CN105678704A - Nonlocal median value blind noise reduction method based on visual perception - Google Patents

Abstract

The invention provides a nonlocal median value blind noise reduction method based on visual perception, comprising the following steps: constructing an impulse noise blind detector based on the visual outlier measure of the pixels in a digital image, wherein the visual outlier measure is obtained by quantizing the visual similarities in impulse noise of different models and fusing different visual feature quantization results; extracting the nonlocal information of the image, and constructing a nonlocal median value calculation model; calculating regularization parameters according to the visual outlier measure and the nonlocal information, and establishing a nonlocal median value regularization item; and building a nonlocal median value noise reduction functional model, and adaptively repairing the noise pixels in the image. According to the blind noise reduction method of the invention, impulse noise of different models and densities in the digital image is processed in a unified manner according to the visual features of impulse noise, image self-similarity and outlier data mining, and the problem that noise pixels are hard to repair effectively due to unknown impulse model, high-density noise and image multi-modal complexity in an actual noise reduction process is solved.

Description

The non local intermediate value blind landing method for de-noising of a kind of view-based access control model perception

Technical field

The present invention relates to technical field of image processing, especially denoising digital picture technology, in particular to the non local intermediate value blind landing method for de-noising of a kind of view-based access control model perception, be suitable for the blind landing to perception model impulse noise non-in digital picture and make an uproar.

Background technology

Impulse noise is the undesired signal that in digital picture, a class is common, in the collection of image, transmission and storage process, produces because of the factor such as imperfect, mistake of imaging system, transmission medium and recording unit. Usually impulse noise is divided into three kinds according to brightness value distribution, it is fixed value, random value and fixed value random value mixed type respectively. Impulse noise in suppression digital picture is prerequisite and the basis of image analysis, understanding and identification, is also a Focal point and difficult point problem in this field. For concrete impulse model, domestic and international research institution and researchist have carried out and have studied widely, obtain a large amount of noise-reduction methods, generally speaking, transform domain can be divided into fall and make an uproar and two classes of making an uproar fall in spatial domain.

The thinking of transform domain noise-reduction method is transformed by observed image, in the transform domain as illustrated restraint speckle, then obtains most final decline by inverse transformation and makes an uproar result. This kind of method represents the openness as priori of territory taking the characteristic distributions of coefficient in transform domain and dictionary, there is powerful multi-resolution analysis and rarefaction representation ability, but it is coefficient complicated operation, optimum configurations and starting condition dependency is strong, and usually there is no overall situation solution, when repairing high-density noise image, complicated image, easily introduce false information, broken ring contrast gradient, as produced " ring ", " ladder ", " overlap ".

Spatial domain noise-reduction method is direct impulse noise mitigation in the territory, space of image, and by contrast, this method application relative maturity in the prior art, falls and make an uproar result also closer to visually-perceptible. The spatial domain minimizing technology of impulse noise is broadly divided into linear processes two class. Mean filter, middle value filtering and innovatory algorithm thereof are the most typical spatial domain filter algorithms, but only utilize average, intermediate value and simply be out of shape noise pixel reparation, and assignment precision is low, it is possible to can cause falling that result of making an uproar is fuzzy or detailed information loss.Theoretical and experiment shows, based on energy functional model canonical impulse noise minimizing technology can restraint speckle effectively, and more intactly protect the details of image. Around the design of regularization model, choosing of regularization parameter, objective function solve three work, domestic and international researchist proposes l₁Norm+guarantor limit regularization term, l₁Norm+total variation item, l₁Norm+partial differential constraint, l₁Norm+l_pThe outstanding algorithms such as norm constraint item. Generally, these methods have desirable anti-acoustic capability mostly, can suppress pulse and effectively protect the details of image, but prerequisite is accurately and reliably prior-constrained, and regularization parameter is chosen rationally. When choosing regularization parameter, algorithm adopts unified definition in advance mostly at present, then the mode by a large amount of optimum experimental. But the noise pixel of different characteristics in image being defined consistent parameter value so that the complex region of image, high-density noise region fidelity and smoothly unbalance, complicated image, high-density noise image are repaired accuracy and are reduced.

