CN104680485B

CN104680485B - A kind of image de-noising method and device based on multiresolution

Info

Publication number: CN104680485B
Application number: CN201310616537.6A
Authority: CN
Inventors: 牛海军; 陈敏杰; 彭晓峰; 林福辉
Original assignee: Spreadtrum Communications Shanghai Co Ltd
Current assignee: Spreadtrum Communications Shanghai Co Ltd
Priority date: 2013-11-27
Filing date: 2013-11-27
Publication date: 2018-04-27
Anticipated expiration: 2033-11-27
Also published as: CN104680485A

Abstract

A kind of image de-noising method and device based on multiresolution, the described method includes：Multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain per single order image information；Decompose inverse to the progress of every single order image information is rebuild, to obtain the approximation subband image corresponding to every single order image information；Bilateral filtering is carried out to the approximation subband image per corresponding to single order image information, to obtain per the image information after single order image information denoising；Based on the image information after initial rank image information denoising as a result, obtaining the image information after original image information denoising, the initial rank image information refers to stratum's image information corresponding to original image information.This method can effectively improve the denoising effect of image.

Description

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

Technical field

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

Background technology

Image denoising is a kind of using more extensive technology in image preprocessing, and the purpose of image denoising is to improve The signal-to-noise ratio of image, the desired character of prominent image.Image is easily subject to the shadow of various factors during acquisition and transmission Ring so that the image collected by imaging sensor often includes noisy image.

Mixed due to described containing noisy noise in image signal and picture signal so that image existing characteristics The problems such as unobvious, not high clarity, so it is generally necessary to image is collected to imaging sensor carries out denoising to carry The signal-to-noise ratio of hi-vision, improves the display effect of image.

Carry out denoising usually to utilize different qualities on frequency domain of noise signal and picture signal to image and Processing, because picture signal is mainly distributed on low frequency region, and noise signal is mainly distributed on high-frequency region, special based on this Point, the method for carrying out denoising to image in the prior art, such as bilateral filtering method, Wavelet noise-eliminating method etc..It is described double Side filtering method can pay close attention to space length relation and the gray-scale relation between neighborhood pixels at the same time during filtering.It is described Wavelet noise-eliminating method then essentially consists in the selection of noise-removed threshold value, the selection of noise-removed threshold value during denoising is carried out to image It is the key of Wavelet Denoising Method, the selection of the noise-removed threshold value is depended primarily upon to the removal effect of picture noise.

But the effect for carrying out denoising to image using various image de-noising methods in the prior art is not fine, is still deposited In some problems.For example, although the bilateral filtering method can effectively remove the noise signal of high-frequency region, but Due to there is partial noise signal to be distributed across low frequency region, the bilateral filtering method cannot be to making an uproar present in low frequency region Acoustic intelligence is effectively removed, and is additionally, since the detail section such as marginal information of image also respectively in high frequency section, is being reached High frequency section denoising effect while, may result in soft edge, image detail section loss.It is described small Ripple denoising method then exists when the noise-removed threshold value chooses improper, and denoising effect also can be undesirable, such as when denoising threshold When value chooses larger, during image denoising, the detailed information of some images can be removed, and if noise-removed threshold value choose compared with Hour, and the problem of denoising dynamics can be caused inadequate., can when carrying out denoising to image using the image de-noising method of the prior art Can be there are image retention noise, the problems such as soft edge.

Correlation technique refers to the U.S. Patent application of Publication No. US2008166064A1.

The content of the invention

The present invention solves the problems, such as it is that image denoising effect is undesirable in the prior art and can not be to the noise of low frequency region The problem of signal is effectively removed.

To solve the above problems, technical solution of the present invention provides a kind of image de-noising method based on multiresolution, it is described Method includes：

Multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain per single order image information；

Decompose inverse to the progress of every single order image information is rebuild, to obtain the approximation subband figure corresponding to every single order image information Picture；

Bilateral filtering is carried out to the approximation subband image per corresponding to single order image information, to obtain per single order image Image information after information denoising；

Based on the image information after initial rank image information denoising as a result, obtaining the letter of the image after original image information denoising Breath, the initial rank image information refer to stratum's image information corresponding to original image information.

Optionally, it is described to include low frequency region image information and high-frequency region image information per single order image information, it is described Being carried out the process of multistage decomposition to original image information based on Multiresolution Decomposition method method is included：

Multistage decompose carries out according to the resolution ratio order from high to low of image, by the higher preceding single order of resolution ratio The input information of the low frequency region image information rear single order relatively low as resolution ratio, based on the input information, is divided The low frequency region image information and high-frequency region image information of the relatively low rear single order of resolution, wherein, by with highest resolution Original image information is as initial rank image information, using the minimum image information of resolution ratio as most high-order image information.

Optionally, the process that inverse decomposition reconstruction is carried out to every single order image information includes：

It is described it is inverse decompose to rebuild carried out according to the resolution ratio order from low to high of multistage image information, based on resolution ratio compared with Image information after the low rear single order denoising preceding single order image information higher to resolution ratio carries out inverse decompose and rebuilds.

Optionally, the method further includes：

It is right to the institute per single order image information after obtaining per the approximation subband image corresponding to single order image information Before the approximation subband image answered carries out bilateral filtering, Anscombe conversion is carried out to the approximation subband image；

It is each to obtain after bilateral filtering is carried out to the approximation subband image per corresponding to single order image information Before image information after rank image information denoising, Anscombe inverse transformations are carried out to the approximation subband image after bilateral filtering.

Optionally, it is described that multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain every single order Image information includes：

Multistage decomposition is carried out based on Multiresolution Decomposition method respectively to multiple passages of original image information, to obtain respectively Every single order image information of each passage, the multiple passage include tri- passages of Y, U, V of image information.

Optionally, every single order image information of each passage includes low frequency region image information and high-frequency region image Information, the process bag that based on Multiresolution Decomposition method multiple passages of original image information are carried out with multistage decomposition respectively Include：

Multistage decompose carries out according to the resolution ratio order from high to low of image, and for each passage, this is led to The input of the low frequency region image information of the higher preceding single order of the resolution ratio in the road rear single order relatively low as the resolution ratio of the passage Information, based on the input information, decompose the relatively low rear single order of resolution ratio for obtaining the passage low frequency region image information and High-frequency region image information, wherein, using the image information of each passage of the original image information with highest resolution as The initial rank image information of the passage, the most High-order Image using the image information of the minimum each passage of resolution ratio as the passage Information.

