CN103578089A - Depth map filtering method and system based on local binary pattern operator guidance - Google Patents

Abstract

The invention discloses a depth map filtering method and system based on local binary pattern operator guidance. The method comprises the steps that firstly, the local binary pattern operator of the adjacent pixel point of a current pixel point in a colorful image is obtained; then, the adjacent pixel point belonging to the same object with the current pixel point is obtained through distinguishing based on the local binary pattern operator of the adjacent pixel point; finally, the adjacent pixel point belonging to the same object with the current pixel point is used as the reference pixel of depth map filtering to obtain the depth map filtering result of the current pixel point. The filtering method is added to an H.264/MVC encoding frame as loop filtering, and the depth map encoding efficiency and the quality of synthetic and virtual view points can be effectively improved.

Description

The depth map filtering method and the system that based on local binary patterns operator, instruct

Technical field

The invention belongs to image filtering field, belong to especially in three-dimensional video-frequency for the depth map filtering field of many viewpoints plus depth video format, be specifically related to a kind of depth map filtering method and system instructing based on local binary patterns operator.

Background technology

When in recent years, 3D film and TV are for spectators bring better degree of depth feeling of immersion and stereoscopic vision impression, also introduced and user between interaction.Stereoscopic vision impression can obtain by synchronization scene different points of view being caught to the multi-viewpoint three-dimensional video (Multi-view video, MVV) that obtain, but the weak point of the three-dimensional video-frequency based on MVV is to provide the free interactivity with user.Therefore, how utilizing the synthetic virtual view of virtual viewpoint rendering technology based on depth image, is current research heat topic.

P.Merkle thinks that depth map has unique data structure, thinks that it is to consist of a large amount of different homogeneous regions, and these homogeneous regions are divided [1] by object edge.Therefore in common hybrid encoding frame, such as H.264/AVC, quantize and module unit encoding operation all can cause the serious artifact effect of object edge in depth map.And depth map provides the geological information in 3D scene, thus the error of depth map will after virtual view cause geometric distortion in synthetic.The error of the pixel in depth map in homogeneous region, with the error of object edge pixel in depth map, the two is compared, and the latter more can have a strong impact on the quality [2] of synthetic virtual view.So keep the acuteness at edge of object extremely important when coding depth figure.

In order to meet the background of depth map codec, K.Oh ^[3]h.264 a loop filter is added in framework, the depth map of coding is carried out to denoising.This kind of loop filter be by limiting geometric distance, image similarity and occurrence frequency, and vicinity points is around operated, and recovers noise in depth map, effectively improved the efficiency of depth map encoding and the quality of synthetic virtual view.But the pixel having damaged in depth map also may reduce the accuracy of reconstructed image vegetarian refreshments, therefore, the method bit rate is lower, and filter effect is poorer.

Thought based on bilateral filtering, liu ^[4]propose a kind of method of three limit filtering, the method, when determining the value of weighting coefficient, increases the structural similarity of having considered cromogram and corresponding depth map.This filtering method can effectively be removed the artifact effect of depth map encoding in encoding-decoding process, and under identical bit, obtains the synthetic viewpoint of better quality.But when around vicinity points only has base point to be similitude, the filtering result of this filtering method just seems unstable.The problem existing in order to solve this filtering method, obtains stable filtering Output rusults, and wave filter need to be eliminated the impact of dissimilar vicinity points around.

Summary of the invention

The problem existing for prior art, the present invention is based on the thought of bilateral filtering, a kind of depth map filtering method and system instructing based on local binary patterns operator proposed, the method is under local binary patterns operator instructs, can effectively differentiate the object whether vicinity points belongs to current pixel point place, and only choose belong to same object with current pixel point vicinity points as carrying out filtering with reference to pixel, thereby obtain stable filtering result.

