CN104680485A - Method and device for denoising image based on multiple resolutions - Google Patents

Abstract

The invention relates to a method and a device for denoising an image based on multiple resolutions. The method comprises the following steps of performing multi-order decomposing on the original image information by a multi-resolution decomposing method, so as to obtain the image information of each order; performing reverse decomposing reconstructing on the image information of each order, so as to obtain an approximate subband image corresponding to the image information of each order; performing bilateral filtering on the approximate subband image corresponding to the image information of each order, so as to obtain the image information after the image information of each order is denoised; according to the image information result after the image information of initial order is denoised, obtaining the image information after the original image information is denoised, wherein the image information of initial order is the order level image information corresponding to the original image information. The method has the advantage that the denoising effect of the image can be effectively improved.

Description

A kind of image de-noising method based on multiresolution and device

Technical field

The present invention relates to image processing field, particularly relate to a kind of image de-noising method based on multiresolution and device.

Background technology

Image denoising is a kind of Application comparison technology widely in Image semantic classification, and the object of image denoising is the signal to noise ratio (S/N ratio) in order to improve image, the desired character of outstanding image.Image is easily subject to the impact of various factors in the process obtained and transmit, and makes the image collected by imageing sensor comprise noisy image often.

Mix containing noisy noise in image signal and picture signal due to described, make the problems such as image existing characteristics is not obvious, sharpness is not high, so usually to need imageing sensor collect image and carry out denoising to improve the signal to noise ratio (S/N ratio) of image, improve the display effect of image.

Different qualities that denoising can utilize noise signal and picture signal usually on frequency domain is carried out to image and processes, because picture signal is mainly distributed in low frequency region, and noise signal is mainly distributed in high-frequency region, based on this feature, existing method of image being carried out to denoising in prior art, such as bilateral filtering method, Wavelet noise-eliminating method etc.Described bilateral filtering method can pay close attention to space length relation between neighborhood pixels and gray-scale relation in the process of filtering simultaneously.Described Wavelet noise-eliminating method is carrying out image in the process of denoising, being then mainly choosing of noise-removed threshold value, and choosing of noise-removed threshold value is the key of Wavelet Denoising Method, depends primarily on choosing of described noise-removed threshold value to the removal effect of picture noise.

But the effect adopting various image de-noising method to carry out denoising to image in prior art is not fine, still there are some problems.Such as, although described bilateral filtering method effectively can be removed the noise signal of high-frequency region, but be distributed in low frequency region owing to there being partial noise signal, described bilateral filtering method effectively can not be removed the noise information existed in low frequency region, and, because the detail section of image is if marginal information is also respectively at HFS, while the denoising effect of the HFS reached, the loss of the detail section of soft edge, image may be caused.Described Wavelet noise-eliminating method then exists when described noise-removed threshold value is chosen improperly time, denoising effect also can be undesirable, such as when noise-removed threshold value is chosen larger, in the process of image denoising, the detailed information of some images can be removed, if and when noise-removed threshold value is chosen less, the problem that denoising dynamics is inadequate can be caused again.When adopting the image de-noising method of prior art to carry out denoising to image, image retention noise may be there is, the problems such as soft edge.

Correlation technique can be the U.S. Patent application of US2008166064A1 with reference to publication number.

Various image de-noising method can be adopted in prior art to carry out denoising to image, such as bilateral filtering method, Wavelet noise-eliminating method etc.

Described bilateral filtering method can pay close attention to space length relation between neighborhood pixels and gray-scale relation in the process of filtering simultaneously, use double wave filtering can preserve image border preferably, and then it is effective smoothing to noise, reach the object of denoising, formula (1) can be adopted to carry out double wave filtering.

$\tilde{I}\left(x\right)=\frac{1}{C}\underset{y\∈N\left(x\right)}{\mathrm{\Σ}}{e}^{\frac{{-\left|\right|y-x\left|\right|}^2}{{2\mathrm{\σ}}_d^2}}{e}^{\frac{{-\left|\right|I\left(y\right)-I\left(x\right)\left|\right|}^2}{{2\mathrm{\σ}}_r^2}}I\left(y\right)---\left(1\right)$

Wherein, C is normalized factor, and described normalized factor C form of Definition as shown in Equation (2).

$C=\underset{y\∈N\left(x\right)}{\mathrm{\Σ}}{e}^{\frac{{-\left|\right|y-x\left|\right|}^2}{{2\mathrm{\σ}}_d^2}}{e}^{\frac{{-\left|\right|I\left(y\right)-I\left(x\right)\left|\right|}^2}{{2\mathrm{\σ}}_r^2}}---\left(2\right)$

Wherein, in formula (1) and formula (2), σ _dfor space length parameter, σ _rfor Gray homogeneity parameter, N (x) is a spatial neighborhood centered by pixel x, and y is a pixel of described spatial neighborhood, and I (x), I (y) are the gray-scale values of pixel x, y, for the gray-scale value of pixel x after bilateral filtering.

The effect of bilateral filtering is by described space length parameter σ _dwith Gray homogeneity parameter σ _rdetermine, σ _dand σ _rrough phenomenon is just there will be close to 0 image, as long as σ once one of them amount _rvariation range within certain scope, just the edge of image is not affected, σ _rcompare σ _dthe details of easier effect diagram picture, from formula (1), σ in bilateral filtering _rand σ _dthe form be multiplied, as long as this just means σ _rand σ _din one close to 0, image just there will be rough phenomenon.

In described Wavelet noise-eliminating method, wavelet transformation is carried out to image, to when carrying out wavelet transformation containing noisy image, the wavelet coefficient of image can be obtained, described wavelet coefficient can be divided into two classes, first kind wavelet coefficient primarily of picture signal detail section caused by, also include the mapped structure of picture noise signal, the amplitude of wavelet coefficient is comparatively large, and number is less, and Equations of The Second Kind wavelet coefficient is primarily of caused by picture noise signal, the amplitude of wavelet coefficient is less, and number is more.

Wavelet noise-eliminating method based on threshold value is the noise suppressing method based on nonparametric model, and the Wavelet noise-eliminating method based on threshold value can pass through formula (3) and realize.

${\mathrm{\σ}}_w=\frac{\mathrm{median}\left(\right|w_{i,j}\left|\right)}{0.06754}---\left(3\right)$

Wherein σ _wfor the noise variance of image, w _i,jfor the pixel value through wavelet transformation after of image in (i, j) position, median represents and carries out median calculation.

Based on the noise variance σ of formula (3) determined image _w, the noise-removed threshold value T used in Wavelet noise-eliminating method can be obtained by formula (4) _h.

