CN107705295A - A kind of image difference detection method based on steadiness factor method - Google Patents
A kind of image difference detection method based on steadiness factor method Download PDFInfo
- Publication number
- CN107705295A CN107705295A CN201710828732.3A CN201710828732A CN107705295A CN 107705295 A CN107705295 A CN 107705295A CN 201710828732 A CN201710828732 A CN 201710828732A CN 107705295 A CN107705295 A CN 107705295A
- Authority
- CN
- China
- Prior art keywords
- image
- matrix
- diff area
- registration
- length
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of image difference detection method for being based on steadiness factor method (RPCA), mainly solves the test problems of the change of divergence in image or video data.Implementation step is:1. obtain different time, different visual angles, the image of Same Scene;2. pair image carries out geometrical registration;3. column vector processing is carried out to the view data after registration respectively, and by all column vector composite matrix X;4. being decomposed using RPCA to matrix X, the corresponding sparse matrix S for including discrepancy information is drawn0;5. according to sparse matrix S0, the filling region of each width image difference dissimilarity is obtained, and miscellaneous noise is filtered out;6. according to filling region result, the centre coordinate and length and width size in each width image difference region are drawn, the diff area is marked in image after registration.Compared with prior art, the present invention has the advantages of non-ideal disturbances various to visual angle, illumination, noise etc. are more sane, available for the diff area detection under multidate unmanned aerial vehicle platform.
Description
Technical field
The invention belongs to technical field of image processing, more particularly to a kind of image difference based on steadiness factor method
Detection method.
Background technology
Image difference detecting system based on unmanned aerial vehicle platform has important use value and wide on military and civilian
Wealthy application prospect, it is the popular domain of the unmanned Study on Monitoring Technology of a new generation, it to various visual angles, multidate unmanned plane by putting down
The picture or video information that platform obtains carry out Difference test, it can be found that the change of potential area-of-interest and the movement of target,
The feature extraction and description to region are realized, and strong foundation is provided for further decision-making.
The differentiation that existing difference detecting method mainly realizes image with gradation of image figure or color change etc. describes, such
The image that method gathers for fixed viewpoint has higher detection performance, can preferably realize discovery and knowledge to diff area
Not.However, in practical application, development and popularization especially using unmanned plane as the unmanned monitoring platform of new generation of representative, it is desirable to
Under conditions of different time different visual angles are not even with monitoring device, the monitoring to diff area is realized.Therefore, if still adopted
To the detection method based on gradation of image figure, the non-ideal factor such as its visual angle difference, illumination condition can severe exacerbation detection
Performance so that Difference test result has a large amount of false-alarms, is not used to judge real difference region.
In order to preferably adapt to the unmanned platform monitoring requirements of a new generation, what can especially be met multiple image while detect should
With demand, now traditional method based on the detection of two width figure differentiation be no longer applicable, it is necessary to using adaptive background structure and
Information extraction technology, by the analysis to multiple image background characteristics and Optimization Solution, the background of certain correlation properties will be met
Image is uniformly identified and extracted, and combines image processing method, realizes the self-adapting detecting to diff area.
The content of the invention
It is an object of the invention to propose a kind of image difference detection method based on steadiness factor method.The present invention
Using the background correlation between multiple image, reduce because color caused by the non-ideal factors such as time, visual angle and illumination disturbs,
The characteristic of adaptive extraction diff area, improves identification robustness, reduces false-alarm point, realizes the nothing under multidate, various visual angles
People's platform adaptive Difference test.
To realize above-mentioned technical purpose, the present invention, which adopts the following technical scheme that, to be achieved.
