difference calculating module, the Grad that is used to calculate the setting of described Grad discriminating module are not the gradient difference of the neighborhood point of 0 point and its preset range;

The mean value computation module is used to calculate gradient difference that described neighborhood gradient difference calculating module obtains and is not the average of 0 point;

Weight computation module, the Grad that is used to calculate the setting of described Grad discriminating module are not the weight of 0 point; With

The Grad computing module, the Grad that is used to calculate the setting of described Grad discriminating module is not 0 the revised Grad of point.

Images