$T_h={\mathrm{\σ}}_w\sqrt{2\ln M}---\left(4\right)$

Wherein, σ _wfor the noise variance of image, M is the sum of all pixels of image.

Described Wavelet noise-eliminating method is carrying out image in the process of denoising, being then mainly choosing of noise-removed threshold value, and choosing of noise-removed threshold value is the key of Wavelet Denoising Method, depends primarily on choosing of described noise-removed threshold value to the removal effect of picture noise.

But all there are some problems in the effect of the image denoising that various image de-noising method of the prior art produces, such as cannot the noise signal being in low frequency region effectively be removed, or in the process of image denoising, can simultaneously by the problem of the detailed information removal in image, after prior art carries out denoising to image, image retention noise may be there is, the problems such as soft edge, therefore, the details of image how can be retained while reducing picture noise, keep the edge of clear picture, namely how obtaining better denoising effect has just become current Image Denoising Technology to need the problem solved.

In order to solve the problem, technical solution of the present invention provides a kind of image de-noising method based on multiresolution.

Fig. 1 is the schematic flow sheet of the image de-noising method based on multiresolution that technical solution of the present invention provides, and as shown in Figure 1, first performs step S101, carries out multistage decomposition, to obtain every single order image information based on Multiresolution Decomposition method to original image information.

Because picture signal is mainly distributed in low frequency region, and noise signal is mainly distributed in high-frequency region, and effectively can find noise based on Multiresolution Decomposition method, so in step S101, first based on Multiresolution Decomposition method, multistage decomposition is carried out to original image information, to obtain the image information of every single order.Original image information can be decomposed into the image information of multiple different resolution by described Multiresolution Decomposition method, such as original image information has the highest resolution, and the low frequency sub-band of the first rank image information obtained after being decomposed by original image information and the resolution of high-frequency sub-band can be respectively 1/4th of the resolution of original image information, the like, the low frequency sub-band of image information obtained after decomposing each time and high-frequency sub-band all only decompose before image information 1/4th resolution, the multistage decomposition of image information can be carried out according to the resolution of image order from high to low.

In the process of carrying out Multiresolution Decomposition, can using the current low frequency region image information will carrying out the image information of decomposing as input image information, and then carry out Multiresolution Decomposition based on described input image information, obtain low frequency region image information and the high-frequency region image information of lower single order, using the input picture of the low frequency region image information of lower single order as single order image information after it, again Multiresolution Decomposition is carried out based on this input picture, the like, can from the original image information with highest resolution, after repeatedly decomposing based on Multiresolution Decomposition method, obtain multistage image information, can using original image information as initial rank image information, the minimum image information of the resolution that last decomposition obtains is as most High-order Image information.

Described Multiresolution Decomposition method can be Wavelet Transform, Gauss's Pyramid transform method, picture contrast Pyramid transform method or gradient pyramid decomposition method etc.

Perform step S102, inverse decomposition is carried out to every single order image information and rebuilds, to obtain the approximation subband image corresponding to every single order image information.

After obtaining multistage image information, just can carry out denoising to every single order image information, when denoising, first need to carry out inverse decomposition to the multistage image information obtained to rebuild, to obtain the approximation subband image corresponding to every single order image information, to carry out denoising to described approximation subband image.

Described inverse decompose rebuild can according in step S101 obtain multistage image information resolution order from low to high carry out, namely first carry out inverse decomposition to most High-order Image information to rebuild, obtain most approximation subband image corresponding to High-order Image information, denoising can be carried out afterwards to the approximation subband image corresponding to most High-order Image information, image information after denoising is carried out the inverse input picture decomposing reconstruction operation as front single order image, high-frequency region information based on described input picture and described front single order image information carries out inverse decomposition reconstruction, obtain the approximation subband image corresponding to described front single order image information, the like, can from the most High-order Image information with lowest resolution, before higher to resolution based on the image information after the rear single order denoising that resolution is lower, single order image information carries out inverse decomposition reconstruction, obtain the approximation subband image corresponding to every single order image information.

It is corresponding that step S102 carries out the inverse Multiresolution Decomposition method adopted in the Methods and steps S101 rebuild of decomposing.Such as, if step S101 adopts Gauss Pyramid transform method, then can adopt in step s 102 and decompose method for reconstructing with inverse corresponding to described Gauss's Pyramid transform method.

For the ease of understanding, the inverse process of reconstruction of decomposing that multistage decomposable process described in step S101 and step S102 carry out can reference diagram 2, as shown in Figure 2,0 rank image information represents original image information, described 0 rank image information also can be called initial rank, and described original image information has the highest resolution.N rank are most High-order Image information, and described most High-order Image information has minimum resolution, original image information obtain after multistage decomposition 1 rank, 2 rank ..., n-2 rank, n-1 rank, n rank image information.Described multistage decomposable process, according to direction shown in Fig. 2 left arrow, carries out from having the image information of high-resolution image information to low resolution, and described inverse decomposition is rebuild and carried out to high-resolution image information according to the image information with low resolution.

Please continue to refer to Fig. 1, perform step S103, bilateral filtering is carried out, to obtain the image information after the denoising of every single order image information to the approximation subband image corresponding to described every single order image information.

In step s 102, for the approximation subband image that every single order image information can obtain corresponding thereto, in step s 103 bilateral filtering is carried out to described approximation subband image, to obtain the image information after the denoising of every single order image information.

Adopt the method for bilateral filtering to carry out denoising to it for every single order image information, the image information after the denoising of current rank can carry out the inverse input image information decomposing reconstruction as its front single order image information.

Perform step S104, based on the image information result after the image information denoising of initial rank, obtain the image information after original image information denoising, described initial rank image information refers to the stratum's image information corresponding to original image information.

In above-mentioned steps, obtain the multistage image of image information based on Multiresolution Decomposition method after, carry out according to the resolution order from low to high of multistage image information, before higher to resolution based on the image information after the rear single order denoising that resolution is lower, single order image information carries out inverse process of decomposing reconstruction, denoising is carried out through successive ignition, inverse process of decomposing reconstruction, can by last inverse process of decomposing reconstruction, obtain the image information after the inverse decomposition reconstruction corresponding to original image, after again bilateral filtering being carried out to it, just can obtain the image information after original image information (also can be called initial rank information) denoising, namely final image denoising result is obtained.

The method is in the process of image denoising, Multiresolution Decomposition method and bilateral filtering method are combined, the noise information of image at different frequency domain effectively can be obtained by Multiresolution Decomposition method, and then effectively can remove the noise information of the high frequency of image, low frequency region in conjunction with the method for bilateral filtering, effectively improve the denoising effect of image.

For enabling above-mentioned purpose of the present invention, feature and advantage more become apparent, and are described in detail specific embodiments of the invention below in conjunction with accompanying drawing.