To sum up, existing majority method is known in impulse model, the raw density of noise, can obtain and fall result of making an uproar preferably when image to be repaired is relatively simple. But consider in actual noise reduction process, seldom can know the complexity of the concrete model of impulse noise in image, density and image to be repaired in advance, thus relating to the blind Detecting of non-perception model impulse noise, self-adapting detecting to different characteristics area pixel, reparation and during to problems such as effective removals of high-density impulse noise in image, existing noise-reduction method is difficult to effectively process.

Summary of the invention

The defect existed for prior art or deficiency; the present invention is intended to propose the non local intermediate value blind landing method for de-noising of a kind of view-based access control model perception; can when the complexity of unknown impulse noise model, noise density and image restraint speckle effectively, and intactly protect the detailed information of image.

Another object of the present invention is to, it is provided that the blind denoising device of the impulse noise of a kind of view-based access control model perception, and a kind of computer system made an uproar for the non local intermediate value blind landing realizing aforementioned view-based access control model perception.

The above-mentioned purpose of the present invention is realized by the technology feature of independent claim, dependent claims by select else or favourable in the way of develop the technology feature of independent claim.

For reaching above-mentioned purpose, a first aspect of the present invention proposes the non local intermediate value blind landing method for de-noising of a kind of view-based access control model perception, comprises the following steps:

Step 1, estimating from group based on the vision of pixel in digital picture, it is to construct impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

Step 2, the non-local information extracting image, it is to construct non local median calculation model;

Step 3, foundation vision estimate from group and non-local information calculates regularization parameter, sets up non local intermediate value regularization term;

Step 4, foundation step 2,3 are set up non local intermediate value and are fallen functional model of making an uproar, and self-adaptation repairs noise pixel in image.

According to the disclosure, the another aspect of the present invention also proposes the blind denoising device of the impulse noise of a kind of view-based access control model perception, comprising:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair noise pixel in image for self-adaptation.

Improvement according to the present invention, a third aspect of the present invention also proposes a kind of computer system made an uproar for the non local intermediate value blind landing realizing view-based access control model perception, and this computer system comprises:

Storer;

One or more treater;

One or more module, this one or more module is stored in which memory and is configured to perform by described one or more treater, and described one or more module comprises the module for performing following process:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair the noise pixel in image for self-adaptation.

Compared with prior art, blind landing proposed by the invention is made an uproar scheme, has significant useful effect:

1. quantizing from visual angle and merged different model impulse noise from character, it is proposed to pixel vision is estimated from group, it is to construct impulse noise blind tester, being that the impulse noise of different model is unified differentiates, it is achieved non-perception model impulse noise blind Detecting;

2. devise non local intermediate value impulse noise minimizing technology, adaptive regularization parameter, fall, in conjunction with non local intermediate value structure, functional model of making an uproar, increase the prior-constrained of objective function, thus improve the reparation precision of noise pixel;

3. at unknown impulse noise model, restraint speckle effectively when the complexity of noise density and image, and intactly protect the detailed information of image.

As long as it is to be understood that aforementioned concepts and all combinations of extra design of describing in further detail below can be regarded as a part for subject matter of the present disclosure when such design is not conflicting. In addition, all combinations of claimed theme are all regarded as a part for subject matter of the present disclosure.

Foregoing and other aspect, embodiment and feature that the present invention instructs can be understood by reference to the accompanying drawings from the following description more comprehensively. Feature and/or the useful effect of other additional aspect such as illustrative embodiments of the present invention will be obvious in the following description, or by the practice of the embodiment instructed according to the present invention is learnt.

Accompanying drawing explanation

Accompanying drawing is not intended to draw in proportion. In the accompanying drawings, each illustrating in each figure be identical or approximately uniform integral part can represent with identical label. For clarity, in each figure, not each integral part is all labeled. Now, the embodiment of all respects of the present invention also will be described with reference to accompanying drawing by example, wherein:

Fig. 1 is the schema of the non local intermediate value blind landing method for de-noising of the view-based access control model perception according to certain embodiments of the invention.