In order to solve the problems of the technologies described above, the present invention adopts following technical scheme:

One, the depth map filtering method that local binary patterns operator instructs, comprises step:

Step 1, for coloured image corresponding to depth image, obtains the local binary patterns operator of the vicinity points of current pixel point;

Step 2, the local binary patterns operator based on vicinity points is differentiated and is obtained belonging to current pixel point the vicinity points of same object, and upgrades the local binary patterns operator of vicinity points;

Step 3, the geometric distance acquisition spatial domain factor of influence according to the vicinity points of current pixel point to current pixel point;

Step 4, according to the vicinity points of current pixel point and current pixel point, the similarity of degree of depth intensity in depth image obtains image area factor of influence;

Step 5, the constraint factor that vicinity points local binary patterns operator, spatial domain factor of influence and the image area factor of influence of take after upgrading is filtration combined weighted coefficient obtains the filtering result of current pixel point.

Above-mentioned steps 1 further comprises that sub-step is:

In coloured image corresponding to depth image, obtain the brightness value of each vicinity points of current pixel point, and the brightness value of each vicinity points and current pixel point is made respectively to difference and get two-value symbol, obtain the local binary patterns operator of vicinity points.

Above-mentioned steps 2 further comprises sub-step:

2-1 is divided into two parts according to the local binary patterns operator of vicinity points by vicinity points;

2-2 judges according to the logical reach of above-mentioned two parts vicinity points and current pixel point whether vicinity points belongs to same object with current pixel point;

2-3 is updated to 1 by the local binary patterns operator that belongs to the vicinity points of same object with current pixel point, and the local binary patterns operator that does not belong to the vicinity points of same object with current pixel point is updated to 0.

The spatial domain factor of influence of step 3 gained is G _q(|| p-c||), wherein, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c.

The image area factor of influence of step 4 gained is G _s(d _p-d _c), wherein, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity.

The depth map filtering result G of the current pixel point of step 5 gained _cfor:

$G_c=\frac{1}{F_c}{\mathrm{\Σ}}_{p\∈w}[G_S(d_p-d_c)\·G_Q\left(\right||p-c|\left|\right)\·{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)\·d_p]$

Wherein:

W is the set of all vicinity points of current pixel point c;

P is the vicinity points of current pixel point c;

G _s(d _p-d _c) be image area factor of influence, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity;

G _q(|| p-c||) be spatial domain factor of influence, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c;

for the local binary patterns operator after vicinity points p renewal, g _pand g _cbe respectively vicinity points p and the current pixel point c brightness value in coloured image corresponding to depth image;

$F_c={\mathrm{\Σ}}_{p\∈w}[G_S(d_p-d_c)\·G_Q\left(\right||p-c|\left|\right)\·{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)],$ For weighting coefficient summation.

Two, the depth map filtering system that local binary patterns operator instructs, comprising:

Local binary patterns operator acquisition module, is used for, for coloured image corresponding to depth image, obtaining the local binary patterns operator of the vicinity points of current pixel point;

Vicinity points judge module, the local binary patterns operator being used for based on vicinity points is differentiated and is obtained belonging to current pixel point the vicinity points of same object, and upgrades the local binary patterns operator of vicinity points;

Spatial domain factor of influence acquisition module, is used for to the geometric distance of current pixel point, obtaining spatial domain factor of influence according to the vicinity points of current pixel point;

Image area factor of influence acquisition module, is used for the similarity of according to the vicinity points of current pixel point and current pixel point degree of depth intensity in depth image to obtain image area factor of influence;

Filtration module, is used for the filtering result that constraint factor that vicinity points local binary patterns operator, spatial domain factor of influence and image area factor of influence after upgrading are filtration combined weighted coefficient obtains current pixel point.

Above-mentioned vicinity points judge module further comprises submodule:

Vicinity points is divided module, is used for, according to the local binary patterns operator of vicinity points, vicinity points is divided into two parts;

Judge module, is used for judging according to the logical reach of above-mentioned two parts vicinity points and current pixel point whether vicinity points belongs to same object with current pixel point;

Local binary patterns operator update module, be used for the local binary patterns operator that belongs to the vicinity points of same object with current pixel point to be updated to 1, the local binary patterns operator that does not belong to the vicinity points of same object with current pixel point is updated to 0.