Optionally, it is described that inverse decomposition reconstruction is carried out to every single order image information, to obtain per corresponding to single order image information Approximation subband image include：

Inverse decomposition reconstruction is carried out respectively to every single order image information of each passage, to obtain every single order figure of each passage Approximation subband image as corresponding to information.

Optionally, every single order image information to each passage carries out the inverse process rebuild of decomposing respectively includes：

It is described it is inverse decompose to rebuild carried out according to the resolution ratio order from low to high of the multistage image information of the passage, be based on The higher preceding single order image of resolution ratio of the image information to the passage after the relatively low rear single order denoising of the resolution ratio of the passage is believed Breath carries out inverse decompose and rebuilds.

Optionally, it is described that bilateral filtering is carried out to the approximation subband image per corresponding to single order image information, to obtain Taking the process of the image information after every single order image information denoising includes：

Bilateral filtering is carried out respectively to the approximation subband image corresponding to every single order image information of each passage, to obtain Image information after every single order image information denoising of each passage.

Optionally, the method further includes：

After the approximation subband image corresponding to every single order image information of each passage is obtained, to the every of each passage Before approximation subband image corresponding to single order image information carries out bilateral filtering respectively, the approximation subband image is carried out Anscombe is converted；

After the approximation subband image corresponding to 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, to the approximation subband image after bilateral filtering into Row Anscombe inverse transformations.

Optionally, the image information based on after initial rank image information denoising is gone as a result, obtaining original image information Image information after making an uproar includes：

Image information after initial rank image information denoising based on each passage is as a result, obtain original image information denoising Image information afterwards, the initial rank image information of each passage refer to the rank in each passage corresponding to original image information Tomographic image information.

Optionally, the Multiresolution Decomposition method includes Wavelet Transform, Gauss Pyramid transform method, picture contrast turriform Any one in decomposition method and gradient pyramid decomposition method.

Optionally, the method further includes：Obtained image noise variance is converted based on Anscombe and determines described pair The Gray homogeneity parameter σ used in the filtering of side_r。

Optionally, it is bigger to carry out the exponent number of the inverse image information decomposed and rebuild, the gray scale used in the bilateral filtering away from From parameter σ_rIt is smaller.

Optionally, it is described to determine to use in the bilateral filtering based on the obtained image noise variance of Anscombe conversion Gray homogeneity parameter σ_rIncluding：

Based on formula：σ_r=f (σ, layer) determines the Gray homogeneity parameter σ in bilateral filtering_r, wherein, σ Anscombe Obtained image noise variance is converted, for layer to carry out the inverse exponent number for decomposing the image information rebuild, f (σ, layer) is to ask Take the Gray homogeneity parameter σ used in bilateral filtering_rFunction.

Optionally, the Gray homogeneity parameterWherein, σ converts obtained image for Anscombe and makes an uproar Sound variance, layer are the exponent number for carrying out the inverse image information decomposed and rebuild.

Optionally, when carrying out bilateral filtering respectively to the approximation subband image corresponding to the same single order image information of each passage Use identical parameter.

Optionally, the identical parameter includes the 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 includes：

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

Reconstruction unit, rebuilds suitable for carrying out inverse decompose to every single order image information, right to obtain the institute per single order image information The approximation subband image answered；

Filter unit, suitable for carrying out bilateral filtering to the approximation subband image per corresponding to single order image information, with Obtain per the image information after single order image information denoising；

Obtaining unit, suitable for based on the image information after initial rank image information denoising as a result, obtaining original image information Image information after denoising, the initial rank image information refer to stratum's image information corresponding to original image information.

Optionally, described device further includes：

First converter unit, suitable for obtaining per the approximation subband image corresponding to single order image information it in reconstruction unit Afterwards, before filter unit carries out bilateral filtering to the approximation subband image per corresponding to single order image information, to described near Anscombe conversion is carried out like sub-band images.

Optionally, the filter unit includes：First inverse transformation subelement, suitable for every single order image information institute It is right before obtaining the image information after every single order image information denoising after corresponding approximation subband image carries out bilateral filtering Approximation subband image after bilateral filtering carries out Anscombe inverse transformations.

Optionally, the resolving cell includes decomposing subelement, suitable for being believed based on Multiresolution Decomposition method original image Multiple passages of breath carry out multistage decomposition respectively, to obtain every single order image information of each passage, the multiple passage respectively Tri- passages of Y, U, V including image information.

Optionally, the reconstruction unit includes rebuilding subelement, suitable for distinguishing every single order image information of each passage Carry out inverse decompose to rebuild, obtain the approximation subband image corresponding to every single order image information of each passage.

Optionally, the filter unit includes filtering subunit, to corresponding to every single order image information of each passage Approximation subband image carries out bilateral filtering respectively, obtains the image information after every single order image information denoising of each passage.

Optionally, described device further includes：Second converter unit, suitable for rebuilding each of each passage of subelement acquisition After approximation subband image corresponding to rank image information, filtering subunit is to corresponding to every single order image information of each passage Approximation subband image carry out bilateral filtering respectively before, to the approximation subband image carry out Anscombe conversion.

Optionally, the filtering subunit includes the second inverse transformation subelement, suitable in every single order figure to each passage After approximation subband image as corresponding to information carries out bilateral filtering respectively, the every single order image information for obtaining each passage is gone Before image information after making an uproar, Anscombe inverse transformations are carried out to the approximation subband image after bilateral filtering.

Optionally, the obtaining unit includes obtaining subelement, is gone suitable for the initial rank image information based on each passage Image information after making an uproar is as a result, obtain the image information after original image information denoising, the initial rank image of each passage Information refers to stratum's image information corresponding to original image information in each passage.

Optionally, described device further includes determination unit, suitable for converting obtained picture noise side based on Anscombe Difference determines the Gray homogeneity parameter σ used in the bilateral filtering_r。

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

Multistage decomposition is carried out to original image information based on Multiresolution Decomposition method and obtains multistage image information, to by described The approximation subband image obtained after every single order image reconstruction of multistage image carries out bilateral filtering, obtains per after single order image denoising Image information, may finally obtain correspond to original image information denoising after image information.This method is in image denoising During, noise signal of the image in different frequency domains can effectively be obtained by Multiresolution Decomposition method, and then combine bilateral filter The method of ripple can effectively remove the high frequency of image, the noise signal of low frequency region, effectively improve the denoising effect of image Fruit.