Fig. 2 a-2d is the image schematic diagram (noise density is 30%) of two class pulse jamming respectively.

Fig. 3 a-3c is the X-ray image and the noise reduction process result schematic diagram thereof that are subject to 50% random value noise jamming respectively.

Fig. 4 a-4c's is be subject to the dry tongue fur image of 70% fixed value noise and noise reduction process result schematic diagram thereof respectively.

Embodiment

In order to more understand the technology contents of the present invention, especially exemplified by specific embodiment and coordinate institute's accompanying drawings to be described as follows.

Each side with reference to the accompanying drawings to describe the present invention in the disclosure, shown in the drawings of the embodiment of many explanations. Embodiment of the present disclosure must not be intended to comprise all aspects of the present invention. It is to be understood that, multiple design presented hereinbefore and embodiment, and describe in more detail below those design and enforcement mode can in many ways in any one is implemented, this is because design disclosed in this invention and embodiment are not limited to any enforcement mode. In addition, aspects more disclosed by the invention can be used alone, or uses with any appropriately combined of other aspects disclosed by the invention.

According to embodiments of the invention, on the whole, the non local intermediate value blind landing method for de-noising of view-based access control model perception proposed by the invention, according to impulse noise visual characteristics, the self-similarity of image own and outlier data digging, the impulse noise of different model, density in unified processing digital images, it is intended to solve the noise pixel that in actual noise reduction process, impulse model the unknown, high-density noise and the multi-modal complicacy of image cause and be difficult to effectively repair problem.

The process made an uproar in whole blind landing roughly comprises two stages, respectively: 1) based on pixel in digital picture vision from group estimate structure impulse noise blind tester, to the blind differentiation of perception model impulse noise non-in digital picture; 2) in the noise pixel assignment stage, setting up cost functional model based on non local intermediate value noise reduction algorithm, self-adaptation repairs the noise pixel in image.

Aforesaid vision is estimated from group, by quantifying the vision general character of different model impulse noise, merges the acquisition of different visual signature quantized result, estimates from group with the vision that this calculates each pixel, for the detection of noise pixel provides tolerance foundation.

Described by following specific embodiment, the vision general character of the noise spike noise of different model is mainly reflected in three aspects: spatial distribution is isolated, connectivity is poor, brightness is abnormal. The blind landing method for de-noising of the present invention is intended to utilize these vision general character to quantize, and quantized result is merged, and calculates the vision of pixel in digital picture with this and estimate from group.

On the basis merged, we build blind tester, to the blind differentiation of perception model impulse noise non-in digital picture. Then, using the cost functional model based on non local intermediate value noise reduction algorithm that the noise in digital picture is carried out self-adaptation reparation.

Shown in accompanying drawing, the realization of the blind landing method for de-noising of previous embodiment is more specifically described.

Shown in composition graphs 1, according to embodiments of the invention, the non local intermediate value blind landing method for de-noising of a kind of view-based access control model perception, comprises the following steps:

Step 1, estimating from group based on the vision of pixel in digital picture, it is to construct impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

Step 2, the non-local information extracting image, it is to construct non local median calculation model;

Step 3, foundation vision estimate from group and non-local information calculates regularization parameter, sets up non local intermediate value regularization term;

Step 4, foundation step 2,3 are set up non local intermediate value and are fallen functional model of making an uproar, and self-adaptation repairs noise pixel in image.

As optional example, abovementioned steps 1, when realizing, comprises impulse noise vision general character and quantizes, merges quantized result and build blind tester two processes, be specifically described respectively below.

First, in conjunction with the vision common feature of the noise spike noise of different model, the process that our paired pulses noise vision general character quantizes comprises: the space of pixel quantizes from group and the brightness of pixel quantizes from group.

A. the space of pixel quantizes from group: according to the data characteristics of digital picture local UNICOM, space is utilized to estimate from group, adopting the abnormal number mining algorithm based on connection property, (IM:isolationmeasurement) IM (i) is estimated in the space calculating any pixel i;

B. the brightness of pixel quantizes from group: based on weber-Fei Xina law, the minimum brightness of goal in research pixel regional area can feel poor, estimates quantizing pixel i in conjunction with local space from group with this and estimates (LTM:luminancetransitionmeasurement) LTM (i) relative to the brightness of its background from group.