Summary of the invention

The problem that the present invention solves is that in prior art, image denoising effect is undesirable and cannot carry out to the noise signal of low frequency region the problem effectively removed.

For solving the problem, technical solution of the present invention provides a kind of image de-noising method based on multiresolution, and described method comprises:

Based on Multiresolution Decomposition method, multistage decomposition is carried out to original image information, to obtain every single order image information;

Carry out inverse decomposition to every single order image information to rebuild, to obtain the approximation subband image corresponding to every single order image information;

Bilateral filtering is carried out, to obtain the image information after the denoising of every single order image information to the approximation subband image corresponding to described every single order image information;

Based on the image information result after the image information denoising of initial rank, obtain the image information after original image information denoising, described initial rank image information refers to the stratum's image information corresponding to original image information.

Optionally, described every single order image information comprises low frequency region image information and high-frequency region image information, describedly comprises the process that original image information carries out multistage decomposition based on Multiresolution Decomposition method method:

Described multistage decomposition is carried out according to the resolution order from high to low of image, using the input information of the low frequency region image information of higher for resolution front single order as the lower rear single order of resolution, based on described input information, decompose the low frequency region image information and the high-frequency region image information that obtain the lower rear single order of resolution, wherein, to there is the original image information of highest resolution as initial rank image information, using image information minimum for resolution as most High-order Image information.

Optionally, describedly the inverse process of rebuilding of decomposing carried out to every single order image information comprise:

Described inverse decomposition is rebuild and is carried out according to the resolution order from low to high of multistage image information, and before higher to resolution based on the image information after the rear single order denoising that resolution is lower, single order image information carries out rebuilding against decomposing.

Optionally, described method also comprises:

After obtaining the approximation subband image corresponding to every single order image information, before bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, Anscombe conversion is carried out to described approximation subband image;

After bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, before obtaining the image information after the denoising of every single order image information, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

Optionally, describedly based on Multiresolution Decomposition method, multistage decomposition is carried out to original image information, comprises to obtain every single order image information:

Carry out multistage decomposition respectively based on multiple passages of Multiresolution Decomposition method to original image information, to obtain every single order image information of each passage respectively, described multiple passage comprises Y, U, V tri-passages of image information.

Optionally, every single order image information of described each passage comprises low frequency region image information and high-frequency region image information, describedly comprises the process that multiple passages of original image information carry out multistage decomposition respectively based on Multiresolution Decomposition method:

Described multistage decomposition is carried out according to the resolution order from high to low of image, for each passage, using the input information of the low frequency region image information of higher for the resolution of this passage front single order as the lower rear single order of the resolution of this passage, based on described input information, decompose the low frequency region image information and high-frequency region image information that obtain the lower rear single order of the resolution of this passage, wherein, the initial rank image information of image information as this passage of each passage of the original image information of highest resolution will be had, using the most High-order Image information of the image information of each passage minimum for resolution as this passage.

Optionally, describedly inverse decomposition carried out to every single order image information rebuild, comprise with the approximation subband image obtained corresponding to every single order image information:

Carry out inverse decomposition respectively to every single order image information of each passage to rebuild, with the approximation subband image corresponding to the every single order image information obtaining each passage.

Optionally, described every single order image information to each passage carries out comprising against decomposing the process of rebuilding respectively:

Described inverse decomposition is rebuild and is carried out according to the resolution order from low to high of the multistage image information of this passage, and before higher based on the resolution of image information to this passage after the rear single order denoising that the resolution of this passage is lower, single order image information carries out rebuilding against decomposing.

Optionally, described bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, comprises with the process of the image information after obtaining the denoising of every single order image information:

Respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, with the image information after the every single order image information denoising obtaining each passage.

Optionally, described method also comprises:

After approximation subband image corresponding to the every single order image information obtaining each passage, before respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, Anscombe conversion is carried out to described approximation subband image;

After the approximation subband image corresponding to the every single order image information to each passage carries out bilateral filtering respectively, before obtaining the image information after every single order image information denoising of each passage, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

Optionally, described based on the image information result after the image information denoising of initial rank, the image information obtained after original image information denoising comprises:

Based on the image information result after the initial rank image information denoising of each passage, obtain the image information after original image information denoising, the initial rank image information of described each passage refers to the stratum's image information in each passage corresponding to original image information.

Optionally, described Multiresolution Decomposition method comprises any one in Wavelet Transform, Gauss's Pyramid transform method, picture contrast Pyramid transform method and gradient pyramid decomposition method.

Optionally, described method also comprises: convert based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _r.

Optionally, the exponent number carrying out the image information that inverse decomposition is rebuild is larger, the Gray homogeneity parameter σ used in described bilateral filtering _rless.

Optionally, describedly convert based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _rcomprise:

Based on formula: σ _r=f (σ, layer) determines the Gray homogeneity parameter σ in bilateral filtering _r, wherein, σ converts the image noise variance obtained for Anscombe, and layer carries out the inverse exponent number decomposing the image information of rebuilding, and f (σ, layer) is for asking for the Gray homogeneity parameter σ used in bilateral filtering _rfunction.

Optionally, described Gray homogeneity parameter wherein, σ converts the image noise variance obtained for Anscombe, and layer carries out the inverse exponent number decomposing the image information of rebuilding.

Optionally, identical parameter is used when bilateral filtering being carried out respectively to the approximation subband image corresponding to the same single order image information of each passage.

Optionally, described identical parameter comprises described Gray homogeneity parameter σ _r.

Technical solution of the present invention also provides a kind of image denoising device based on multiresolution, and described device comprises:

Resolving cell, is suitable for carrying out multistage decomposition based on Multiresolution Decomposition method to original image information, obtains every single order image information;

Reconstruction unit, is suitable for carrying out inverse decomposition to every single order image information and rebuilds, to obtain the approximation subband image corresponding to every single order image information;

Filter unit, is suitable for carrying out bilateral filtering, to obtain the image information after the denoising of every single order image information to the approximation subband image corresponding to described every single order image information;

Obtain unit, be suitable for the image information result after based on the image information denoising of initial rank, obtain the image information after original image information denoising, described initial rank image information refers to the stratum's image information corresponding to original image information.

Optionally, described device also comprises:

First converter unit, be suitable for after reconstruction unit obtains the approximation subband image corresponding to every single order image information, filter unit carries out Anscombe conversion to described approximation subband image before carrying out bilateral filtering to the approximation subband image corresponding to described every single order image information.