A kind of image difference detection method based on steadiness factor method comprises the following steps:
S1:Using optical camera obtain different time, different visual angles, Same Scene multiple image, obtain image
Number is M;
S2:Geometrical registration is carried out to M width image, the image data table after registration is obtained and is shown as matrix X1To XMOr Xi, i takes
Value 1 is to M, XiRepresent the matrix of the view data after the i-th width image registration;
S3:Respectively to matrix X1To XMColumn vector processing is carried out, obtains column vector η1To ηM, then by column vector η1To ηM
It is combined into matrix X, X=[η1,...,ηM];
S4:The matrix X is decomposed using steadiness factor method, draw corresponding to comprising discrepancy information
Sparse matrix S0;
S5:By the sparse matrix S0Each row pull into again and XiDimension identical matrix, S0The i-th row pull into
Matrix is designated asFor i values 1 to M, the matrix pulled into is the discrepancy information of each width image;
S6:It is rightCarry out miscellaneous noise to filter out, the discrepancy that the same area is connected merges into one
Diff area, and each diff area after filtering out is marked successively, it is designated asExtremelyOrL values 1 are to Li,
LiFor the number of the diff area obtained in the i-th width image,Represent the the-th block diff area in the i-th width image;
S7:Every piece of diff area is calculated, draws diff area(i=1 ..., M, l=1 ..., Li) corresponding
Centre coordinate (mi_l,ni_l), length lengthi_lWith height heighti_l, and mark the diff area information of the diff area
For xi_out_l=[mi_l,ni_l,lengthi_l,heighti_l] (i_out_l=i_out_1 ..., i_out_Li, i=1 ...,
M);
S8:Utilize xi_out_l(i_out_l=i_out_1 ..., i_out_Li, i=1 ..., M) and obtain the i-th width image
Diff area information matrix(i=1 ..., M), the diff area is believed
Cease the image X of diff area mark after registration corresponding to matrixiIn.
In certain embodiments, in the step S2, geometrical registration is carried out to M width image, obtains the image after registration
Data are expressed as matrix X1To XMOr Xi, i values 1 to M, XiThe matrix of the view data after the i-th width image registration is represented, including
Following steps:Geometrical registration is carried out to the M width image, on the basis of piece image, calculated using SIFT operators or its improvement
Method carries out geometrical registration successively to other images, by other described images transform to the visual angle consistent with the piece image and
Scene size, registration after view data be designated as matrix X1To XMOr Xi, i values 1 to M, XiAfter representing the i-th width image registration
The matrix of view data.
In certain embodiments, in the step S4, the matrix X is divided using steadiness factor method
Solution, draw the corresponding sparse matrix S for including discrepancy information0, specifically include following steps:
Matrix X decomposition model is:X=L0+S0+N0, wherein, L0、S0And N0For three submatrixs after decomposition, L0It is low
Order matrix, N0Represent residual noise, S0For sparse matrix, matrix S0It is identical with matrix X dimension;Extracted according to optimal model
Go out L0、S0And N0:
min||L0||*+μ||S0||1
s.t.||X-L0-S0||F<δ
Wherein, | | | |11- norms are sought in expression, | | | |FF- norms are sought in expression, | | | |*Nuclear norm is sought in expression, and δ is to set
Fixed constant, μ represent weight factor and μ>0, min represent to minimize, and s.t. is that subject to write a Chinese character in simplified form expression " being constrained to ", whole
The implication of individual equation is to meet that constraints is | | X-L0-S0||F<Under conditions of δ so that object function | | L0||*+μ||S0|
|1Value it is minimum.
In certain embodiments, in the step S5, by the sparse matrix S0Each row pull into again and XiDimension
Identical matrix, S0The i-th matrix for pulling into of row be designated asI values 1 to M, for the discrepancy of each width image believe by the matrix pulled into
Breath, the step specifically include following steps:For the sparse matrix S0I-th row S0(i), by S0(i) according to from top to bottom
Order divide b column vector, b is image XiThe length of corresponding transverse axis, the element number of the b column vector is a, a
For image XiThe length of the corresponding longitudinal axis, the b column vector is combined as corresponding matrix by stripe sequenceWherein, i takes
Value 1 is to M.
In certain embodiments, it is right in the step S6Carry out miscellaneous noise to filter out, the same area is connected
Discrepancy merges into a diff area, and each diff area after filtering out is marked successively, is designated asExtremely
OrThe step specifically includes following steps:
Image prior thresholding T is set, the criterion of the threshold sets is:T is mesh interested to be detected in M width images after registration
Mark the minimum value of resolution cell number corresponding to the size of type;
(i=1 ..., M) the individual matrix pulled into for i-thFollowing three step is performed successively:
S1:It is rightCandy operator edge detections are carried out, closed curve confirmation is carried out to edge after detection;
S2:Edge for belonging to closed curve, processing is filled to it, and calculates its pixel number, works as filling
When the pixel number in the closed curve is more than or equal to thresholding T afterwards, it is one to define all differences point in the closed curve
Diff area;
S3:It is less than thresholding T for non-closed curve or closed curve but pixel number, by its marginal information or difference section
Domain information zero setting, that is, filter out, and is designated as successively for the remaining each diff area retained after filtering outExtremelyOrRepresent the the-th block diff area in the i-th width image, l values 1 to Li, LiFor the difference obtained in the i-th width image
The number in region.