Multistage decomposition is carried out based on Multiresolution Decomposition method respectively to multiple passages of original image information, it is each to obtain Every single order image information of passage, carries out inverse decomposition reconstruction to every single order image information of each passage, obtains each logical respectively Approximation subband image corresponding to every single order image information in road, to the approximation corresponding to every single order image information of each passage Sub-band images carry out bilateral filtering respectively, the image information after every single order image information denoising of each passage are obtained, based on every Image information after the original image information denoising of a passage is as a result, may finally obtain after corresponding to original image information denoising Image information.Multiresolution Decomposition method and bilateral filtering method can be combined, the noise of low frequency region can believed Breath is effectively removed, and effectively improves the denoising effect of image.

To every single order image of the multistage image or every single order image of each passage, carry out against after decomposing reconstruction Before obtained approximation subband image carries out bilateral filtering, converted by variance stability（Such as converted by Anscombe）Can be with Poisson noise present in image information is converted to the noise of Gaussian Profile, so as to be carried out under Gauss model at denoising Reason, obtains effective denoising effect.

Obtained image is made an uproar in being converted based on the exponent number for carrying out the inverse image information decomposed and rebuild and by variance stability Sound variance determines the Gray homogeneity parameter used in bilateral filtering, can carry out bilateral filtering in the image information to not same order During, using the Gray homogeneity parameter being adapted with the rank image information, adaptive denoising is realized, improves image Denoising effect.

For each passage same single order image information carry out bilateral filtering during, will be based on determined by luminance channel Gray homogeneity parameter and the window weight being thus calculated are used for the denoising of chrominance channel, it is possible to prevente effectively from false colors Edge or texture, avoid the fuzzy of marginalisation, obtain effective global de-noising effect.

Brief description of the drawings

Fig. 1 is the flow diagram for 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 inverse schematic diagram for decomposing reconstruction；

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

Fig. 4 is the flow diagram of the image de-noising method provided by Embodiment 2 of the present invention based on multiresolution；

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

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

Embodiment

Various image de-noising methods can be used to carry out denoising to image in the prior art, for example, it is bilateral filtering method, small Ripple denoising method etc..

The bilateral filtering method can be paid close attention at the same time during filtering space length relation between neighborhood pixels with And gray-scale relation, image border can be preferably preserved using double wave filtering, and then effectively noise is carried out smoothly, to reach The purpose made an uproar, can use formula（1）Carry out double wave filtering.

Wherein, C is normalization factor, the normalization factor C form of Definition such as formula（2）It is shown.

Wherein, in formula（1）And formula（2）In, σ_dFor space length parameter, σ_rFor Gray homogeneity parameter, N (x) is with picture A spatial neighborhood centered on vegetarian refreshments x, y are a pixel of the spatial neighborhood, and I (x), I (y) are pixel x, y's Gray value,For the gray value of pixel x after bilateral filtering.

The effect of bilateral filtering is by the space length parameter σ_dWith Gray homogeneity parameter σ_rDetermine, σ_dAnd σ_rOnce wherein Rough phenomenon just occurs close to 0 image in one amount, as long as σ_rExcursion within the scope of certain, just to image Edge do not influence, σ_rCompare σ_dThe details of image is easily influenced, by formula（1）Understand, σ in bilateral filtering_rAnd σ_dIt is to be multiplied Form, as long as this means that σ_rAnd σ_dAs soon as close to 0 in, rough phenomenon occurs in image.

In the Wavelet noise-eliminating method to image carry out wavelet transformation, to containing noisy image carry out wavelet transformation when, The wavelet coefficient of image can be obtained, the wavelet coefficient can be divided into two classes, first kind wavelet coefficient is mainly believed by image Number detail section caused by, also include the mapped structure of picture noise signal, the amplitude of wavelet coefficient is larger, number compared with Few, the second class wavelet coefficient is mainly as caused by picture noise signal, and the amplitude of wavelet coefficient is smaller, and number is more.

Wavelet noise-eliminating method based on threshold value is the noise suppressing method based on nonparametric model, and the small echo based on threshold value is gone Method for de-noising can pass through formula（3）Realize.

Wherein σ_wFor the noise variance of image, w_i,jExist for image（I, j）Pixel at position is after wavelet transformation Value, median represent to carry out median calculation.

Based on formula（3）The noise variance σ of identified image_w, pass through formula（4）It can obtain in Wavelet noise-eliminating method Used noise-removed threshold value T_h。

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

The Wavelet noise-eliminating method then essentially consists in the selection of noise-removed threshold value, goes during denoising is carried out to image The selection for threshold value of making an uproar is the key of Wavelet Denoising Method, and the choosing of the noise-removed threshold value is depended primarily upon to the removal effect of picture noise Take.

But the effect of image denoising all there are problems that caused by various image de-noising methods of the prior art, example Such as the noise signal in low frequency region can not effectively be removed, or during image denoising, will can be schemed at the same time The problem of detailed information as in removes, may there are image retention noise, image after the prior art carries out denoising to image How the problems such as edge blurry, therefore, the details of image is retained while picture noise is reduced, keeps the side of image clearly How edge, i.e., obtain more preferable denoising effect just into current Image Denoising Technology problem to be solved.

To solve the above-mentioned problems, technical solution of the present invention provides a kind of image de-noising method based on multiresolution.

Fig. 1 be technical solution of the present invention provide the image de-noising method based on multiresolution flow diagram, such as Fig. 1 It is shown, step S101 is first carried out, multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, it is each to obtain Rank image information.

Since picture signal is mainly distributed on low frequency region, and noise signal is mainly distributed on high-frequency region, and based on more Resolution decomposition method can effectively find noise, so in step S101, be primarily based on Multiresolution Decomposition method to original graph As the multistage decomposition of information progress, to obtain the image information per single order.The Multiresolution Decomposition method can believe original image Breath is decomposed into the image information of multiple and different resolution ratio, such as original image information has highest resolution ratio, and by original graph The low frequency sub-band of the first rank image information obtained after being decomposed as information and the resolution ratio of high-frequency sub-band can be respectively original The a quarter of the resolution ratio of beginning image information, and so on, the low frequency sub-band of image information that is obtained after decomposing each time and High-frequency sub-band all only decomposed before image information a quarter resolution ratio, the multistage decomposition of image information can be by Carried out according to the resolution ratio order from high to low of image.