Optionally, described filter unit comprises: the first inverse transformation subelement, be suitable for after bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, before obtaining the image information after the denoising of every single order image information, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

Optionally, described resolving cell comprises decomposition subelement, be suitable for carrying out multistage decomposition respectively based on multiple passages of Multiresolution Decomposition method to original image information, to obtain every single order image information of each passage respectively, described multiple passage comprises Y, U, V tri-passages of image information.

Optionally, described reconstruction unit comprises reconstruction subelement, is suitable for carrying out inverse decomposition respectively to every single order image information of each passage and rebuilds, obtain the approximation subband image corresponding to every single order image information of each passage.

Optionally, described filter unit comprises filtering subelement, carries out bilateral filtering respectively, obtain the image information after every single order image information denoising of each passage to the approximation subband image corresponding to every single order image information of each passage.

Optionally, described device also comprises: the second converter unit, be suitable for after rebuilding the approximation subband image corresponding to every single order image information that subelement obtains each passage, filtering subelement carries out Anscombe conversion to described approximation subband image before carrying out bilateral filtering respectively to the approximation subband image corresponding to every single order image information of each passage.

Optionally, described filtering subelement comprises the second inverse transformation subelement, be suitable for after the approximation subband image corresponding to the every single order image information to each passage carries out bilateral filtering respectively, before obtaining the image information after every single order image information denoising of each passage, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

Optionally, described acquisition unit comprises acquisition subelement, be suitable for the image information result after based on the initial rank image information denoising of each passage, obtain the image information after original image information denoising, the initial rank image information of described each passage refers to the stratum's image information in each passage corresponding to original image information.

Optionally, described device also comprises determining unit, is suitable for converting based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _r.

Compared with prior art, technical scheme of the present invention has the following advantages:

Based on Multiresolution Decomposition method, multistage decomposition is carried out to original image information and obtain multistage image information, bilateral filtering is carried out to by the approximation subband image obtained after every single order image reconstruction of described multistage image, obtain the image information after every single order image denoising, finally can obtain the image information after corresponding to original image information denoising.The method is in the process of image denoising, the noise signal of image at different frequency domain effectively can be obtained by Multiresolution Decomposition method, and then effectively can remove the noise signal of the high frequency of image, low frequency region in conjunction with the method for bilateral filtering, effectively improve the denoising effect of image.

Multistage decomposition is carried out respectively based on multiple passages of Multiresolution Decomposition method to original image information, to obtain every single order image information of each passage, carry out inverse decomposition respectively to every single order image information of each passage to rebuild, obtain the approximation subband image corresponding to every single order image information of each passage, respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, obtain the image information after every single order image information denoising of each passage, based on the image information result after the original image information denoising of each passage, finally can obtain the image information after corresponding to original image information denoising.Multiresolution Decomposition method and bilateral filtering method can be combined, effectively can remove the noise information of low frequency region, effectively improve the denoising effect of image.

To every single order image of described multistage image or every single order image of each passage, before the approximation subband image carrying out obtaining after inverse decomposition is rebuild carries out bilateral filtering, the poisson noise existed in image information can be converted to the noise of Gaussian distribution by variance stability conversion (such as being converted by Anscombe), thus denoising can be carried out under Gauss model, obtain effective denoising effect.

Based on the Gray homogeneity parameter of carrying out the inverse exponent number decomposing the image information of rebuilding and the image noise variance obtained in being converted by variance stability and determining to use in bilateral filtering, can carry out in the process of bilateral filtering to the image information of not same order, use the Gray homogeneity parameter adapted with this rank image information, achieve adaptive denoising, improve image denoising effect.

Same single order image information for each passage is carried out in the process of bilateral filtering, the denoising of chrominance channel will be used for based on the determined Gray homogeneity parameter of luminance channel and the window weight that calculates thus, can effectively avoid false colors edge or texture, avoid the fuzzy of marginalisation, obtain effective global de-noising effect.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the image de-noising method based on multiresolution that technical solution of the present invention provides;

Fig. 2 is the Multiresolution Decomposition that technical solution of the present invention provides and the schematic diagram rebuild against decomposition;

Fig. 3 is the schematic flow sheet of the image de-noising method based on multiresolution that the embodiment of the present invention one provides;

Fig. 4 is the schematic flow sheet of the image de-noising method based on multiresolution that the embodiment of the present invention two provides;

Fig. 5 is the schematic flow sheet of the image de-noising method based on multiresolution that the embodiment of the present invention three provides;

Fig. 6 is the schematic flow sheet of the image de-noising method based on multiresolution that the embodiment of the present invention four provides.

Embodiment

Embodiment one

In the present embodiment, carry out in the process of denoising based on Multiresolution Decomposition method to original image information, before the approximation subband image obtained after being rebuild by inverse decomposition the multistage image obtained carries out bilateral filtering, by variance stability conversion (such as Anscombe conversion), the poisson noise existed in image information is converted to the noise of Gaussian distribution.

Fig. 3 is the schematic flow sheet of the image de-noising method based on multiresolution that the present embodiment provides, and as shown in Figure 3, first performs step S301, carries out multistage decomposition, to obtain every single order image information based on Multiresolution Decomposition method to original image information.Step S301 please refer to step S101.

Perform step S302, inverse decomposition is carried out to every single order image information and rebuilds, to obtain the approximation subband image corresponding to every single order image information.Step S302 please refer to step S102.

Perform step S303, Anscombe conversion is carried out to the approximation subband image corresponding to described every single order image information.

Anscombe conversion is carried out to the approximation subband image obtained in step S302, the poisson noise existed in image information can be converted to the noise of Gaussian distribution, and then the variance yields of noise can be obtained, can be carried out under the denoising of image information being transformed into Gauss model by described Anscombe conversion.

Described Anscombe conversion can realize based on formula (5).

$f\left(z\right)=2\sqrt{z+\frac{3}{8}}---\left(5\right)$

Wherein z is the value of the pixel of obeying Poisson distribution, the value of the pixel that f (z) is Gaussian distributed.

Anscombe based on formula (5) converts, and can obtain f (z) approximate obedience noise variance is the Gaussian distribution of 1.

Perform step S304, bilateral filtering is carried out to the approximation subband image carried out corresponding to the every single order image information after Anscombe conversion.

The method of bilateral filtering is adopted to carry out denoising to it for the approximation subband image corresponding to the every single order image information after Anscombe conversion.

Perform step S305, Anscombe inverse transformation is carried out to the approximation subband image corresponding to the every single order image information after bilateral filtering.

Carry out Anscombe inverse transformation to the approximation subband image after the bilateral filtering obtained in step S304, the method that the method for inverse transformation can adopt those skilled in the art to know carries out inverse transformation, does not repeat them here.

Perform step S306, obtain the image information after the denoising of every single order image information.

Based on obtain in step S305 carry out Anscombe inverse transformation after image information namely can obtain the image information after the image information denoising of current rank.