In certain embodiments, in the step S7, every piece of diff area is calculated, draws diff area
Corresponding centre coordinate (mi_l,ni_l), length lengthi_lWith height heighti_l, and mark the diff area of the diff area
Information is xi_out_l=[mi_l,ni_l,lengthi_l,heighti_l], the step specifically includes following steps:
Every piece of diff area is calculated, for diff area(i=1 ..., M, l=1 ..., Li), it is somebody's turn to do
The transverse axis maximum and minimum value of diff area are respectively bi_l_maxAnd bi_l_min, the longitudinal axis maximum and minimum of the diff area
Value is respectively ai_l_maxAnd ai_l_min;
As defined as follows, the diff area is calculatedCentre coordinate (mi_l,ni_l), length lengthi_lAnd height
heighti_l:
mi_l=(bi_l_max+bi_l_min)/2
ni_l=(ai_l_max+ai_l_min)/2
lengthi_l=bi_l_max-bi_l_min
heighti_l=ai_l_max-ai_l_min
Mark the diff areaDiff area information be xi_out_l:
xi_out_l=[mi_l,ni_l,lengthi_l,heighti_l]
Beneficial effects of the present invention are:1) method that prior art is mainly subtracted each other using image pixel, in the ideal case
There can be preferable diff area Detection results, but in actual applications, difference to the same area IMAQ visual angle,
Selection of reference frame difference between the difference and multiple image of the non-ideal conditions such as illumination, can all cause detection performance drastically under
Drop, produces a large amount of clutter false-alarms, is not used to the judgement of successive image diff area information.Background is miscellaneous between the present invention utilizes image
The strong correlation that ripple is distributed in structure and color information, the clutter background of multiple image is extracted while can adaptively, is dropped
It is low due to false-alarm caused by non-ideal error.2) conventional method is based primarily upon the image difference detection of two images, if needing several
Image comparison with this, it is necessary to be traveled through, and because the selection of benchmark image is different, cause result difference larger.The present invention utilizes low
Order Matrix Properties, under conditions of computing capability license, great amount of images can be handled simultaneously, overall ask only is needed once to a problem
Solution preocess, the problems such as avoiding data traversal, while benchmark image need not be selected, but obtain all images and share part conduct
Background image, diff area obtained from entering have universality relative to other all images.3) present invention in implementation process not
Increase any hardware constraint, while there is stronger usage range, in change platform, various visual angles, multidate observation condition
And unmanned intelligence system is first-class can meet to apply.
Brief description of the drawings
Fig. 1 is a kind of flow chart of image difference detection method based on steadiness factor method of the present invention;
Fig. 2 a are the width scene image A that unmanned aerial vehicle platform obtains;
Fig. 2 b are the Same Scene image B that unmanned aerial vehicle platform obtains after image A is obtained more days;
Fig. 3 is image A and image B Characteristic points match result;
Fig. 4 a are the image after image A registrations;
Fig. 4 b are the image after registrations of the image B on the basis of image A;
Fig. 5 is the image difference testing result that PCA of the present invention obtains;
Fig. 6 is the edge detection results of image difference dissimilarity;
Fig. 7 is the filling result after rim detection to enclosed region;
Fig. 8 a are the result after being filtered out to the progress non-occlusion region removal of filling result and miscellaneous noise;
Fig. 8 b are to mark Fig. 8 a diff area result in the image A after registration;
Fig. 9 a are the diff area testing result obtained using conventional method;
Fig. 9 b are to mark Fig. 9 a diff area result in the image A after registration.
Embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Be easy to describe, illustrate only in accompanying drawing to about the related part of invention.
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combination.Describe the application in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Reference picture 1, for a kind of flow chart of image difference detection method based on steadiness factor method of the present invention.
The image difference detection method based on steadiness factor method comprises the following steps:
Step 101, using optical camera obtain different time, different visual angles, Same Scene multiple image, obtain image
Number be M.
The multiple image of Same Scene is obtained under different time, different visual angles using optical camera, obtains of image
Number is M, and the image information of each image is expressed as matrix Xorg_1To Xorg_MOr Xorg_i, i takes 1 to M, Xorg_iRepresent optics
The image information of image acquired in camera ith.
Wherein, the image information X of each imageorg_iThe data format of (i=1 ..., M) for a three-dimensional array a × b ×
C, wherein the first dimension a is the geometric longitudinal axis length of image, the second dimension b is the geometry transverse axis length of image, and third dimension c is image face
Color information.