, can be by the low frequency region figure for the image information that currently decomposed during Multiresolution Decomposition is carried out Picture information carries out Multiresolution Decomposition as input image information, and then based on the input image information, obtains lower single order Low frequency region image information and high-frequency region image information, using the low frequency region image information of lower single order as single order figure behind As the input picture of information, Multiresolution Decomposition is carried out based on this input picture again, and so on, can be from best result The original image information of resolution starts, and is carried out based on Multiresolution Decomposition method after repeatedly decomposing, obtains multistage image information, can be with Using original image information as initial rank image information, the minimum image information of obtained resolution ratio is decomposed for the last time as most High-order Image information.

The Multiresolution Decomposition method can be Wavelet Transform, Gauss Pyramid transform method, picture contrast Pyramid transform Method or gradient pyramid decomposition method etc..

Step S102 is performed, inverse decompose is carried out to every single order image information and is rebuild, it is right to obtain the institute per single order image information The approximation subband image answered.

After multistage image information is obtained, it is possible to denoising is carried out to every single order image information, in denoising, it is necessary first to Inverse decompose of multistage image information progress of acquisition is rebuild, to obtain the approximation subband image corresponding to every single order image information, To carry out denoising to the approximation subband image.

It is described it is inverse decompose rebuild can according to resulting multistage image information in step S101 resolution ratio from low to high Order carries out, i.e., inverse decompose is carried out to most high-order image information first rebuilds, and obtains the approximation most corresponding to high-order image information Sub-band images, can carry out denoising, by denoising to the approximation subband image corresponding to most high-order image information afterwards Image information afterwards carries out the inverse input picture for decomposing reconstruction operation as preceding single order image, based on the input picture and institute The high-frequency region information of single order image information carries out inverse decomposition reconstruction before stating, and obtains near corresponding to the preceding single order image information Like sub-band images, and so on, can be relatively low based on resolution ratio since the most high-order image information with lowest resolution Image information afterwards after the single order denoising preceding single order image information higher to resolution ratio carries out inverse decompose and rebuilds, and obtains every single order figure Approximation subband image as corresponding to information.

It is corresponding with the Multiresolution Decomposition method employed in step S101 that step S102 carries out the inverse method for decomposing reconstruction. For example, if step S101 uses Gauss Pyramid transform method, can use and the Gauss Pyramid transform in step s 102 Inverse decomposition method for reconstructing corresponding to method.

In order to make it easy to understand, the inverse decomposition weight that the multistage decomposable process and step S102 described in step S101 are carried out The process of building may be referred to Fig. 2, as shown in Fig. 2, 0 rank image information represents original image information, the 0 rank image information can also Referred to as initial rank, the original image information have highest resolution ratio.N ranks are most high-order image information, the most high-order figure As information has minimum resolution ratio, original image information obtained after multistage decomposition 1 rank, 2 ranks ..., n-2 ranks, n-1 Rank, n rank image informations.The multistage decomposable process is believed according to direction shown in Fig. 2 left arrows from high-resolution image The image information for ceasing low resolution carries out, and inverse decompose is rebuild according to the image information with low resolution to high-resolution Image information carry out.

Please continue to refer to Fig. 1, perform step S103, to the approximation subband image per corresponding to single order image information into Row bilateral filtering, to obtain per the image information after single order image information denoising.

In step s 102, approximation subband image corresponding thereto can be obtained for every single order image information, Bilateral filtering is carried out to the approximation subband image in step S103, is believed with obtaining per the image after single order image information denoising Breath.

Denoising, the figure after current rank denoising carry out it using the method for bilateral filtering for every single order image information As information can carry out the inverse input image information for decomposing reconstruction as its preceding single order image information.

Step S104 is performed, based on the image information after initial rank image information denoising as a result, obtaining original image information Image information after denoising, the initial rank image information refer to stratum's image information corresponding to original image information.

In above-mentioned steps, after the multistage image that image information is obtained based on Multiresolution Decomposition method, according to multistage image The order of the resolution ratio of information from low to high carries out, based on the image information after the relatively low rear single order denoising of resolution ratio to resolution ratio Higher preceding single order image information carries out the process that inverse decomposition is rebuild, and denoising, the inverse mistake for decomposing reconstruction are carried out by successive ignition Journey, can decompose the process rebuild by the way that last time is inverse, obtain the image letter after the inverse decomposition corresponding to original image is rebuild Breath, then after carrying out bilateral filtering to it, it is possible to obtain original image information（It is referred to as initial rank information）After denoising Image information, that is, obtain final image denoising result.

Multiresolution Decomposition method and bilateral filtering method are combined, pass through during image denoising by this method Multiresolution Decomposition method can effectively obtain noise information of the image in different frequency domains, and then the method for combining bilateral filtering can be with The noise information of high frequency, low frequency region to image is effectively removed, and effectively improves the denoising effect of image.

It is understandable to enable the above objects, features and advantages of the present invention to become apparent, below in conjunction with the accompanying drawings to the present invention Specific embodiment be described in detail.

Embodiment one

In the present embodiment, during carrying out denoising to original image information based on Multiresolution Decomposition method, to obtaining Multistage image rebuild by inverse decomposition after before obtained approximation subband image carries out bilateral filtering, converted by variance stability （Such as Anscombe conversion）Poisson noise present in image information is converted to the noise of Gaussian Profile.

Fig. 3 is the flow diagram of the image de-noising method provided in this embodiment based on multiresolution, as shown in figure 3, Step S301 is first carried out, multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain per single order figure As information.Step S301 refer to step S101.

Step S302 is performed, inverse decompose is carried out to every single order image information and is rebuild, it is right to obtain the institute per single order image information The approximation subband image answered.Step S302 refer to step S102.

Step S303 is performed, Anscombe changes are carried out to the approximation subband image per corresponding to single order image information Change.

Anscombe conversion is carried out to the approximation subband image obtained in step S302, can be by present in image information Poisson noise is converted to the noise of Gaussian Profile, and then can obtain the variance yields of noise, can by Anscombe conversion Carried out with will be transformed into the denoising of image information under Gauss model.

The Anscombe conversion can be based on formula（5）Realize.