Perform step S307, based on the image information result after the image information denoising of initial rank, obtain the image information after original image information denoising.Step S307 please refer to step S104.

In the present embodiment, before the approximation subband image obtained after every single order image information of described multistage image being carried out to inverse decomposition and rebuilding carries out bilateral filtering, the poisson noise existed in image information can be converted to the noise of Gaussian distribution by Anscombe conversion, thus denoising can be carried out under Gauss model, obtain effective denoising effect.

Embodiment two

In the present embodiment, carrying out in the process of denoising based on Multiresolution Decomposition method to original image information, based on carry out inverse decompose the image information of rebuilding exponent number and convert by variance stability the Gray homogeneity parameter that the image noise variance obtained in (such as Anscombe converts) determines to use in bilateral filtering.

Fig. 4 is the schematic flow sheet of the image de-noising method based on multiresolution that the present embodiment provides, and as shown in Figure 4, first performs step S401, carries out multistage decomposition, to obtain every single order image information based on Multiresolution Decomposition method to original image information.

Perform step S402, inverse decomposition is carried out to every single order image information and rebuilds, to obtain the approximation subband image corresponding to every single order image information.

Perform step S403, Anscombe conversion is carried out to the approximation subband image corresponding to described every single order image information.

Step S401 to step S403 please refer to embodiment one step S301 to step S303.

Perform step S404, convert the Gray homogeneity parameter of use when the image noise variance determination pairing approximation sub-band images obtained carries out bilateral filtering based on Anscombe.

In step S403 when carrying out Anscombe conversion to described approximation subband image, image noise variance σ can be obtained, owing to considering that the approximation subband image that resolution is lower can adopt less Gray homogeneity parameter σ when carrying out bilateral filtering _r, problem excessively fuzzy after can avoiding image information denoising like this, so the Gray homogeneity parameter σ carrying out bilateral filtering _rcan be associated with the current inverse exponent number decomposing the image information of rebuilding of multiresolution that carries out.The exponent number of the less correspondence of the resolution due to image information is larger, so, when the exponent number carrying out the image information that inverse decomposition is rebuild is larger, the Gray homogeneity parameter σ now used in described bilateral filtering _rcan be less.

Described Gray homogeneity parameter σ can be determined based on formula (6) _r.

σ _r=f(σ,layer) （6）

Wherein, σ converts the image noise variance obtained for Anscombe, and layer carries out the inverse exponent number decomposing the image information of rebuilding, and f (σ, layer) is for asking for the Gray homogeneity parameter σ used in bilateral filtering _rfunction.

Described f (σ, layer) can be the inversely proportional function of image noise variance σ and layer, such as described Gray homogeneity parameter σ _rformula (7) can be passed through determine.

${\mathrm{\σ}}_r=\frac{\mathrm{\σ}}{\mathrm{layer}}---\left(7\right)$

Perform step S405, bilateral filtering is carried out to the approximation subband image carried out corresponding to the every single order image information after Anscombe conversion.

Based on Gray homogeneity parameter σ determined in step S404 _rbilateral filtering is carried out to the approximation subband image corresponding to every single order image information.

Perform step S406, Anscombe inverse transformation is carried out to the approximation subband image corresponding to the every single order image information after bilateral filtering.

Perform step S407, obtain the image information after the denoising of every single order image information.

Perform step S408, based on the image information result after the image information denoising of initial rank, obtain the image information after original image information denoising.

Step S406 to step S408 please refer to embodiment one step S305 to step S307.

In the present embodiment, when carrying out bilateral filtering to image information, without the need to artificial, filtering parameter is set, the inverse exponent number decomposing the image information of rebuilding can be carried out according to current, adaptively determine Gray homogeneity parameter, can carry out in the process of bilateral filtering to the image information of not same order, use the Gray homogeneity parameter adapted with this rank image information, realize the effect of adaptive denoising.

Embodiment three

Color image information can be divided into the image information of multiple passage, in the present embodiment, is described for multichannel color image information.Fig. 5 is the schematic flow sheet of the image de-noising method based on multiresolution that the present embodiment provides, as shown in Figure 5, first step S501 is performed, multistage decomposition is carried out respectively, to obtain every single order image information of each passage respectively based on multiple passages of Multiresolution Decomposition method to original image information.

The image information of described multiple passage can be the image information comprising Y, U, V tri-passages.

Step S501 can refer step S101, in the present embodiment, need all to perform to multiple passage the operation being similar to step S101, for Y, U, V passage, need to carry out multistage decomposition to the original image information of Y passage based on Multiresolution Decomposition method, to obtain every single order image information of Y passage, also need to carry out multistage decomposition based on Multiresolution Decomposition method respectively to U, V passage, to obtain every single order image information of U passage and V passage respectively.

Described multistage decomposition is carried out according to the resolution order from high to low of image, for each passage, using the input information of the low frequency region image information of higher for the resolution of this passage front single order as the lower rear single order of the resolution of this passage, based on described input information, decompose the low frequency region image information and high-frequency region image information that obtain the lower rear single order of the resolution of this passage, wherein, the initial rank image information of image information as this passage of each passage of the original image information of highest resolution will be had, using the most High-order Image information of the image information of each passage minimum for resolution as this passage.

Perform step S502, inverse decomposition is carried out respectively to every single order image information of each passage and rebuilds, with the approximation subband image corresponding to the every single order image information obtaining each passage.

Step S502 can refer step S102, the difference of step S502 and step S102 is, in the present embodiment, need to carry out inverse decomposition respectively to every single order image information of each passage to rebuild, the every single order image information for each passage can obtain corresponding approximation subband image.

Described inverse decomposition is rebuild and is carried out according to the resolution order from low to high of the multistage image information of this passage, and before higher based on the resolution of image information to this passage after the rear single order denoising that the resolution of this passage is lower, single order image information carries out rebuilding against decomposing.

Decomposing process of reconstruction for inverse described in the multistage decomposable process described in step S501 and step S502 still can reference diagram 2, for each passage need to perform similar Fig. 2 multistage decomposition, inversely decompose the process of rebuilding.

Perform step S503, respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, with the image information after the every single order image information denoising obtaining each passage.

Perform step S504, based on the image information result after the initial rank image information denoising of each passage, obtain the image information after original image information denoising.

The initial rank image information of described each passage refers to the stratum's image information in each passage corresponding to original image information.

In the method, for having multichannel image information, Multiresolution Decomposition method and bilateral filtering method being combined, effectively can remove the noise information of low frequency region, effectively improve the denoising effect of image.

Embodiment four

In the present embodiment, be described to carry out denoising based on the image de-noising method of multiresolution to the image information with Y, U, V tri-passages.