Step 102, geometrical registration is carried out to M width image, obtains the image data table after registration and be shown as matrix Xi。
To above-mentioned M width image Xorg_1To Xorg_MRegistration is carried out, the characteristic point of image is chosen first, with piece image
Xorg_1On the basis of, other each width images are calculated respectively to the projection square of piece image using SIFT operators or its innovatory algorithm
Battle array, and corresponding geometric transformation is done, obtain the data after registration and be expressed as matrix X1To XMOr Xi, XiRepresent the i-th width image registration
The matrix of view data afterwards, i take 1 to M.
The matrix X of view data after registrationiWith it is registering before view data matrix Xorg_iDimension it is identical, its data lattice
Formula remains as three-dimensional array a × b × c, and the now lower geometrical relationship and yardstick for reflecting object of respective coordinates between each width image
Information is consistent, i.e., visual angle is consistent with scene size.But passing through geometrical registration, image is projected to piece image
Xorg_1Visual angle and picture size structure on, now part original digital image data is lost because geometry rotates, meanwhile, after registering
Part image data is zero.
In some optional implementations of the present embodiment, geometrical registration is carried out to above-mentioned M width image, can be chosen
Any piece image in M width images is stated as benchmark, other images are carried out successively using SIFT operators or its innovatory algorithm
Geometrical registration, other above-mentioned images are transformed into the visual angle consistent with selected benchmark image and scene size, after registering
View data is designated as matrix X1To XM。
Step 103, respectively to matrix X1To XMColumn vector processing is carried out, obtains column vector η1To ηM, then by column vector
η1To ηMIt is combined into matrix X, X=[η1,...,ηM]。
Matrix XiThe length of (i=1 ..., M) in longitudinal axis dimension is a, matrix XiIt is b in the length of transverse axis dimension;Namely
Say, matrix XiLine number be a, columns b.To matrix XiColumn vectorization processing is carried out to comprise the following steps:By matrix Xi(i=
1 ..., M) each row extract, then according to being respectively listed in matrix XiIn order, by matrix XiEach row be combined to one
In row, column vector η is formedi(i=1 ..., M);It is readily apparent that the matrix X of a × b dimensionsiAfter being handled through column vectorization, turn
The column vector η that ab × 1 is tieed up is changed intoi, so matrix X=[η1,...,ηM] dimension be ab × M.
Understand that matrix X is the matrix for including all image informations by forming for matrix X.If property in former scene
Body does not change in different observed images, i.e. X1≈X2…≈XM, wherein, being approximately equal to the appearance reason of symbol is:Nothing
The image information deviation caused by difference such as effect of jitter, influence of noise and the illumination of people's machine platform, blowing, visual angle, then, this
When matrix X be low-rank matrix that an approximate order is 1;If former scene changes between different images, then X can consider
It is to be made up of two parts, one is the low-rank matrix obtained by static scene, and another is the sparse square for including region of variation
Battle array (has openness) for region of variation relative scene.
Step 104, matrix X is decomposed using steadiness factor method, draw corresponding to include discrepancy information
Sparse matrix S0。
Steadiness factor method (RPCA, robust principal component analysis) is used for several
Image is changed the separation in region and non-changing region.In steadiness factor method, matrix X decomposition model is:
X=L0+S0+N0
Wherein, L0、S0And N0For three submatrixs after decomposition, S0For sparse matrix, L0It is low-rank matrix, N0Represent remaining
Noise.
Then, L is extracted according to optimal model0、S0And N0:
min||L0||*+μ||S0||1
s.t.||X-L0-S0||F<δ
Wherein, | | | |11- norms are sought in expression, | | | |FF- norms are sought in expression, | | | |*Nuclear norm is sought in expression, specifically
Ground says, | | L0||*Represent L0In all singular values sum, δ is the constant of setting, and μ represents weight factor and μ>0.In above formula, min tables
Show minimum, s.t. is that subject to write a Chinese character in simplified form expressions " being constrained to ", and the implication of whole equation is is meeting that constraints is |
|X-L0-S0||F<Under conditions of δ so that object function | | L0||*+μ||S0||1Value it is minimum.
Step 105, by the sparse matrix S0Each row pull into again and XiDimension identical matrix, S0I-th row draw
Into matrix be designated asFor i values 1 to M, the matrix pulled into is the discrepancy information of each width image.
By analysis, the S obtained in step 1040Include discrepancy information, its each row have corresponded to difference
Change point distribution situation between image, passes through the inversion opposite with step 103, i.e. S0In each row pull into again for Xi
Dimension identical matrix, that is, the discrepancy information of each width image is obtained.Specifically, sparse matrix S0I-th be classified as S0(i), will
S0(i) b column vector is divided in accordance with the order from top to bottom, and b is image XiThe length of corresponding transverse axis, the b column vector
Element number is a, and a is image XiThe length of the corresponding longitudinal axis, the b column vector is combined as by stripe sequence corresponding to
MatrixWherein, i values 1 are to M.