The spatial domain factor of influence that above-mentioned spatial domain factor of influence acquisition module obtains is G _q(|| p-c||), wherein, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c.

Above-mentioned image area factor of influence is G _s(d _p-d _c), wherein, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity.

The filtering result G of the current pixel point that above-mentioned filtration module obtains _cfor:

$G_c=\frac{1}{F_c}{\mathrm{\Σ}}_{p\∈w}[G_S(d_p-d_c)\·G_Q\left(\right||p-c|\left|\right)\·{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)\·d_p]$

Wherein:

W is the set of all vicinity points of current pixel point c;

P is the vicinity points of current pixel point c;

G _s(d _p-d _c) be image area factor of influence, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity;

G _q(|| p-c||) be spatial domain factor of influence, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c;

for the local binary patterns operator after vicinity points p renewal, g _pand g _cbe respectively vicinity points p and the current pixel point c brightness value in coloured image corresponding to depth image;

$F_c={\mathrm{\Σ}}_{p\∈w}[G_S(d_p-d_c)\·G_Q\left(\right||p-c|\left|\right)\·{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)],$ For weighting coefficient summation.

Compared with prior art, the present invention has following feature and beneficial effect:

The present invention is based on bilateral filtering thought, proposed a kind of depth map filtering method instructing based on local binary patterns operator.Local binary patterns operator (LBP) is at first by TImo Ojala ^[5]propose, as a kind of mode of description microtexture of stalwartness, local binary patterns operator has been described circular to being uniformly distributed of vicinity points around in territory, city.Local binary patterns operator (LBP) can the various structural information features of Precise Representation, such as: edge, turning, sharp point and flat site.

And the input pixel that depth map wave filter is selected, need to belong to current pixel point the vicinity points of an object, so the local binary patterns operator based on coloured image corresponding to depth image, can judge whether vicinity points belongs to same object with current pixel point.In addition,, even if dull gray scale is carried out change quantization operation, local binary patterns operator (LBP) also has good robustness ^[6], in depth map, slight quantization error is not enough to affect the accuracy of LBP description scheme information.Therefore, with the local binary patterns operator of coloured image, control the input reference pixel of wave filter, compare liu ^[4]with the brightness value of coloured image, come calculation of filtered weighting coefficient to have more robustness.

Existing depth map filtering method can not be distinguished the object whether vicinity points belongs to current pixel point place, and the vicinity points that makes not belong to same object with current pixel point is also used to be used as the input of weighted filtering, causes filtering unstable result.And the present invention is under local binary patterns operator instructs, can effectively differentiate the object whether vicinity points belongs to current pixel point place, and only choose belong to same object with current pixel point vicinity points as carrying out filtering with reference to pixel, thereby can obtain stable, accurate filtering result.

H.264/MVC depth map filtering method of the present invention is joined in coding framework as loop filtering, by removing the noise and increasing deep video code efficiency of depth map, can further improve based on virtual visual point image subjective and objective quality corresponding to the colored plus depth form three-dimensional video-frequency of many viewpoints.

embodiment

The invention discloses a kind of depth map filtering method instructing based on local binary patterns operator, ultimate principle is: the local binary patterns operator that obtains the coloured image that depth map is corresponding, under the local binary patterns operator of coloured image instructs, whether vicinity points and the current pixel point of effectively differentiating current pixel point belong to same object, and only choose the vicinity points that belongs to same object with current pixel point, as the reference pixel of depth map filtering.

Below in conjunction with specific embodiment, further illustrate the inventive method, concrete steps are as follows:

Step 1, for coloured image corresponding to depth image, obtains the local binary patterns operator LBP of the vicinity points of current pixel point _p.R.

Local binary patterns operator LBP operates take P the vicinity points of current pixel point in the circular symmetric area that the center of circle, the R of take are radius, therefore, in this concrete enforcement, LBP is designated as to LBP _p.R.Above-mentioned take current pixel point as the center of circle, the R of the take circle that is radius the adjacent windows that is current pixel point place, radius R is controlled the area of space size of local binary patterns operator; The number P of vicinity points, controls the quantization degree of local binary patterns operator on space, angle ^[5].