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

Based on formula（5）Anscombe conversion, can obtain that f (z) is approximate to obey the Gaussian Profile that noise variance is 1.

Step S304 is performed, to the approximation subband image corresponding to every single order image information after progress Anscombe conversion Carry out bilateral filtering.

The approximation subband image corresponding to every single order image information after being converted for Anscombe is using bilateral filtering Method carries out denoising to it.

Step S305 is performed, the approximation subband image corresponding to every single order image information after bilateral filtering is carried out Anscombe inverse transformations.

Anscombe inverse transformations are carried out to the approximation subband image after the bilateral filtering that is obtained in step S304, inverse transformation Method can use method known to a person skilled in the art to carry out inverse transformation, and details are not described herein.

Step S306 is performed, is obtained per the image information after single order image information denoising.

Current rank can be obtained based on the image information after obtained progress Anscombe inverse transformations in step S305 Image information after image information denoising.

Step S307 is performed, based on the image information after initial rank image information denoising as a result, obtaining original image information Image information after denoising.Step S307 refer to step S104.

In the present embodiment, obtained after inverse decomposition is carried out to every single order image information of the multistage image and is rebuild near Before carrying out bilateral filtering like sub-band images, poisson noise present in image information can be converted to by Anscombe conversion The noise of Gaussian Profile, so as to carry out denoising under Gauss model, obtains effective denoising effect.

Embodiment two

In the present embodiment, during denoising is carried out to original image information based on Multiresolution Decomposition method, it is based on Carry out the exponent number of the inverse image information decomposed and rebuild and converted by variance stability（Such as Anscombe conversion）In obtained by Image noise variance determine the Gray homogeneity parameter used in bilateral filtering.

Fig. 4 is the flow diagram of the image de-noising method provided in this embodiment based on multiresolution, as shown in figure 4, Step S401 is first carried out, multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain per single order figure As information.

Step S402 is performed, inverse decompose is carried out to every single order image information and is rebuild, it is right to obtain the institute per single order image information The approximation subband image answered.

Step S403 is performed, Anscombe changes are carried out to the approximation subband image per corresponding to single order image information Change.

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

Perform step S404, based on Anscombe convert obtained image noise variance determine pairing approximation sub-band images into The Gray homogeneity parameter used during row bilateral filtering.

When step S403 is carrying out Anscombe conversion to the approximation subband image, image noise variance can be obtained σ, due to consideration that the relatively low approximation subband image of resolution ratio can use less Gray homogeneity to join when carrying out bilateral filtering Number σ_r, so can be to avoid excessive the problem of obscuring after image information denoising, so carrying out the Gray homogeneity parameter σ of bilateral filtering_r Can be associated with the current exponent number for carrying out the inverse image information for decomposing reconstruction of multiresolution.Since the resolution ratio of image information is got over Small corresponding exponent number is bigger, so, when the exponent number for carrying out the inverse image information decomposed and rebuild is bigger, the bilateral filtering at this time The middle Gray homogeneity parameter σ used_rCan be smaller.

It can be based on formula（6）Determine the Gray homogeneity parameter σ_r。

σ_r=f(σ,layer) （6）

Wherein, σ converts obtained image noise variance for Anscombe, and layer is to carry out the inverse image for decomposing and rebuilding The exponent number of information, f (σ, layer) are the Gray homogeneity parameter σ for asking for using in bilateral filtering_rFunction.

The f (σ, layer) can be image noise variance σ and function inversely proportional layer, for example, the gray scale away from From parameter σ_rFormula can be passed through（7）It is determined.

Step S405 is performed, to the approximation subband image corresponding to every single order image information after progress Anscombe conversion Carry out bilateral filtering.

Based on identified Gray homogeneity parameter σ in step S404_rTo the approximation subband corresponding to every single order image information Image carries out bilateral filtering.

Step S406 is performed, the approximation subband image corresponding to every single order image information after bilateral filtering is carried out Anscombe inverse transformations.

Step S407 is performed, is obtained per the image information after single order image information denoising.

Step S408 is performed, based on the image information after initial rank image information denoising as a result, obtaining original image information Image information after denoising.

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

In the present embodiment, when carrying out bilateral filtering to image information, without artificial setting filtering parameter, Ke Yigen According to the current exponent number for carrying out the inverse image information decomposed and rebuild, adaptive definite Gray homogeneity parameter can be to not same order During image information carries out bilateral filtering, using the Gray homogeneity parameter being adapted with the rank image information, realize adaptive Answer the effect of image denoising.

Embodiment three

Color image information can be divided into the image information of multiple passages, in the present embodiment, with the cromogram of multichannel As being illustrated exemplified by information.Fig. 5 is the flow diagram of the image de-noising method provided in this embodiment based on multiresolution, As shown in figure 5, step S501 is first carried out, multiple passages of original image information are carried out respectively based on Multiresolution Decomposition method Multistage decomposition, to obtain every single order image information of each passage respectively.

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

Step S501 may be referred to step S101, in the present embodiment, it is necessary to be performed both by being similar to step to multiple passages The operation of S101, it is by taking Y, U, V passage as an example, it is necessary to more based on the progress of Multiresolution Decomposition method to the original image information of Y passages Rank is decomposed, to obtain every single order image information of Y passages, it is also necessary to the progress of Multiresolution Decomposition method is based respectively on to U, V passage Multistage decomposition, to obtain every single order image information of U passages and V passages respectively.

Step S502 is performed, inverse decomposition reconstruction is carried out respectively to every single order image information of each passage, it is each to obtain Approximation subband image corresponding to every single order image information of passage.

Step S502 may be referred to step S102, and step S502 and the difference of step S102 are, in the present embodiment In rebuild, it is necessary to respectively carry out inverse decompose to every single order image information of each passage, for every single order figure of each passage As information can obtain corresponding approximation subband image.

The inverse decomposition reconstruction process described in multistage decomposable process and step S502 described in for step S501 can still join Fig. 2 is examined, is required for performing the multistage decomposition of similar Fig. 2, the inverse process for decomposing reconstruction for each passage.

Step S503 is performed, the approximation subband image corresponding to every single order image information of each passage is carried out respectively double Side filters, to obtain the image information after every single order image information denoising of each passage.

Step S504 is performed, the image information after the initial rank image information denoising based on each passage is as a result, obtain original Image information after beginning image information denoising.