Step 106, it is rightCarry out miscellaneous noise to filter out, the discrepancy that the same area is connected merges into a difference section
Domain, and each diff area after filtering out is marked successively, it is designated as
Image prior thresholding T is set, the criterion of the threshold sets is:T is mesh interested to be detected in M width images after registration
Mark the minimum value of resolution cell number corresponding to the size of type.
(i=1 ..., M) the individual matrix pulled into for i-thFollowing three step is performed successively:
S1:It is rightCandy operator edge detections are carried out, obtain the marginal information of discrepancy, are believed according to the edge of discrepancy
Breath carries out closed curve confirmation to edge.Because discrepancy caused by feature changes often has certain geometry and can gather
Class is enclosed region, therefore can carry out edge extraction to it, and non-closed curve is filtered out.
S2:Edge for belonging to closed curve, processing is filled to it, and calculates its pixel number (area),
The image difference dissimilarity caused by being disturbed compared to clutter, diff area caused by feature changes is larger, therefore, should after filling
When pixel number (area) in closed curve is more than or equal to thresholding T, defines the point of all differences in the closed curve and belong to
The curve inner region of one diff area, the i.e. closed curve is a diff area.
S3:It is less than thresholding T for non-closed curve or closed curve but pixel number, by above-mentioned non-closed curve or picture
The marginal information of closed curve of the vegetarian refreshments number (area) less than thresholding T or diff area information zero setting, that is, filter out, for filtering out
The remaining each diff area retained afterwards is designated as successivelyExtremelyOrL values 1 are to Li, LiFor in the i-th width image
The number of obtained diff area,The the-th block diff area in the i-th width image is represented, i takes 1 to M.
Step 107, every piece of diff area is calculated, draws diff areaCorresponding centre coordinate (mi_l,
ni_l), length lengthi_lWith height heighti_l, and it is x to mark the diff area information of the diff areai_out_l=[mi_l,
ni_l,lengthi_l,heighti_l]。
Every piece of diff area is calculated, for diff area(i=1 ..., M, l=1 ..., Li), it is somebody's turn to do
The transverse axis maximum and minimum value of diff area are respectively bi_l_maxAnd bi_l_min, the longitudinal axis maximum and minimum of the diff area
Value is respectively ai_l_maxAnd ai_l_min。
As defined as follows, above-mentioned diff area is calculatedCentre coordinate (mi_l,ni_l), length lengthi_lAnd height
heighti_l:
mi_l=(bi_l_max+bi_l_min)/2
ni_l=(ai_l_max+ai_l_min)/2
lengthi_l=bi_l_max-bi_l_min
heighti_l=ai_l_max-ai_l_min
Mark above-mentioned diff areaDiff area information be xi_out_l:xi_out_l=[mi_l,ni_l,lengthi_l,
heighti_l]。
Step 108, x is utilizedi_out_lObtain the diff area information matrix of the i-th width imageDiff area corresponding to the diff area information matrix is labeled in and matched somebody with somebody
Image X after standardiIn.
Utilize xi_out_lObtain the diff area information matrix of the i-th width imageIts data dimension is Li× 4, according to the diff area information matrix,
And then diff area corresponding to the diff area information matrix is labeled in the image X after first wife's standardiIn.The difference section of mark
Domain is the difference section of each image and background image, and the shared part that background image refers to above-mentioned M images is (static in scene
Background).
It can be seen from above-mentioned analysis:Present invention mainly solves multiple image in different visual angles, different phases and difference
Diff area test problems under the non-ideal conditions such as illumination.The present invention is by the data vector of multiple image, and by several figures
As the data of vectorization put together form a new matrix, due in scene static background in multiple image have it is stronger
Color and structural dependence, it can be considered that its form part a low-rank matrix is shown as in new matrix, and
The part that diff area is formed is a sparse matrix, using such characteristic, using sane principal component analytical method logarithm
According to being handled, the isolated sparse matrix for including different information, finally realize to the effective of multiple image the change of divergence
Detection.