Merge the result that the space of aforementioned pixel quantizes from group from the brightness of group's quantification and pixel, the vision that we can obtain each pixel estimates (VPOM:visualperceptionoutliermeasurement) VPOM (i) from group, and constructs on this basis based on the impulse noise blind tester that pixel vision is estimated from group.

Owing to the computation model of non-local mean Denoising Algorithm derives from being that independent variable(s) function solves mnm. taking P.

min_P(Σ_j∈P(i)ω_i,j|P-P_j|²)

In formula, i represents object pixel, and j represents non local pixel, P_jBeing the image block centered by j, P (i) is the self-similar pixel search window of pixel i, ω_i,jIt it is the similarity of pixel i and j.

Consider the nonlinear characteristic of impulse noise, average in non-local mean algorithm is solved and converts intermediate value to and solve the assignment accuracy that can improve noise pixel, seek the non local median algorithm of weighted median of all self-similar pixel of object pixel i, solve the mnm. that P is independent variable(s) function and obtain intermediate value.

Therefore, in aforesaid step 2, we construct non local median calculation model by following mode:

min_P(Σ_j∈P(i)ω_i,j|P-P_j|²)

As the aforementioned, i represents object pixel, and j represents non local pixel, P_jBeing the image block centered by j, P (i) is the self-similar pixel search window of pixel i, ω_i,jIt it is the similarity of pixel i and j.

Therefore solve the mnm. that the P obtained is independent variable(s) function and namely obtain non local intermediate value.

Meanwhile, owing to noise pixel is very little to the reparation of i contribution, so estimating the fuzzy membership with pixel from group and optimize weights omega in conjunction with pixel in certain embodiments_i,j. In order to solve corresponding weighting average fast, as preferred example, the way of setting threshold restriction self-similar pixel number is adopted to solve mnm., to reduce computation complexity, it is to increase the speed of convergence of computation model.

Certainly, in further embodiments, it is also possible to adopt other known mode existing to solve, do not repeat them here.

We are taking energy functional model as framework, utilize non local intermediate value structure canonical item, set up and fall functional model of making an uproar, more specifically will describe in following content.

In step 3, we are with the detected result (in iterative process pixel i be judged as number of times T (i) of noise pixel) of the vision of pixel from group's degree, noise pixel, analyze the local information in region residing for pixel, for pixel determines regularization parameter adaptively:

λ (i)=λ₀f₁(VPOM(i))f₂(T(i))

λ₀It is initial regularization parameter, f₁And f₂It it is weighting function.

In step 4, we combine local canonical item, non local median calculation model and the non local intermediate value regularization term of adaptive regularization parametric configuration.

Shown in Fig. 1, the flow process adopting the non local intermediate value blind landing method for de-noising of view-based access control model perception of previous embodiment that digital picture carries out blind noise reduction process is carried out exemplary explanation.

Step 1, the observed image u inputting a width and being subject to impulse noise interference

The image u of this step input is subject to impulse noise interference, but concrete impulse model is unknown, it may be that fixed value impulse model or random value impulse model, certainly it could even be possible to be the two mixture model. Such as the example of Fig. 2 a-2d, the noise density of the picture in figure is 30%.

In step 2, calculating observation image u, the space of any pixel i, brightness are estimated from group, and fusion calculation result obtains the vision of each pixel in image and estimates from group

A. in computed image u, the space of any pixel i is estimated from group

According to people's eye to the visually-perceptible of brightness in 9 × 9 fields centered by pixel i, utilize the variable threshold value LUT (u of following formula_l), wherein l=i+k, k ∈ [-4,4], calculates UNICOM's pixel chain of this pixel, and with maximum UNICOM's pixel chain, the number of pixels namely comprising the maximum pixel chain of number of pixels is to define the connectivity parameters C of this pixel.

In formula, u_lRepresenting the brightness of current pixel l in pixel chain, LUT (l) is the variable threshold value by background of the brightness of pixel l.