Brightness value by vicinity points in coloured image corresponding to depth image is designated as g _p, p represents p vicinity points, p=0, and 1 ..., P-1, the number that P is vicinity points.By the brightness value g of each vicinity points _prespectively with the brightness value g of current pixel point _cmake the poor luminance difference (g that obtains _p-g _c), due to (g _p-g _c) and g _cseparate, therefore the brightness value g of vicinity points p in coloured image _pjoint distribution see formula (1):

T＝T(g _c,g ₀,g ₁,...,g _P-1) （1）

Luminance difference (the g of vicinity points p and current pixel point c _p-g _c) joint distribution ^[7]see formula (2):

T≈T(g ₀-g _c,g ₁-g _c,...,g _P-1-g _c) （2）

Luminance difference (g _p-g _c) symbol can not change along with the change of average brightness value, therefore, can use two-value symbol s (g _p-g _c) replacement (g _p-g _c), see formula (3):

$s(g_p-g_c)=\left\{\begin{array}{c}1,g_p-g_c\&GreaterEqual;0\\ 0,g_p-g_c<0\end{array}\right.---\left(3\right)$

In order to describe more accurately object edge, the present invention does not carry out the rotary manipulation of local binary patterns operator ^[5].LBP _p.Rbe equivalent to two-value joint distribution, see formula (4) and (5):

LBP _P.R(g _p-g _c)＝s(g _p-g _c) （4）

LBP _P.R＝T(s(g ₀-g _c),s(g ₁-g _c),...,s(g _P-1-g _c)) （5）

Wherein, p is p vicinity points, p=0, and 1 ..., P-1.

Step 2, the local binary patterns operator LBP based on vicinity points _p.Rvicinity points is judged to selection.

The local binary patterns operator LBP of vicinity points in coloured image _p.Rthere is the characteristic of describing object edge.According to the local binary patterns operator LBP of vicinity points _p.Rvalue is divided into two parts by the vicinity points in adjacent windows, and the local binary patterns operator value of a part of vicinity points is 1, and the local binary patterns operator value of another part vicinity points is 0.

Calculate respectively above-mentioned two parts vicinity points to the logical reach sum of current pixel point _zeroand sum _one:

sum _one＝Σ _p∈w[LBP _P.R(g _p-g _c)·|d _p-d _c|] （6）

sum _zero＝Σ _p∈w[|1-LBP _P.R(g _p-g _c)|·|d _p-d _c|] （7）

Wherein:

Sum _zerofor the logical reach of the local binary patterns operator value vicinity points that is 0 to current pixel point;

Sum _onefor the logical reach of the local binary patterns operator value vicinity points that is 1 to current pixel point;

LBP _p.R(g _p-g _c)=s (g _p-g _c), be the local binary patterns operator of vicinity points p, s (g _p-g _c) be the luminance difference (g of vicinity points p and current pixel point c _p-g _c) two-value symbol;

W is the set of the interior vicinity points of adjacent windows of current pixel point;

P is p vicinity points;

D _cfor the degree of depth intensity of current pixel point c in depth image;

D _pfor the degree of depth intensity of vicinity points p in depth image.

That a part of vicinity points and current pixel point that logical reach is less belong to same object, the local binary patterns operator of this part vicinity points is updated to 1, the local binary patterns operator of another part vicinity points is updated to 0, sees formula (8):

With the local binary patterns operator upgrading

as the filtration combined weighted restricted coefficients of equation factor, can avoid not belonging to current pixel point the interference of the vicinity points of same object, eliminate the instability of filtering.

Step 3, obtains spatial domain factor of influence.