Advantages of the present invention can be illustrated by following measured data:
Same Scene is observed using unmanned aerial vehicle platform, the front and rear image obtained twice (is herein as shown in Figure 2 table
State intuitively, image number is 2, and the inventive method can be equally used in multiple image contrast), wherein, Fig. 2 a and Fig. 2 b are obtained
Time interval be two days, collecting device used is completely the same.It can be seen that exist between two images certain visual angle difference and
Notable difference be present in the illumination condition of image.Comparison diagram 2a and Fig. 2 b, it can be seen that many places feature changes in figure be present, for example,
Get on the car on road.
In order to realize the differentiation detection to image, it is necessary to carry out registration to image, generally use is based on SIFT operators
And its innovatory algorithm carries out Image Feature Point Matching, as shown in figure 3, corresponding to the obvious target of the feature such as house, automobile
Feature Points Matching better performances.After the transformed matrix between image is obtained using characteristic point, projective transformation is carried out to image.Herein
Registration is to carry out projective transformation on the basis of image A, therefore to image B, and the image after change is as shown in Figure 4 b.Fig. 4 a are image
Image after A registrations, because image A registrations are on the basis of the figure itself, so without projection operation, Fig. 4 b are image B with spy
Projection transform is carried out on the basis of sign point registration, it can be seen that the image after image B projection transforms is with image A in geometrical relationship
It is completely the same, but due to rotating and changing, there is part image data missing in upper left and the upper right corner, while in original digital image data
Upper left, upper right and lower right corner sub-image data are lost, in order that the image B after image A and conversion is completely registering, to image A
Same loss of data processing is done, as shown in Fig. 4 a upper lefts and the upper right corner.
Using the method for the present invention, i.e., multiple image is handled using steadiness factor method, included
The sparse matrix of diff area, the first row of sparse matrix is taken out, and the column vector is become again and turned to and original image A dimensions
Consistent image, its image result is obtained as shown in figure 5, the figure i.e. Difference test result.By by A pairs of the figure and original image
Than, it can be seen that the region wherein included is mainly made up of image A and image B diff area, and this point demonstrates this
The validity of inventive method.
In order to further improve diff area detection performance, image is handled, detects obtained difference to it first
Region carries out edge extraction, as shown in Figure 6.Secondly, often there are necessarily several using discrepancy caused by feature changes
What structure and the characteristics of being closed curve can be clustered as enclosed region, the marginal information of enclosed region, pixel number is more than
Edge equal to the closed curve of 150 pixels is filled, as shown in Figure 7.Finally, its non-closed boundary curve is carried out
The diff area for filtering out and pixel number in closed curve being less than to 150 pixels is filtered out that (threshold sets are herein
150, its numerical values recited is the number of the pixel cell shared by 1/3rd of a typical dolly, therefore can realize to automobile amount
The detection and extraction of the change of divergence of level target), its image after filtering out is as shown in Figure 8 a.Preferably to be carried out to testing result
Checking, the testing result in Fig. 8 a is marked again in image A after registration, as shown in the oval solid line mark in Fig. 8 b,
As can be seen that its diff area is all detected and marked.Further to realize contrast, Fig. 9 a are to use traditional frame-to-frame differences method
Obtained diff area testing result, Fig. 9 b are to mark Fig. 9 a diff area result in the image A after registration, wherein,
The result of mistake is outlined with chain line in Fig. 9 b, it can be seen that obvious image missing inspection and false retrieval result be present, this is mainly
The disturbance of color of image difference causes caused by the difference of unmanned aerial vehicle platform visual angle.
In summary, static scene is in the strong correlation of color and geometry, reduction between the present invention utilizes multiple image
Due to disturbance caused by the factors such as visual angle, phase, platform, acquisition condition, missing inspection and false retrieval are reduced, improves objective self-adapting detection
Performance, improve sane diff area detection performance.
Obviously, those skilled in the art can carry out various changes and modification without departing from the present invention to patent of the present invention
Spirit and scope.So if these modifications and variations of the invention belong to the model of the claims in the present invention and its equivalent technologies
Within enclosing, then the present invention is also intended to comprising including these changes and modification.