The window of centered by pixel l 5 × 5 calculates 10 pixels minimum with pixel i luminance difference, the connectivity parameters finding out these pixels gets its intermediate value C1, and then get the intermediate value C2 of the connectivity parameters of all pixels in whole 5 × 5 windows, get connectivity measurement IM (i) of ratio as pixel i of the two.

B. in computed image u, the brightness of any pixel i is estimated from group

Estimating from group according to space, the α calculating this image block according to following formula in the window of centered by pixel i 5 × 5 cuts out the background luminance of average as regional area:

In formula, u_αBeing that α cuts out average, n is the number of pixel in image block, u_kIt is kth the value after n pixel is arranged from small to large, gets α=18 here.

In the window of centered by pixel i 5 × 5, calculate 10 pixels minimum with pixel i luminance difference, calculate the luminance difference S of these pixels and pixel i_t, t ∈ [1,10], calculates the local visual luminance difference of pixel according to Fei Xina law, as shown in the formula:

C. merging brightness to estimate from group and estimate from group with space, in computed image u, the vision of any pixel i estimates VPOM (i) from group, and this numerical value is the foundation judging whether this pixel i belongs to noise pixel.

VPOM (i)=β IM (i)+γ LTM (i)

In upper formula, β and γ be brightness from the fusion coefficients that group estimates with space is estimated from group, get β=γ=0.5 here.

Step 3, utilize the vision of each pixel in image u to estimate from group, build the blind tester of following formula, by threshold value T^kNoise pixel in detected image:

T^k=T^k-10.9, k=1,2,3 ... K_max

In upper formula, k is iteration number of times (noise reduction process of the present invention takes iterative processing), K_maxIt it is maximum iteration time.

Step 4, the non-local information extracting any pixel i in image u

Choose the image block of centered by pixel i 21 × 21, utilize following kernel function to calculate the self-similar pixel weights omega of pixel i in this block of pixels_i,j:

In upper formula, u_iAnd u_jIt is the pixel value of i and j respectively, λ=16.

Step 5, the regularization parameter of any pixel i in computed image u

With the detected result (in iterative process pixel i be judged as number of times T (i) of noise pixel) of the vision of pixel from group's degree, noise pixel, for pixel determines regularization parameter adaptively.

λ (i)=λ₀f₁(VPOM(i))f₂(T(i))

λ₀It is initial regularization parameter, gets λ here₀=0.01, f₁And f₂It is weighting function, wherein,

In upper formula, VPOM (i+k) is that the vision of the field interior pixel of 3 × 3 centered by pixel i is from group's measure value.

K in upper formula_maxRepresent the maximum iteration time in noise reduction process process.

Step 6, sets up target and falls the functional model F that makes an uproar_r:R^M×N→ R, carries out valuation reparation to pixel to be repaired in image u (noise pixel that step 3 detects out)

In upper formula, V ≡ 1,2 ..., M} × 1,2 ..., N}, it represents that a width size is the image of M × N, and r represents and falls reparation image of making an uproar, and i here represents target pixel points, and j is the self-similar pixel of pixel i, and Q is 49 pixels the most similar to i in self-similar pixel, r_iRepresent the reparation result of pixel i, r_jRepresenting the reparation result of pixel j, the value of λ is 16.

By asking for F_rU noise pixel valuation in image is repaired by the mnm. of (), the falling to make an uproar of the present embodiment is iterative algorithm, and by iteration progressively to the impulse noise detection reparation in image, image u repairs the most at last.

Blind landing method for de-noising described by embodiment as previously, utilizes the method that digital picture is carried out the example of noise reduction process below in conjunction with some, further describes its noise reduction.

1) experiment condition windows8, CPUInter (R) Core (TM) i5,2.5GHz, software platform is Matlab7.9.1.

First data that emulation is chosen are the X-ray images of the random value sound pollution being subject to 50%, and such as Fig. 3 a, the 2nd data are subject to 70% fixed value sound pollution tongue fur image 4a. 3rd data are subject to the Lena image of 30% mixed noise interference, Baboon image, Goldhill image, Boat image, Pepper image.