Each vicinity points p comprising in adjacent windows for current pixel point c place, p=0,1 ..., P-1, P is the number of the vicinity points that comprises in adjacent windows, calculates respectively the geometric distance of each vicinity points p and current pixel point c || p-c||, with G _q(|| p-c||) be the spatial domain factor of influence that vicinity points p is corresponding, G _qfor Gaussian function.G _q(|| p-c||) can be expressed as follows:

$G_Q\left(\right||p-c|\left|\right)=\exp (-\frac{{|x_p-x_c|}^2+{|y_p-y_c|}^2}{{\mathrm{\&sigma;}}_Q^2})---\left(9\right)$

Wherein:

(x _p, y _p) be the locus coordinate of vicinity points p;

(x _c, y _c) be the locus coordinate of current pixel point c;

σ _qfor standard deviation, be experience factor, be the optimal value obtaining by test of many times, in this concrete enforcement, be set as 1.

Adopt G _q(|| the p-c||) factor of influence to filtration combined weighted coefficient as spatial domain.

Step 4, obtains image area factor of influence.

The vicinity points p comprising in adjacent windows for current pixel point c place, p=0,1 ..., P-1, P is the number of the vicinity points that comprises in adjacent windows, calculates respectively the degree of depth intensity d of each vicinity points p in depth map _pdegree of depth intensity d with current pixel point c in depth map _csimilarity (d _p-d _c), with G _s(d _p-d _c) be the image area factor of influence that vicinity points p is corresponding, G _sfor Gaussian function.G _s(d _p-d _c) can be expressed as follows:

$G_S(d_p-d_c)=\exp (-\frac{{|d_p-d_c|}^2}{{\mathrm{\&sigma;}}_s^2})---\left(10\right)$

Wherein, σ _sfor standard deviation, be experience factor, be the optimal value obtaining by test of many times, be set as 0.1.

Adopt G _s(d _p-d _c) factor of influence as image area to filtration combined weighted coefficient.

Step 5, compute depth figure filtering result

By the local binary patterns operator of step 2 gained vicinity points

the step 3 gained spatial domain image factor and the step 4 gained image area image factor, as the vicinity points p degree of depth intensity d in depth map in constraint adjacent windows _pweighting coefficient.

Degree of depth intensity d to each vicinity points _pbe weighted summation, obtain the filtering result G of current pixel point c in depth map _c:

$G_c=\frac{1}{F_c}{\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)\&CenterDot;d_p]---\left(11\right)$

Wherein:

W is the set of all vicinity points of current pixel point c;

G _s(d _p-d _c) be image area factor of influence, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth map _pdegree of depth intensity d with current pixel point c in depth map _csimilarity;

G _q(|| p-c||) be spatial domain factor of influence, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c;

for vicinity points p after upgrading local binary patterns operator, g _pfor the brightness value of vicinity points p in coloured image corresponding to depth image, g _cfor the brightness value of current pixel point c in coloured image corresponding to depth image;

$F_c={\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)],$ For weighting coefficient summation, for standardization.

H.264/MVC the depth map filtering method of above-mentioned steps 1～5 is joined in coding framework, as loop filtering, by removing the noise and increasing deep video code efficiency of depth map, further improved based on virtual visual point image subjective and objective quality corresponding to the colored plus depth form three-dimensional video-frequency of many viewpoints.

The depth map filtering method instructing based on local binary patterns operator (LBP) that the present invention proposes, the PSNR(Y-PSNR that makes new synthetic virtual view quality and depth map encoding efficiency on average obtain 0.45dB and 0.66dB) lifting, is also greatly improved on subjective vision with stylish synthetic virtual visual point image.

Filtering method of the present invention is inserted into the software platform of JMVC6.0 as loop filtering ^[9].For deep video encoding setting Q _pvalue 22,27,32 and 37, color video encoding Q _pbe set to 22.Deep video and color video 100 frames of all encoding, arrange reference frame between two time domain reference frames and two viewpoints." Ballet " that the resolution that cycle tests provides for Microsoft is 1024 * 768 and " Breakdancer ".Viewpoint 0 and viewpoint 2 are chosen as reference view, and viewpoint 1 is synthetic virtual view, and virtual view adopts VSRS2.3 software synthetic.Contrast is the depth filtering algorithm [4] of optimum at present, and the PSNR that the inventive method makes depth map encoding efficiency on average obtain 0.66dB promotes, and new synthetic virtual view objective quality is on average promoted to 0.45dB.