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

In the method, for the image information with multichannel, by Multiresolution Decomposition method and bilateral filtering method into Row combines, and the noise information of low frequency region can effectively be removed, effectively improve the denoising effect of image.

Example IV

In the present embodiment, with based on the image de-noising method of multiresolution to tri- passages of Y, U, V image believe Breath illustrated exemplified by denoising.

During denoising, the approximation subband that is obtained after being rebuild to the obtained multistage image of each passage by inverse decomposition Before image carries out bilateral filtering, poisson noise present in image information of changing commanders is become by Anscombe and is converted to Gaussian Profile Noise, and obtained picture noise in being converted based on the exponent number for carrying out the inverse image information decomposed and rebuild and by Anscombe Variance determines the Gray homogeneity parameter used in bilateral filtering.

Fig. 6 is the flow diagram of the image de-noising method provided in this embodiment based on multiresolution, as shown in fig. 6, Step S601 is first carried out, multistage decomposition is carried out based on Multiresolution Decomposition method respectively to multiple passages of original image information, To obtain every single order image information of each passage respectively.

When described image color space is YUV color spaces, then Multiresolution Decomposition method can be based on to original image Tri- passages of Y, U, V carry out multistage decomposition respectively, in the present embodiment, the image of each passage is decomposed into 4 rank images Information.

By taking Y passages as an example, the image information identifier of the Y passages of original image information can be schemed for 0 rank of the Y passages The 0 rank image information, the image of the high-frequency region of the 1st rank image information is decomposed into based on Multiresolution Decomposition method by picture information The image information of information and low frequency region, the low frequency sub-band of the 1st rank image information and the resolution ratio of high-frequency sub-band can divide Not Wei 0 rank image information a quarter；Is further used as with the image information of the low frequency region of the 1st rank image information The input image information of 2 rank image informations, based on the input image information, using Multiresolution Decomposition method by the 1st rank Image information is decomposed into the image information of high-frequency region and the image information of low frequency region of the 2nd rank image information, the 2nd rank The low frequency sub-band of image information and the resolution ratio of high-frequency sub-band can be respectively a quarter of the 1st rank image information；With such Push away, the 3rd rank image information can be obtained, the low frequency sub-band of the 3rd rank image information and the resolution ratio of high-frequency sub-band can divide Not Wei 2 rank image informations a quarter.Using the above method, 0 rank, 1 rank, 2 ranks and 3 rank image informations are can obtain altogether.Similarly, All 4 rank image informations of U passages and all 4 rank image informations of V passages can be obtained in corresponding manner.

Step S602 is performed, the inverse higher-order image information rebuild of decomposing is not carried out to each passage and carries out inverse decomposition respectively Rebuild, to obtain the approximation subband image corresponding to the rank image information of each passage.

After the multistage image information of tri- passages of Y, U, V is obtained based on step S601, rebuild to image information When, can be carried out by inverse decompose and rebuild for the most high-order of each passage, that is, the 3rd rank image information by step S602 first, The approximation subband image corresponding to the 3rd rank image information of Y passages can be obtained, can similarly obtain U passage V passages respectively The corresponding approximation subband image of 3rd rank image information institute.If in step S601 it is current do not carry out it is inverse decompose rebuild compared with High-order Image information is the 2nd rank image information（It can be appreciated that it is the highest not carried out in the inverse image information decomposed and rebuild Rank image information）, then what is completed in this step is to carry out inverse decomposition respectively to the 2nd rank image information of tri- passages of Y, U, V Rebuild, the approximation subband image corresponding to the 2nd rank image information of tri- passages of Y, U, V can be respectively obtained, and so on, obtain To the approximation subband image corresponding to the rank image information of each passage.

It should be noted that in this step, the inverse higher-order image information decomposed and rebuild is not being carried out to each passage It is based on the denoising adjacent and lower than the rank image resolution ratio with the rank resolution ratio during carrying out inverse decomposition reconstruction respectively Rear image information carries out inverse decomposing what is rebuild, for example, if be currently to the 2nd rank image information of Y passages respectively into Row is then that the inverse decomposition of image information progress after the 3rd rank image information denoising based on Y passages is rebuild against reconstruction is decomposed.

In by tri- passages of Y, U, V each passage do not carry out it is inverse decompose the higher-order image information rebuild respectively into Inverse decompose of row is rebuild, and after obtaining the approximation subband image corresponding to the rank image information of each passage, performs step S603.

Step S603, Anscombe is carried out to the approximation subband image corresponding to the rank image information of each passage Conversion.

Step S603 refer to one step S303 of embodiment, is with step S303 differences, is pair in this step Approximation subband image corresponding to the rank image information of each passage carries out Anscombe conversion, needs herein to Y, U, V tri- Approximation subband image corresponding to the rank image information of a passage carries out Anscombe conversion, such as the to Y passages the 3rd respectively Approximation subband image corresponding to rank image information carries out Anscombe conversion, to corresponding to the 3rd rank image information of U, V passage Approximation subband image carry out Anscombe conversion respectively.

Step S604 is performed, converting obtained image noise variance based on Anscombe determines to the approximation subband figure As the Gray homogeneity parameter used when carrying out bilateral filtering.

It may be referred to the ash that two step S404 of embodiment determines to use when carrying out bilateral filtering to the approximation subband image Spend distance parameter σ_r, due to there is tri- passages of Y, U, V, so accordingly corresponding to the rank image information to each passage When approximation subband image carries out Anscombe conversion, it can obtain and corresponding three figures of described tri- passage difference of Y, U and V As noise variance σ_Y、σ_UAnd σ_V, and then can determine after being converted to Anscombe, the rank image information of tri- passages of Y, U and V The Gray homogeneity parameter σ that corresponding approximation subband image uses when being filtered_rValue can beWith

Step S605 is performed, to the approximation corresponding to the rank image information of each passage after progress Anscombe conversion Sub-band images carry out bilateral filtering.

Based on the Gray homogeneity parameter σ of identified corresponding passage in step S604_rValue, schemes the rank of current channel Approximation subband image as corresponding to information carries out bilateral filtering.

For example, if what is obtained in step S604 is that Y passages carry out the 3rd rank image information after Anscombe conversion Corresponding approximation subband image, then be that Y passages are carried out corresponding to the 3rd rank image information after Anscombe conversion herein Approximation subband image is filtered, and can obtain the image information after the 3rd rank image information denoising of Y passages.For U, V passage The image information of every single order bilateral filtering is carried out using same method.