In denoising process, before the approximation subband image obtained after being rebuild by inverse decomposition the multistage image of each passage obtained carries out bilateral filtering, by Anscombe conversion the poisson noise existed in image information is converted to the noise of Gaussian distribution, and based on the Gray homogeneity parameter of carrying out the inverse exponent number decomposing the image information of rebuilding and the image noise variance obtained in being converted by Anscombe and determining to use in bilateral filtering.

Fig. 6 is the schematic flow sheet of the image de-noising method based on multiresolution that the present embodiment provides, as shown in Figure 6, first step S601 is performed, multistage decomposition is carried out respectively, to obtain every single order image information of each passage respectively based on multiple passages of Multiresolution Decomposition method to original image information.

When described image color space is YUV color space, then can carries out multistage decomposition respectively based on Multiresolution Decomposition method Y, U, V tri-passages to original image, in the present embodiment, the image of each passage is all decomposed into 4 rank image informations.

For Y passage, it can be 0 rank image information of described Y passage by the image information identifier of the Y passage of original image information, based on Multiresolution Decomposition method, described 0 rank image information is decomposed into the image information of high-frequency region and the image information of low frequency region of the 1st rank image information, the described low frequency sub-band of the 1st rank image information and the resolution of high-frequency sub-band can be respectively 1/4th of 0 rank image information; The input image information of the 2nd rank image information is further used as with the image information of the low frequency region of described 1st rank image information, based on described input image information, adopt Multiresolution Decomposition method described 1st rank image information to be decomposed into the image information of high-frequency region and the image information of low frequency region of the 2nd rank image information, the described low frequency sub-band of the 2nd rank image information and the resolution of high-frequency sub-band can be respectively 1/4th of the 1st rank image information; By that analogy, can obtain the 3rd rank image information, the described low frequency sub-band of the 3rd rank image information and the resolution of high-frequency sub-band can be respectively 1/4th of 2 rank image informations.Adopt said method, 0 rank, 1 rank, 2 rank and 3 rank image informations can be obtained altogether.In like manner, method all 4 rank image informations of U passage and all 4 rank image informations of V passage can be obtained accordingly.

Perform step S602, the inverse higher-order image information of rebuilding of decomposing is not carried out to each passage and carries out inverse decomposition respectively and rebuild, with the approximation subband image corresponding to this rank image information obtaining each passage.

After the multistage image information obtaining Y, U, V tri-passages based on step S601, when rebuilding image information, first step S602 can be passed through, to the most high-order of each passage, namely the 3rd rank image information carries out inverse decomposition reconstruction, the approximation subband image corresponding to the 3rd rank image information of Y passage can be obtained, in like manner can obtain respectively the 3rd rank image information of U passage V passage distinguish corresponding approximation subband image.If current in step S601, not carry out the inverse higher-order image information of rebuilding of decomposing be the 2nd rank image information (also can be understood as is do not carry out inverse most High-order Image information of decomposing in the image information of rebuilding), what then complete in this step is carry out inverse decomposition respectively to the 2nd rank image information of Y, U, V tri-passages to rebuild, the approximation subband image corresponding to the 2nd rank image information of Y, U, V tri-passages can be obtained respectively, the like, this approximation subband image corresponding to image information of rank obtaining each passage.

It should be noted that, in this step, carry out in the process of rebuilding against decomposition respectively in the higher-order image information of each passage not being carried out to inverse decomposition reconstruction, be based on adjacent with this rank resolution and lower than this rank image resolution ratio denoising after image information carry out inverse decomposition and rebuild, for example, carry out inverse decomposition respectively to the 2nd rank image information of Y passage to rebuild if current, be then based on the 3rd rank image information denoising of Y passage after image information carry out inverse decomposition and rebuild.

Carry out inverse decomposition respectively and rebuild not carried out the inverse higher-order image information of rebuilding of decomposing by each passage in Y, U, V tri-passages, after this approximation subband image corresponding to image information of rank obtaining each passage, perform step S603.

Step S603, carries out Anscombe conversion to the approximation subband image corresponding to this rank image information of described each passage.

Step S603 please refer to embodiment one step S303, be with step S303 difference, that Anscombe conversion is carried out to the approximation subband image corresponding to this rank image information of each passage in this step, need to carry out Anscombe conversion respectively to the approximation subband image corresponding to this rank image information of Y, U, V tri-passages at this, such as Anscombe conversion is carried out to the approximation subband image corresponding to the 3rd rank image information of Y passage, respectively Anscombe conversion is carried out to the approximation subband image corresponding to the 3rd rank image information of U, V passage.

Perform step S604, convert the Gray homogeneity parameter of use when the image noise variance obtained is determined to carry out bilateral filtering to described approximation subband image based on Anscombe.

Can the Gray homogeneity parameter σ of reference example two step S404 use when determining to carry out bilateral filtering to described approximation subband image _r, owing to there being Y, U, V tri-passages, so accordingly when the approximation subband image corresponding to this rank image information to each passage carries out Anscombe conversion, three the image noise variance σs corresponding respectively with described Y, U and V tri-passages can be obtained _y, σ _uand σ _v, and then can determine after Anscombe conversion, the Gray homogeneity parameter σ of use when this of Y, U and V tri-passages approximation subband image corresponding to image information of rank carries out filtering _rvalue can be with

Perform step S605, bilateral filtering is carried out to the approximation subband image corresponding to this rank image information of carrying out each passage after Anscombe conversion.

Based on the Gray homogeneity parameter σ of the passage of correspondence determined in step S604 _rvalue, carries out bilateral filtering to the approximation subband image corresponding to this rank image information of current channel.

For example, if what obtain in step S604 is the approximation subband image corresponding to the 3rd rank image information after Y passage carries out Anscombe conversion, be then that filtering is carried out to the approximation subband image that Y passage carries out corresponding to the 3rd rank image information after Anscombe conversion herein, the image information after the 3rd rank image information denoising of Y passage can be obtained.Image information for every single order of U, V passage is adopted to use the same method and is carried out bilateral filtering.

Perform step S606, Anscombe inverse transformation is carried out to the approximation subband image corresponding to this rank image information of each passage after bilateral filtering.

Perform step S607, obtain the image information after this rank image information denoising of each passage.

Based on obtain in step S606 carry out Anscombe inverse transformation after image information can obtain the image information after this rank image information denoising.

Perform step S608, judge that whether the resolution of this rank image information of each passage be the resolution of the initial rank image information of this passage.

In step S608, need to determine that whether this rank image information resolution of each passage be the resolution of the initial rank image information of this passage.