In literary composition, relate to following list of references:

[1]P.Merkle,J.B.Singla,K.Muller,and T.Wiegand,“Correlation histogram analysis of depth-enhanced3d video coding,”in Image Processing(ICIP),201017th IEEE International Conference on,sept.2010,pp.2605–2608.

[2]Y.Morvan P.Merkle and A.Smolic,“The effects of multi-view depth video compression on multi-view rendering,”Signal Processing:Image Communication,vol.24,no.1-2,pp.73C88,January2009.

[3]Kwan-Jung Oh,A.Vetro,and Yo-Sung Ho,“Depth coding using a boundary reconstruction filter for3-d video systems,”Circuits and Systems for Video Technology,IEEE Transactions on,vol.21,no.3,pp.350–359,march2011.

[4]Shujie Liu,PoLin Lai,Dong Tian,and Chang Wen Chen,“New depth coding techniques with utilization of corresponding video,”Broadcasting,IEEE Transactions on,vol.57,no.2,pp.551–561,june2011.

[5]T.Ojala,M.Pietikainen,and T.Maenpaa,“Multi-resolution gray-scale and rotation invariant texture classification with local binary patterns,”Pattern Analysis and Machine Intelligence,IEEE Transactions on,vol.24,no.7,pp.971–987,jul2002.

[6]B.Widrow,I.Kollar,and Ming-Chang Liu,“Statistical theory of quantization,”Instrumentation and Measurement,IEEE Transactions on,vol.45,no.2,pp.353–361,apr1996.

[7]C.Dorea P.Yin P.Lai,A.Ortega and C.Gomila,“Statistical theory of quantization,”in Proc.of Visual Communic.and Image Proc.,VCIP’09.San Jose,CA,USA,,Jan.2009.

[8]Timo Ojala,Kimmo Valkealahti,Erkki Oja,and MattiPietikinen,“Texture discrimination with multidimensional distributions of signed gray-level differences,”Pattern Recognition,vol.34,no.3,pp.727–739,2001.

Claims (10)

1. the depth map filtering method instructing based on local binary patterns operator, is characterized in that, comprises step:

Step 1, for coloured image corresponding to depth image, obtains the local binary patterns operator of the vicinity points of current pixel point;

Step 2, the local binary patterns operator based on vicinity points is differentiated and is obtained belonging to current pixel point the vicinity points of same object, and upgrades the local binary patterns operator of vicinity points;

Step 3, the geometric distance acquisition spatial domain factor of influence according to the vicinity points of current pixel point to current pixel point;

Step 4, according to the vicinity points of current pixel point and current pixel point, the similarity of degree of depth intensity in depth image obtains image area factor of influence;

Step 5, the constraint factor that vicinity points local binary patterns operator, spatial domain factor of influence and the image area factor of influence of take after upgrading is filtration combined weighted coefficient obtains the filtering result of current pixel point.

2. the depth map filtering method instructing based on local binary patterns operator as claimed in claim 1, is characterized in that:

Above-mentioned steps 2 further comprises sub-step:

2-1 is divided into two parts according to the local binary patterns operator of vicinity points by vicinity points;

2-2 judges according to the logical reach of above-mentioned two parts vicinity points and current pixel point whether vicinity points belongs to same object with current pixel point;

2-3 is updated to 1 by the local binary patterns operator that belongs to the vicinity points of same object with current pixel point, and the local binary patterns operator that does not belong to the vicinity points of same object with current pixel point is updated to 0.

3. the depth map filtering method instructing based on local binary patterns operator as claimed in claim 1, is characterized in that:

Spatial domain factor of influence described in step 3 is G _q(|| p-c||), wherein, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c.

4. the depth map filtering method instructing based on local binary patterns operator as claimed in claim 1, is characterized in that:

Image area factor of influence described in step 4 is G _s(d _p-d _c), wherein, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity.