Claims (6)
1. a kind of image difference detection method based on steadiness factor method, it is characterised in that methods described includes following
Step:
S1:Using optical camera obtain different time, different visual angles, Same Scene multiple image, the number for obtaining image is
M;
S2:Geometrical registration is carried out to M width image, the image data table after registration is obtained and is shown as matrix X1To XMOr Xi, i values 1 to
M, XiRepresent the matrix of the view data after the i-th width image registration;
S3:Respectively to matrix X1To XMColumn vector processing is carried out, obtains column vector η1To ηM, then by column vector η1To ηMCombination
Into matrix X, X=[η1,...,ηM];
S4:The matrix X is decomposed using steadiness factor method, drawn corresponding sparse comprising discrepancy information
Matrix S0;
S5:By the sparse matrix S0Each row pull into again and XiDimension identical matrix, S0The i-th matrix for pulling into of row
It is designated asFor i values 1 to M, the matrix pulled into is the discrepancy information of each width image;
S6:It is rightCarry out miscellaneous noise to filter out, the discrepancy that the same area is connected merges into a difference
Region, and each diff area after filtering out is marked successively, it is designated asExtremelyOrL values 1 are to Li, LiFor
The number of the diff area obtained in i-th width image,Represent the the-th block diff area in the i-th width image;
S7:Every piece of diff area is calculated, draws diff areaIn corresponding
Heart coordinate (mi_l,ni_l), length lengthi_lWith height heighti_l, and mark the diff area information of the diff area to be
xi_out_l=[mi_l,ni_l,lengthi_l,heighti_l] (i_out_l=i_out_1 ..., i_out_Li, i=1 ..., M);
S8:Utilize xi_out_l(i_out_l=i_out_1 ..., i_out_Li, i=1 ..., M) and obtain the difference of the i-th width image
Area information matrixBy the diff area information square
The image X of diff area mark after registration corresponding to battle arrayiIn.
A kind of 2. image difference detection method based on steadiness factor method as claimed in claim 1, it is characterised in that
In the step S2, geometrical registration is carried out to M width image, the image data table after registration is obtained and is shown as matrix X1To XMOr Xi,
I values 1 are to M, XiThe matrix of the view data after the i-th width image registration is represented, is comprised the following steps:
Geometrical registration is carried out to the M width image, on the basis of piece image, using SIFT operators or its innovatory algorithm to it
He carries out geometrical registration by image successively, and other described images are transformed into the visual angle consistent with the piece image and scene is big
Small, the view data after registration is designated as matrix X1To XMOr Xi, i values 1 to M, XiRepresent the picture number after the i-th width image registration
According to matrix.
A kind of 3. image difference detection method based on steadiness factor method as claimed in claim 1, it is characterised in that
In the step S4, the matrix X is decomposed using steadiness factor method, drawn corresponding comprising discrepancy letter
The sparse matrix S of breath0, specifically include following steps:
Matrix X decomposition model is:X=L0+S0+N0, wherein, L0、S0And N0For three submatrixs after decomposition, L0It is low-rank square
Battle array, N0Represent residual noise, S0For sparse matrix;
L is extracted according to optimal model0、S0And N0:
min||L0||*+μ||S0||1
s.t.||X-L0-S0||F<δ
Wherein, | | | |11- norms are sought in expression, | | | |FF- norms are sought in expression, | | | |*Nuclear norm is sought in expression, and δ is setting
Constant, μ represent weight factor and μ>0, min represents to minimize, and s.t. writes a Chinese character in simplified form expression " being constrained to " for subject to's, whole side
The implication of journey is to meet that constraints is | | X-L0-S0||F<Under conditions of δ so that object function | | L0||*+μ||S0||1's
Value is minimum.
A kind of 4. image difference detection method based on steadiness factor method as claimed in claim 3, it is characterised in that
In the step S5, by the sparse matrix S0Each row pull into again and XiDimension identical matrix, S0I-th row draw
Into matrix be designated asI values 1 to M, the matrix pulled into be each width image discrepancy information, the step specifically include with
Lower step:
For the sparse matrix S0I-th row S0(i), by S0(i) b column vector is divided in accordance with the order from top to bottom, and b is
Image XiThe length of corresponding transverse axis, the element number of the b column vector is a, and a is image XiThe length of the corresponding longitudinal axis,
The b column vector is combined as corresponding matrix by stripe sequenceWherein, i values 1 are to M.
A kind of 5. image difference detection method based on steadiness factor method as claimed in claim 4, it is characterised in that
It is right in the step S6Miscellaneous noise is carried out to filter out, the discrepancy that the same area is connected merges into a diff area,
And each diff area after filtering out is marked successively, it is designated asExtremelyOrThe step specifically includes following
Step:
Image prior thresholding T is set, the criterion of the threshold sets is:T is interesting target class to be detected in M width images after registration
The minimum value of resolution cell number corresponding to the size of type;
(i=1 ..., M) the individual matrix pulled into for i-thFollowing three step is performed successively:
S1:It is rightCandy operator edge detections are carried out, closed curve confirmation is carried out to edge after detection;
S2:Edge for belonging to closed curve, processing is filled to it, and calculates its pixel number, should after filling
When pixel number in closed curve is more than or equal to thresholding T, it is a difference to define all differences point in the closed curve
Region;
S3:It is less than thresholding T for non-closed curve or closed curve but pixel number, its marginal information or diff area is believed
Zero setting is ceased, that is, filters out, is designated as successively for the remaining each diff area retained after filtering outExtremelyOr
Represent the the-th block diff area in the i-th width image, l values 1 to Li, LiFor the number of the diff area obtained in the i-th width image.