Step S606 is performed, to the approximation subband image corresponding to the rank image information of each passage after bilateral filtering Carry out Anscombe inverse transformations.

Step S607 is performed, obtains the image information after the rank image information denoising of each passage.

The rank image can be obtained based on the image information after obtained progress Anscombe inverse transformations in step S606 Image information after information denoising.

Perform step S608, judge each passage the rank image information resolution ratio whether be the passage initial rank figure As the resolution ratio of information.

In step S608 it needs to be determined that the rank image information resolution ratio of each passage whether be the passage initial rank The resolution ratio of image information.

After the image information after the denoising of the rank image information of each passage is obtained based on step S607, step is performed Rapid S608, judge each passage the rank image information resolution ratio whether be the passage corresponding point of initial rank image information Resolution, that is, judge image information corresponding to the rank of each passage resolution ratio whether by each passage original image it is right The resolution ratio for the 0 rank image information answered.If it is, having obtained the image information after original image denoising, step is performed S609；Otherwise return and perform step S602.

For example, totally 4 rank image information is obtained if decomposed in step S601, most high-order is the 3rd rank, then if step Obtained in S607 be tri- passages of Y, U, V the 3rd rank image information denoising after image information, then should return to step S602, continue to tri- passages of Y, U, V do not carry out it is inverse decompose the higher-order image information rebuild, i.e. the 2nd rank image information into Inverse decompose of row is rebuild.If step S607 obtains being the image information after the 0th rank image information denoising of tri- passages of Y, U, V, Then obtain the image information after original image denoising, it should perform step S609.

Step S609, terminates the denoising to image information.

It will be understood by those skilled in the art that the image de-noising method described above based on multiresolution, it is possibility to have Other conversion embodiments, for example, being to multiple in step S602 into step S608, in each step in the present embodiment The image information of the phase same order of passage is operated accordingly respectively, and multiple passage phase same orders can be obtained in each step The corresponding operating result of image information institute.In other embodiments, class only first can also be performed to one of passage The step of being similar to step S602 to step S608, i.e., first perform above-mentioned steps to single passage, obtain the original graph of single passage As the image information after information denoising, then above-mentioned steps are performed respectively to other passages successively, obtain the original of each passage respectively Image information after beginning image information denoising, and then can equally obtain the final denoising result of original image information.

And for example, in the present embodiment, while Anscombe conversion is employed, and based on resulting in Anscombe conversion The image noise variance method that determines the Gray homogeneity parameter used in bilateral filtering, in other embodiments, can also Anscombe transform methods are only included, are not limited herein.

In addition, in the present embodiment, approximation corresponding to the rank image information of each passage is determined in step S604 The Gray homogeneity parameter σ used when carrying out bilateral filtering with image_rWhen, different gray scales is determined respectively according to different passages Distance parameterIn other embodiments, can also only determine gray scale corresponding to Y passages away from From parameter, and U, V passage use the Gray homogeneity parameter determined by Y passages in filtering.This is because Y passages, Containing information such as edge, textures i.e. in the image information of luminance channel, it is based on so utilizing containing information such as the edge, textures Luminance channel determined byWhen carrying out bilateral filtering to U and V chrominance channels, can effectively it prevent because some chrominance passband Edge or texture information unobvious in road information and cause the denoising dynamics of denoising dynamics and other chrominance channels to differ, most False colors edge is caused either texture or to cause edge blurry eventually.Therefore, for the same single order image information of each passage During carrying out bilateral filtering, by the Gray homogeneity parameter based on determined by luminance channel and the window being thus calculated Weight is also used for the denoising of chrominance channel, it is possible to prevente effectively from false colors edge or texture, avoid the fuzzy of marginalisation, obtain Obtain effective global de-noising effect.

Although present disclosure is as above, the present invention is not limited to this.Any those skilled in the art, are not departing from this In the spirit and scope of invention, it can make various changes or modifications, therefore protection scope of the present invention should be with claim institute Subject to the scope of restriction.

Claims

A kind of 1. image de-noising method based on multiresolution, it is characterised in that including：

Multistage decomposition is carried out to original image information based on Multiresolution Decomposition method, to obtain per single order image information；

Decompose inverse to the progress of every single order image information is rebuild, to obtain the approximation subband image corresponding to every single order image information；

Bilateral filtering is carried out to the approximation subband image per corresponding to single order image information, to obtain per single order image information Image information after denoising；

Based on the image information after initial rank image information denoising as a result, obtaining the image information after original image information denoising, The initial rank image information refers to stratum's image information corresponding to original image information；

Wherein, further include：After obtaining per the approximation subband image corresponding to single order image information, to described per single order image Before approximation subband image corresponding to information carries out bilateral filtering, Anscombe conversion is carried out to the approximation subband image；

After bilateral filtering is carried out to the approximation subband image per corresponding to single order image information, to obtain per single order figure Before the image information after information denoising, Anscombe inverse transformations are carried out to the approximation subband image after bilateral filtering；

Obtained image noise variance is converted based on the exponent number and Anscombe for carrying out the inverse image information decomposed and rebuild to determine The Gray homogeneity parameter σ used in the bilateral filtering_r。
2. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that described per single order image letter Breath includes low frequency region image information and high-frequency region image information, and the Multiresolution Decomposition method method that is based on is to original image The process that information carries out multistage decomposition includes：

Multistage decompose carries out according to the resolution ratio order from high to low of image, by the low frequency of the higher preceding single order of resolution ratio The input information of the regional image information rear single order relatively low as resolution ratio, based on the input information, decomposition obtains resolution ratio The low frequency region image information and high-frequency region image information of relatively low rear single order, wherein, by with the original of highest resolution Image information is as initial rank image information, using the minimum image information of resolution ratio as most high-order image information.
3. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that described to every single order image Information, which carries out the inverse process rebuild of decomposing, to be included：

It is described it is inverse decompose to rebuild carried out according to the resolution ratio order from low to high of multistage image information, it is relatively low based on resolution ratio Image information afterwards after the single order denoising preceding single order image information higher to resolution ratio carries out inverse decompose and rebuilds.
4. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that described to be based on multiresolution Decomposition method carries out multistage decomposition to original image information, is included with obtaining per single order image information：