5. the depth map filtering method instructing based on local binary patterns operator as claimed in claim 1, is characterized in that:

The depth map filtering result G of the current pixel point that step 5 obtains _cfor:

$G_c=\frac{1}{F_c}{\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)\&CenterDot;d_p]$

Wherein:

W is the set of all vicinity points of current pixel point c;

P is the vicinity points of current pixel point c;

G _s(d _p-d _c) be image area factor of influence, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity;

G _q(|| p-c||) be spatial domain factor of influence, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c;

for the local binary patterns operator after vicinity points p renewal, g _pand g _cbe respectively vicinity points p and the current pixel point c brightness value in coloured image corresponding to depth image;

$F_c={\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)],$ For weighting coefficient summation.

6. the depth map filtering system instructing based on local binary patterns operator, is characterized in that, comprising:

Local binary patterns operator acquisition module, is used for, for coloured image corresponding to depth image, obtaining the local binary patterns operator of the vicinity points of current pixel point;

Vicinity points judge module, the local binary patterns operator being used for based on vicinity points is differentiated and is obtained belonging to current pixel point the vicinity points of same object, and upgrades the local binary patterns operator of vicinity points;

Spatial domain factor of influence acquisition module, is used for to the geometric distance of current pixel point, obtaining spatial domain factor of influence according to the vicinity points of current pixel point;

Image area factor of influence acquisition module, is used for the similarity of according to the vicinity points of current pixel point and current pixel point degree of depth intensity in depth image to obtain image area factor of influence;

Filtration module, is used for the filtering result that constraint factor that vicinity points local binary patterns operator, spatial domain factor of influence and image area factor of influence after upgrading are filtration combined weighted coefficient obtains current pixel point.

7. the depth map filtering system instructing based on local binary patterns operator as claimed in claim 6, is characterized in that:

Described vicinity points judge module further comprises submodule:

Vicinity points is divided module, is used for, according to the local binary patterns operator of vicinity points, vicinity points is divided into two parts;

Judge module, is used for judging according to the logical reach of above-mentioned two parts vicinity points and current pixel point whether vicinity points belongs to same object with current pixel point;

Local binary patterns operator update module, be used for the local binary patterns operator that belongs to the vicinity points of same object with current pixel point to be updated to 1, the local binary patterns operator that does not belong to the vicinity points of same object with current pixel point is updated to 0.

8. the depth map filtering system instructing based on local binary patterns operator as claimed in claim 6, is characterized in that:

The spatial domain factor of influence that described spatial domain factor of influence acquisition module obtains is G _q(|| p-c||), wherein, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c.

9. the depth map filtering system instructing based on local binary patterns operator as claimed in claim 6, is characterized in that:

Described image area factor of influence is G _s(d _p-d _c), wherein, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth image _pdegree of depth intensity d with current pixel point c in depth image _csimilarity.

10. the depth map filtering system instructing based on local binary patterns operator as claimed in claim 6, is characterized in that:

The filtering result G of the current pixel point that described filtration module obtains _cfor:

$G_c=\frac{1}{F_c}{\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)\&CenterDot;d_p]$

Wherein:

W is the set of all vicinity points of current pixel point c;

G _s(d _p-d _c) be image area factor of influence, G _sfor Gaussian function, d _p-d _cfor the vicinity points p of the current pixel point c degree of depth intensity d in depth map _pdegree of depth intensity d with current pixel point c in depth map _csimilarity;

G _q(|| p-c||) be spatial domain factor of influence, G _qfor Gaussian function, || p-c|| is that the vicinity points p of current pixel point c is to the geometric distance of current pixel point c;

for vicinity points p after upgrading local binary patterns operator, g _pfor the brightness value of vicinity points p in coloured image corresponding to depth image, g _cfor the brightness value of current pixel point c in coloured image corresponding to depth image;

$F_c={\mathrm{\&Sigma;}}_{p\&Element;w}[G_S(d_p-d_c)\&CenterDot;G_Q\left(\right||p-c|\left|\right)\&CenterDot;{\mathrm{LBP}}_{P.R}^{\mathrm{uni}}(g_p-g_c)],$ For weighting coefficient summation.