After image information after the denoising of this rank image information obtaining each passage based on step S607, perform step S608, judge that whether the resolution of this rank image information of each passage be resolution corresponding to the initial rank image information of this passage, namely judge the resolution of the 0 rank image information of resolution whether corresponding to the original image of each passage of the image information corresponding to these rank of each passage.If so, then obtain the image information after original image denoising, perform step S609; Otherwise return and perform step S602.

For example, if decompose in step S601 and obtain totally 4 rank image informations, most high-order is the 3rd rank, if what then obtain in step S607 is image information after the 3rd rank image information denoising of Y, U, V tri-passages, then step S602 should be returned, continue to decompose the higher-order image information of rebuilding to not the carrying out of Y, U, V tri-passages is inverse, namely the 2nd rank image information carries out rebuilding against decomposing.If step S607 obtains being the image information after the 0th rank image information denoising of Y, U, V tri-passages, then obtain the image information after original image denoising, should step S609 be performed.

Step S609, terminates the denoising to image information.

It will be appreciated by those skilled in the art that, image de-noising method based on multiresolution described above, also other conversion embodiment can be had, such as, in the present embodiment, in step S602 in step S608, be all that the image information of the phase same order of multiple passage is operated respectively accordingly in each step, can obtain in each step the image information of multiple passage phase same order distinguish corresponding operating result.In other embodiments, also first only can perform to one of them passage the step being similar to step S602 to step S608, namely first above-mentioned steps is performed to single passage, obtain the image information after the original image information denoising of single passage, successively respectively above-mentioned steps is performed to other passage again, obtain the image information after the original image information denoising of each passage respectively, and then the final denoising result of original image information can be obtained equally.

And for example, in the present embodiment, have employed Anscombe conversion simultaneously, and the method for the Gray homogeneity parameter used in bilateral filtering is determined based on the image noise variance obtained in Anscombe conversion, in other embodiments, also can only include Anscombe transform method, not limit at this.

In addition, in the present embodiment, the Gray homogeneity parameter σ of the use when the approximation subband image corresponding to this rank image information that step S604 determines each passage carries out bilateral filtering _rtime, determine different Gray homogeneity parameters respectively according to different passages in other embodiments, the Gray homogeneity parameter corresponding to Y passage can also only be determined , and U, V passage adopts when filtering by the determined Gray homogeneity parameter of Y passage .This is due to Y passage, i.e. containing the information such as edge, texture in the image information of luminance channel, so utilize determined based on the luminance channel containing information such as described edge, textures when bilateral filtering is carried out to U and V chrominance channel, can effectively prevent because in certain chrominance channel information edge or texture information not obvious and cause denoising dynamics not identical with the denoising dynamics of other chrominance channel, finally cause false colors edge or texture, or cause edge fog.Therefore, same single order image information for each passage is carried out in the process of bilateral filtering, the denoising of chrominance channel will be also used for based on the determined Gray homogeneity parameter of luminance channel and the window weight that calculates thus, can effectively avoid false colors edge or texture, avoid the fuzzy of marginalisation, obtain effective global de-noising effect.

Although the present invention discloses as above, the present invention is not defined in this.Any those skilled in the art, without departing from the spirit and scope of the present invention, all can make various changes or modifications, and therefore protection scope of the present invention should be as the criterion with claim limited range.

Claims (28)

1. based on an image de-noising method for multiresolution, it is characterized in that, comprising:

Based on Multiresolution Decomposition method, multistage decomposition is carried out to original image information, to obtain every single order image information;

Carry out inverse decomposition to every single order image information to rebuild, to obtain the approximation subband image corresponding to every single order image information;

Bilateral filtering is carried out, to obtain the image information after the denoising of every single order image information to the approximation subband image corresponding to described every single order image information;

Based on the image information result after the image information denoising of initial rank, obtain the image information after original image information denoising, described initial rank image information refers to the stratum's image information corresponding to original image information.

2. as claimed in claim 1 based on the image de-noising method of multiresolution, it is characterized in that, described every single order image information comprises low frequency region image information and high-frequency region image information, describedly comprises the process that original image information carries out multistage decomposition based on Multiresolution Decomposition method method:

Described multistage decomposition is carried out according to the resolution order from high to low of image, using the input information of the low frequency region image information of higher for resolution front single order as the lower rear single order of resolution, based on described input information, decompose the low frequency region image information and the high-frequency region image information that obtain the lower rear single order of resolution, wherein, to there is the original image information of highest resolution as initial rank image information, using image information minimum for resolution as most High-order Image information.

3. as claimed in claim 1 based on the image de-noising method of multiresolution, it is characterized in that, describedly the inverse process of rebuilding of decomposing is carried out to every single order image information comprise:

Described inverse decomposition is rebuild and is carried out according to the resolution order from low to high of multistage image information, and before higher to resolution based on the image information after the rear single order denoising that resolution is lower, single order image information carries out rebuilding against decomposing.

4. the image de-noising method based on multiresolution as described in any one of claims 1 to 3, is characterized in that, also comprise:

After obtaining the approximation subband image corresponding to every single order image information, before bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, Anscombe conversion is carried out to described approximation subband image;

After bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, before obtaining the image information after the denoising of every single order image information, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

5. as claimed in claim 1 based on the image de-noising method of multiresolution, it is characterized in that, describedly based on Multiresolution Decomposition method, multistage decomposition carried out to original image information, comprise to obtain every single order image information:

Carry out multistage decomposition respectively based on multiple passages of Multiresolution Decomposition method to original image information, to obtain every single order image information of each passage respectively, described multiple passage comprises Y, U, V tri-passages of image information.

6. as claimed in claim 5 based on the image de-noising method of multiresolution, it is characterized in that, every single order image information of described each passage comprises low frequency region image information and high-frequency region image information, describedly comprises the process that multiple passages of original image information carry out multistage decomposition respectively based on Multiresolution Decomposition method:

Described multistage decomposition is carried out according to the resolution order from high to low of image, for each passage, using the input information of the low frequency region image information of higher for the resolution of this passage front single order as the lower rear single order of the resolution of this passage, based on described input information, decompose the low frequency region image information and high-frequency region image information that obtain the lower rear single order of the resolution of this passage, wherein, the initial rank image information of image information as this passage of each passage of the original image information of highest resolution will be had, using the most High-order Image information of the image information of each passage minimum for resolution as this passage.

7. as claimed in claim 5 based on the image de-noising method of multiresolution, it is characterized in that, describedly inverse decomposition is carried out to every single order image information rebuild, comprise with the approximation subband image obtained corresponding to every single order image information:

Carry out inverse decomposition respectively to every single order image information of each passage to rebuild, with the approximation subband image corresponding to the every single order image information obtaining each passage.