2) experiment content and result

The method of traditional non-local mean method (NLM method) and previous embodiment is used to process first, the 2nd experimental data under these experimental conditions respectively. What NLM method obtained falls result of making an uproar such as Fig. 3 b, Fig. 4 b; What the present invention obtained falls result of making an uproar such as Fig. 3 c, Fig. 4 c.

Relatively Fig. 3 b, Fig. 3 c and Fig. 4 b, Fig. 4 c can find out, it is more serious that image detail loss of making an uproar falls in the NLM method of prior art, has there is destruction to a certain degree in the edge of image, the method of present invention not only can effectively remove the details that noise can keep again image simultaneously, is obviously better than the NLM method of prior art vision effect.

Utilizing NLM method and the inventive method to the 3rd data noise reduction process under these experimental conditions, calculate the overall peaks signal to noise ratio PSNR and mean absolute error MAE that fall result images of making an uproar, result is such as table 1.

The PSNR value of the image repair result of table 1 30% impulse noise interference and MAE value

As can be seen from Table 1, the method for present invention is obviously better than NLM algorithm in the noise reduction of mixture model impulse noise, and the PSNR value of its reflection picture quality is higher, and the MAE value of reflection image detail loss is less.

To sum up; the inventive method has obvious advantage compared with traditional NLM method in the various types of impulse noise of removal; anti-acoustic capability is better; gained result images PSNR significantly improves; details protection is more complete; and achieve unknown impulse model noise blind Detecting and to the self-adaptation reparation of noise pixel in high-density noise image and complicated image, it is to increase the accuracy that noise pixel is repaired.

According to the disclosure, the present invention also relates to the blind denoising device of the impulse noise of a kind of view-based access control model perception, comprising:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair noise pixel in image for self-adaptation.

It is to be understood that, the first module that the present embodiment proposes, the 2nd module, the 3rd module and four module, its function, effect and effect are illustrated in the description of the non local intermediate value blind landing method for de-noising of above view-based access control model perception, its implementation and in the aforementioned embodiment about blind landing method for de-noising, done exemplary illustration, do not repeat them here.

Aforementioned embodiments according to the present invention, the non local intermediate value blind landing method for de-noising of such as view-based access control model perception and the blind denoising device of non local intermediate value of view-based access control model perception, the present invention also proposes a kind of computer system made an uproar for the non local intermediate value blind landing realizing view-based access control model perception, and this computer system comprises:

Storer;

One or more treater;

One or more module, this one or more module is stored in which memory and is configured to perform by described one or more treater, and described one or more module comprises the module for performing following process:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair the noise pixel in image for self-adaptation.

It is to be understood that aforesaid storer is used for program of depositing and data, for performing for described treater. These storeies can be such as take disk as the storer of storage media, or storer based on flash memory chip etc.

Obviously, in the computer system of the present embodiment, these modules stored, can perform by one or more treater and realize the blind noise reduction process described by non local intermediate value blind landing method for de-noising of aforementioned view-based access control model perception, reach the blind Detecting to unknown impulse model noise and the self-adaptation reparation to noise pixel in high-density noise image and complicated image, it is to increase the accuracy that noise pixel is repaired.

Although the present invention with better embodiment disclose as above, so itself and be not used to limit the present invention.Persond having ordinary knowledge in the technical field of the present invention, without departing from the spirit and scope of the present invention, when being used for a variety of modifications and variations. Therefore, protection scope of the present invention is when being as the criterion depending on those as defined in claim.

Claims (8)

1. the non local intermediate value blind landing method for de-noising of a view-based access control model perception, it is characterised in that, comprise the following steps:

Step 1, vision based on pixel in digital picture estimate structure impulse noise blind tester from group, and described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

Step 2, the non-local information extracting image, it is to construct non local median calculation model;

Step 3, foundation vision estimate from group and non-local information calculates regularization parameter, sets up non local intermediate value regularization term;

Step 4, foundation step 2,3 are set up non local intermediate value and are fallen functional model of making an uproar, and self-adaptation repairs noise pixel in image.