Multistage decomposition is carried out based on Multiresolution Decomposition method respectively to multiple passages of original image information, it is each to obtain respectively Every single order image information of passage, the multiple passage include tri- passages of Y, U, V of image information.
5. the image de-noising method based on multiresolution as claimed in claim 4, it is characterised in that each passage it is every Single order image information includes low frequency region image information and high-frequency region image information, and the Multiresolution Decomposition method that is based on is to original The process that multiple passages of beginning image information carry out multistage decomposition respectively includes：

Multistage decompose carries out according to the resolution ratio order from high to low of image, for each passage, by the passage The input information of the low frequency region image information of the higher preceding single order of the resolution ratio rear single order relatively low as the resolution ratio of the passage, Based on the input information, the low frequency region image information and high frequency region of the relatively low rear single order of resolution ratio for obtaining the passage are decomposed Area image information, wherein, using the image information of each passage of the original image information with highest resolution as the passage Initial rank image information, the most high-order image information using the image information of the minimum each passage of resolution ratio as the passage.
6. the image de-noising method based on multiresolution as claimed in claim 4, it is characterised in that described to every single order image Information carries out inverse decompose and rebuilds, and is included with obtaining per the approximation subband image corresponding to single order image information：

Inverse decomposition reconstruction is carried out respectively to every single order image information of each passage, to obtain every single order image of each passage letter The corresponding approximation subband image of breath.
7. the image de-noising method based on multiresolution as claimed in claim 6, it is characterised in that described to each passage Carrying out the inverse process rebuild of decomposing respectively per single order image information includes：

It is described it is inverse decompose to rebuild carried out according to the resolution ratio order from low to high of the multistage image information of the passage, it is logical based on this The higher preceding single order image information of resolution ratio of the image information to the passage after the relatively low rear single order denoising of the resolution ratio in road into Inverse decompose of row is rebuild.
8. the image de-noising method based on multiresolution as claimed in claim 6, it is characterised in that described to every single order Approximation subband image corresponding to image information carries out bilateral filtering, to obtain per the image information after single order image information denoising Process include：

Bilateral filtering is carried out respectively to the approximation subband image corresponding to every single order image information of each passage, it is each to obtain Image information after every single order image information denoising of passage.
9. the image de-noising method based on multiresolution as claimed in claim 8, it is characterised in that described to be schemed based on initial rank As the image information after information denoising as a result, the image information obtained after original image information denoising includes：

After image information after initial rank image information denoising based on each passage is as a result, obtain original image information denoising Image information, the initial rank image information of each passage refer to stratum's figure in each passage corresponding to original image information As information.
10. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that the multiresolution point Solution includes appointing in Wavelet Transform, Gauss Pyramid transform method, picture contrast Pyramid transform method and gradient pyramid decomposition method Meaning is a kind of.
11. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that carry out inverse decompose and rebuild Image information exponent number it is bigger, the Gray homogeneity parameter σ used in the bilateral filtering_rIt is smaller.
12. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that described inverse based on carrying out Decompose the exponent number for the image information rebuild and Anscombe converts obtained image noise variance and determines in the bilateral filtering The Gray homogeneity parameter σ used_rIncluding：

Based on formula：σ_r=f (σ, layer) determines the Gray homogeneity parameter σ in bilateral filtering_r, wherein, σ converts for Anscombe Obtained image noise variance, layer are the exponent number for carrying out the inverse image information decomposed and rebuild, and f (σ, layer) is double to ask for The Gray homogeneity parameter σ used in the filtering of side_rFunction.
13. the image de-noising method based on multiresolution as claimed in claim 1, it is characterised in that the Gray homogeneity ginseng NumberWherein, σ converts obtained image noise variance for Anscombe, and layer rebuilds to carry out inverse decompose Image information exponent number.
14. the image de-noising method based on multiresolution as claimed in claim 8, it is characterised in that to the same of each passage Using identical parameter when approximation subband image corresponding to rank image information carries out bilateral filtering respectively.
15. the image de-noising method based on multiresolution as claimed in claim 14, it is characterised in that the identical parameter Including the Gray homogeneity parameter σ_r。
A kind of 16. image denoising device based on multiresolution, it is characterised in that including：

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

Reconstruction unit, rebuilds suitable for carrying out inverse decompose to every single order image information, to obtain per corresponding to single order image information Approximation subband image；

Filter unit, suitable for carrying out bilateral filtering to the approximation subband image per corresponding to single order image information, to obtain Per the image information after single order image information denoising；

Obtaining unit, suitable for based on the image information after initial rank image information denoising as a result, obtaining original image information denoising Image information afterwards, the initial rank image information refer to stratum's image information corresponding to original image information；

Wherein, further include：First converter unit, suitable for obtaining the approximation subband corresponding to every single order image information in reconstruction unit It is right before filter unit carries out bilateral filtering to the approximation subband image per corresponding to single order image information after image The approximation subband image carries out Anscombe conversion；

First inverse transformation subelement, suitable for carrying out bilateral filter to the approximation subband image per corresponding to single order image information After ripple, obtain per before the image information after single order image information denoising, the approximation subband image after bilateral filtering is carried out Anscombe inverse transformations；

Determination unit, suitable for converting obtained image based on the exponent number and Anscombe that carry out the inverse image information for decomposing reconstruction Noise variance determines the Gray homogeneity parameter σ used in the bilateral filtering_r。
17. the image denoising device based on multiresolution as claimed in claim 16, it is characterised in that the resolving cell bag Decomposition subelement is included, suitable for carrying out multistage decomposition respectively to multiple passages of original image information based on Multiresolution Decomposition method, To obtain every single order image information of each passage respectively, the multiple passage includes tri- passages of Y, U, V of image information.
18. the image denoising device based on multiresolution as claimed in claim 17, it is characterised in that the reconstruction unit bag Reconstruction subelement is included, is rebuild suitable for carrying out inverse decompose respectively to every single order image information of each passage, obtains each passage Per the approximation subband image corresponding to single order image information.
19. the image denoising device based on multiresolution as claimed in claim 18, it is characterised in that the filter unit bag Filtering subunit is included, bilateral filtering is carried out respectively 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.
20. the image denoising device based on multiresolution as claimed in claim 19, it is characterised in that the obtaining unit bag Acquisition subelement is included, suitable for the image information after the initial rank image information denoising based on each passage as a result, obtaining original graph As the image information after information denoising, the initial rank image information of each passage refers to original image information in each passage Corresponding stratum's image information.