8. as claimed in claim 7 based on the image de-noising method of multiresolution, it is characterized in that, described every single order image information to each passage is carried out the inverse process of rebuilding of decomposing respectively and is comprised:

Described inverse decomposition is rebuild and is carried out according to the resolution order from low to high of the multistage image information of this passage, and before higher based on the resolution of image information to this passage after the rear single order denoising that the resolution of this passage is lower, single order image information carries out rebuilding against decomposing.

9. as claimed in claim 7 based on the image de-noising method of multiresolution, it is characterized in that, described bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, comprises with the process of the image information after obtaining the denoising of every single order image information:

Respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, with the image information after the every single order image information denoising obtaining each passage.

10., as claimed in claim 9 based on the image de-noising method of multiresolution, it is characterized in that, also comprise:

After approximation subband image corresponding to the every single order image information obtaining each passage, before respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, Anscombe conversion is carried out to described approximation subband image;

After the approximation subband image corresponding to the every single order image information to each passage carries out bilateral filtering respectively, before obtaining the image information after every single order image information denoising of each passage, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

11. as claimed in claim 9 based on the image de-noising method of multiresolution, and it is characterized in that, described based on the image information result after the image information denoising of initial rank, the image information obtained after original image information denoising comprises:

Based on the image information result after the initial rank image information denoising of each passage, obtain the image information after original image information denoising, the initial rank image information of described each passage refers to the stratum's image information in each passage corresponding to original image information.

12., as claimed in claim 1 based on the image de-noising methods of multiresolution, is characterized in that, described Multiresolution Decomposition method comprise in Wavelet Transform, Gauss's Pyramid transform method, picture contrast Pyramid transform method and gradient pyramid decomposition method any one.

13., as claimed in claim 10 based on the image de-noising methods of multiresolution, is characterized in that, also comprise: convert based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _r.

14., as claimed in claim 13 based on the image de-noising method of multiresolution, is characterized in that, the exponent number carrying out the image information that inverse decomposition is rebuild is larger, the Gray homogeneity parameter σ used in described bilateral filtering _rless.

15., as claimed in claim 13 based on the image de-noising methods of multiresolution, is characterized in that, describedly convert based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _rcomprise:

Based on formula: σ _r=f (σ, layer) determines the Gray homogeneity parameter σ in bilateral filtering _r, wherein, σ converts the image noise variance obtained for Anscombe, and layer carries out the inverse exponent number decomposing the image information of rebuilding, and f (σ, layer) is for asking for the Gray homogeneity parameter σ used in bilateral filtering _rfunction.

16., as claimed in claim 13 based on the image de-noising method of multiresolution, is characterized in that, described Gray homogeneity parameter wherein, σ converts the image noise variance obtained for Anscombe, and layer carries out the inverse exponent number decomposing the image information of rebuilding.

17., as claimed in claim 9 based on the image de-noising methods of multiresolution, is characterized in that, use identical parameter when carrying out bilateral filtering respectively to the approximation subband image corresponding to the same single order image information of each passage.

18. as claimed in claim 17 based on the image de-noising method of multiresolution, and it is characterized in that, described identical parameter comprises described Gray homogeneity parameter σ _r.

19. 1 kinds, based on the image denoising device of multiresolution, is characterized in that, comprising:

Resolving cell, is suitable for carrying out multistage decomposition based on Multiresolution Decomposition method to original image information, obtains every single order image information;

Reconstruction unit, is suitable for carrying out inverse decomposition to every single order image information and rebuilds, to obtain the approximation subband image corresponding to every single order image information;

Filter unit, is suitable for carrying out bilateral filtering, to obtain the image information after the denoising of every single order image information to the approximation subband image corresponding to described every single order image information;

Obtain unit, be suitable for the image information result after based on the image information denoising of initial rank, obtain the image information after original image information denoising, described initial rank image information refers to the stratum's image information corresponding to original image information.

20., as claimed in claim 19 based on the image denoising device of multiresolution, is characterized in that, also comprise:

First converter unit, be suitable for after reconstruction unit obtains the approximation subband image corresponding to every single order image information, filter unit carries out Anscombe conversion to described approximation subband image before carrying out bilateral filtering to the approximation subband image corresponding to described every single order image information.

21. as claimed in claim 20 based on the image denoising device of multiresolution, it is characterized in that, described filter unit comprises: the first inverse transformation subelement, be suitable for after bilateral filtering is carried out to the approximation subband image corresponding to described every single order image information, before obtaining the image information after the denoising of every single order image information, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

22. image denoising devices based on multiresolution as described in any one of claim 19 to 21, it is characterized in that, described resolving cell comprises decomposition subelement, be suitable for carrying out multistage decomposition respectively based on multiple passages of Multiresolution Decomposition method to original image information, to obtain every single order image information of each passage respectively, described multiple passage comprises Y, U, V tri-passages of image information.

23. as claimed in claim 22 based on the image denoising device of multiresolution, it is characterized in that, described reconstruction unit comprises reconstruction subelement, be suitable for carrying out inverse decomposition respectively to every single order image information of each passage to rebuild, obtain the approximation subband image corresponding to every single order image information of each passage.

24. as claimed in claim 23 based on the image denoising device of multiresolution, it is characterized in that, described filter unit comprises filtering subelement, respectively bilateral filtering is carried out to the approximation subband image corresponding to every single order image information of each passage, obtains the image information after every single order image information denoising of each passage.

25. as claimed in claim 24 based on the image denoising device of multiresolution, it is characterized in that, also comprise: the second converter unit, be suitable for after rebuilding the approximation subband image corresponding to every single order image information that subelement obtains each passage, filtering subelement carries out Anscombe conversion to described approximation subband image before carrying out bilateral filtering respectively to the approximation subband image corresponding to every single order image information of each passage.

26. as claimed in claim 25 based on the image denoising device of multiresolution, it is characterized in that, described filtering subelement comprises the second inverse transformation subelement, be suitable for after the approximation subband image corresponding to the every single order image information to each passage carries out bilateral filtering respectively, before obtaining the image information after every single order image information denoising of each passage, Anscombe inverse transformation is carried out to the approximation subband image after bilateral filtering.

27. as claimed in claim 24 based on the image denoising device of multiresolution, it is characterized in that, described acquisition unit comprises acquisition subelement, be suitable for the image information result after based on the initial rank image information denoising of each passage, obtain the image information after original image information denoising, the initial rank image information of described each passage refers to the stratum's image information in each passage corresponding to original image information.

28., as claimed in claim 25 based on the image denoising devices of multiresolution, is characterized in that, also comprise: determining unit, are suitable for converting based on Anscombe the Gray homogeneity parameter σ that the image noise variance obtained determines to use in described bilateral filtering _r.