2. the non local intermediate value blind landing method for de-noising of view-based access control model perception according to claim 1, it is characterised in that, in described step 1, based on pixel in digital picture vision from group estimate structure impulse noise blind tester specific implementation comprise:

1-1) calculate a space being subject to any pixel i in the image u of impulse noise interference to estimate from group;

1-2) calculate earlier figures to estimate from group as the brightness of any pixel i in u;

1-3) merging brightness to estimate from group and estimate from group with space, in computed image u, the vision of any pixel i estimates VPOM (i) from group, and this numerical value is the foundation judging whether this pixel i belongs to noise pixel; And

1-4) utilize the vision of each pixel in image u to estimate VPOM (i) from group, build the blind tester of following formula, by threshold value T^kNoise pixel in detected image:

T^k=T^k-10.9, k=1,2,3 ... K_max

In upper formula, k is iteration number of times, K_maxIt it is maximum iteration time.

3. the non local intermediate value blind landing method for de-noising of view-based access control model perception according to claim 2, it is characterised in that, described step 1-1) in, the account form that the space of any pixel i is estimated from group is as follows:

According to people's eye to the visually-perceptible of brightness in 9 × 9 fields centered by pixel i, utilize the variable threshold value LUT (u of following formula_l), wherein l=i+k, k ∈ [-4,4], calculates UNICOM's pixel chain of this pixel, and with maximum UNICOM's pixel chain, the number of pixels namely comprising the maximum pixel chain of number of pixels is to define the connectivity parameters C of this pixel:

$LUT\left(u_l\right)=\left\{\begin{array}{c}17(1-\sqrt{\frac{u_l}{127}})+3,0\≤u_l\≤127\\ \frac{3}{128}(u_l-127)+3,128\≤u_l\≤255\end{array}\right.$

In formula, u_lRepresenting the brightness of current pixel l in pixel chain, LUT (l) is the variable threshold value by background of the brightness of pixel l;

The window of centered by pixel l 5 × 5 calculates 10 pixels minimum with pixel i luminance difference, the connectivity parameters finding out these pixels gets its intermediate value C1, and then get the intermediate value C2 of the connectivity parameters of all pixels in whole 5 × 5 windows, getting connectivity measurement IM (i) of ratio as pixel i of the two, the space that this connectivity measurement IM (i) is pixel i is estimated from group;

$IM\left(i\right)=\frac{C_1}{C_2};$

Further, abovementioned steps 1-2) in the account form estimated from group of brightness as follows:

Estimating from group according to space, the α calculating this image block according to following formula in the window of centered by pixel i 5 × 5 cuts out the background luminance of average as regional area:

$u_{\mathrm{\α}}=\frac{1}{n-\mathrm{\α}}\underset{k=(\mathrm{\α}/2)+1}{\overset{n-(\mathrm{\α}/2)-1}{\Σ}}{u}_k$

In formula, u_αBeing that α cuts out average, n is the number of pixel in image block, u_kIt is kth the value after n pixel is arranged from small to large, α=18;

In the window of centered by pixel i 5 × 5, calculate 10 pixels minimum with pixel i luminance difference, calculate the luminance difference S of these pixels and pixel i_t, t ∈ [1,10], calculates the local visual luminance difference of pixel according to Fei Xina law, as shown in the formula:

$LTM\left(i\right)=\underset{t=1}{\overset{10}{\Σ}}{S}_t\log \frac{S_t}{u_{\mathrm{\α}}}$

The brightness that local visual luminance difference LTM (i) calculated according to this is pixel i is estimated from group;

And in step 1-3) in, utilizing following formula fusion brightness to estimate from group and estimate from group with space, in computed image u, the vision of any pixel i estimates VPOM (i) from group:

VPOM (i)=β IM (i)+γ LTM (i)

In upper formula, β and γ be brightness from the fusion coefficients that group estimates with space is estimated from group, get β=γ=0.5.

4. the non local intermediate value blind landing method for de-noising of view-based access control model perception according to claim 3, it is characterised in that, the specific implementation of abovementioned steps 2 comprises:

Choose the image block of centered by pixel i 21 × 21, utilize following kernel function to calculate the self-similar pixel weights omega of pixel i in this block of pixels_i,j:

${\mathrm{\ω}}_{i,j}=\exp \left(-\frac{\left|\right|u_i-u_j\left|\right|}{\mathrm{\λ}}\right)$

In upper formula, u_iAnd u_jIt is the pixel value of i and j respectively, λ=16.

5. the non local intermediate value blind landing method for de-noising of view-based access control model perception according to claim 1, it is characterised in that, the specific implementation of described step 3 comprises:

With the detected result of the vision of pixel from group's degree, noise pixel, for pixel determines regularization parameter λ (i) adaptively:

λ (i)=λ₀f₁(VPOM(i))f₂(T(i))

λ₀It is initial regularization parameter, gets λ here₀=0.01, f₁And f₂Being weighting function, T (i) is the detected result of noise pixel, and namely in iterative process, pixel i is judged as the number of times of noise pixel;

Wherein,

$f_1\left(VPOM\right(i\left)\right)=min\{1,\frac{VPOM\left(i\right)}{median\{VPOM(i+k),-1\≤k\≤1\}}\}$

In upper formula, VPOM (i+k) is that the vision of the field interior pixel of 3 × 3 centered by pixel i is from group's measure value;

$f_2\left(T\right(i\left)\right)=\left\{\begin{array}{c}\frac{T\left(i\right)+1}{K_{max}},K_{max}>3\\ \frac{1}{K_{max}},K_{max}\≤3\end{array}\right.$

K in upper formula_maxRepresent the maximum iteration time in noise reduction process process.

6. the non local intermediate value blind landing method for de-noising of view-based access control model perception according to claim 1, it is characterised in that, the specific implementation of described step 4 comprises:

The non local intermediate value regularization term that the non local median calculation model built according to abovementioned steps 2 and step 3 are set up is carried out establishing target and is fallen the functional model F that makes an uproar_r:R^M×N→ R, carries out valuation reparation to pixel to be repaired in image u:

$F_r\left(u\right)=\underset{i\∈V}{\Σ}\left|\right|r_i-u_i\left|\right|+\mathrm{\λ}\underset{i\∈V}{\Σ}\underset{j\∈Q}{\Σ}{\mathrm{\ω}}_{i,j}|r_i-r_j|$

V ≡ in upper formula 1,2 ..., M} × 1,2 ..., N}, it represents that a width size is the image of M × N, and r represents and falls reparation image of making an uproar, and i here represents target pixel points, and j is the self-similar pixel of pixel i, and Q is 49 pixels the most similar to i in self-similar pixel, r_iRepresent the reparation result of pixel i, r_jRepresenting the reparation result of pixel j, the value of λ is 16;

By asking for F_rU noise pixel valuation in image is repaired by the mnm. of (), by iteration progressively to the impulse noise detection reparation in image, image u repairs the most at last.

7. the blind denoising device of the impulse noise of a view-based access control model perception, it is characterised in that, comprising:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair noise pixel in image for self-adaptation.

8. the computer system made an uproar for realizing the non local intermediate value blind landing of view-based access control model perception, it is characterised in that, this computer system comprises:

Storer;

One or more treater;

One or more module, this one or more module is stored in which memory and is configured to perform by described one or more treater, and described one or more module comprises the module for performing following process:

For estimating from group based on the vision of pixel in digital picture, it is to construct the first module of impulse noise blind tester, described vision estimates the vision general character of different model impulse noise by quantifying from group, merges different visual signature quantized result and obtains;

For extracting the non-local information of image, it is to construct the 2nd module of non local median calculation model;

For estimating and non-local information calculating regularization parameter from group according to vision, set up the 3rd module of non local intermediate value regularization term;

Non local intermediate value regularization term for setting up according to the non local median calculation model constructed by aforementioned 2nd module and the 3rd module builds non local intermediate value and falls functional model of making an uproar, and this non local intermediate value is fallen functional model of making an uproar and is configured to repair the noise pixel in image for self-adaptation.