A kind of 6. image difference detection method based on steadiness factor method as claimed in claim 5, it is characterised in that
In the step s 7, every piece of diff area is calculated, draws diff areaCorresponding centre coordinate (mi_l,ni_l), it is long
Spend lengthi_lWith height heighti_l, and it is x to mark the diff area information of the diff areai_out_l=[mi_l,ni_l,
lengthi_l,heighti_l], the step specifically includes following steps:
Every piece of diff area is calculated, for diff areaObtain the difference
The transverse axis maximum and minimum value in region are respectively bi_l_maxAnd bi_l_min, the longitudinal axis maximum and minimum value point of the diff area
Wei not ai_l_maxAnd ai_l_min;
As defined as follows, the diff area is calculatedCentre coordinate (mi_l,ni_l), length lengthi_lAnd height
heighti_l:
mi_l=(bi_l_max+bi_l_min)/2
ni_l=(ai_l_max+ai_l_min)/2
lengthi_l=bi_l_max-bi_l_min
heighti_l=ai_l_max-ai_l_min
Mark the diff areaDiff area information be xi_out_l:
xi_out_l=[mi_l,ni_l,lengthi_l,heighti_l]。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710828732.3A CN107705295B (en) | 2017-09-14 | 2017-09-14 | Image difference detection method based on robust principal component analysis method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710828732.3A CN107705295B (en) | 2017-09-14 | 2017-09-14 | Image difference detection method based on robust principal component analysis method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107705295A true CN107705295A (en) | 2018-02-16 |
CN107705295B CN107705295B (en) | 2021-11-30 |
Family
ID=61172498
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710828732.3A Active CN107705295B (en) | 2017-09-14 | 2017-09-14 | Image difference detection method based on robust principal component analysis method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107705295B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109741377A (en) * | 2018-11-30 | 2019-05-10 | 四川译讯信息科技有限公司 | A kind of image difference detection method |
CN110782459A (en) * | 2019-01-08 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Image processing method and device |
CN113189543A (en) * | 2021-04-27 | 2021-07-30 | 哈尔滨工程大学 | Interference suppression method based on motion compensation robust principal component analysis |
CN113393498A (en) * | 2021-05-26 | 2021-09-14 | 上海联影医疗科技股份有限公司 | Image registration method and device, computer equipment and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101369310A (en) * | 2008-09-27 | 2009-02-18 | 北京航空航天大学 | Robust human face expression recognition method |
CN102509264A (en) * | 2011-11-01 | 2012-06-20 | 武汉大学 | Image-segmentation-based scanning image dedusting method |
CN103901416A (en) * | 2014-03-31 | 2014-07-02 | 西安电子科技大学 | Multichannel clutter suppression method based on robust principal component analysis |
US20170154413A1 (en) * | 2015-11-27 | 2017-06-01 | Toshiba Medical Systems Corporation | Dynamic image denoising using a sparse representation |
-
2017
- 2017-09-14 CN CN201710828732.3A patent/CN107705295B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101369310A (en) * | 2008-09-27 | 2009-02-18 | 北京航空航天大学 | Robust human face expression recognition method |
CN102509264A (en) * | 2011-11-01 | 2012-06-20 | 武汉大学 | Image-segmentation-based scanning image dedusting method |
CN103901416A (en) * | 2014-03-31 | 2014-07-02 | 西安电子科技大学 | Multichannel clutter suppression method based on robust principal component analysis |
US20170154413A1 (en) * | 2015-11-27 | 2017-06-01 | Toshiba Medical Systems Corporation | Dynamic image denoising using a sparse representation |
Non-Patent Citations (2)
Title |
---|
ZHANG BAOHUA 等: "Multi一focus image fusion based on sparse decomposition and background detection", 《DIGITAL SIGNAL PROCESSING》 * |
王向阳 等: "采用多任务稳健主成分分析的运动目标分割", 《应用科学学报》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109741377A (en) * | 2018-11-30 | 2019-05-10 | 四川译讯信息科技有限公司 | A kind of image difference detection method |
CN109741377B (en) * | 2018-11-30 | 2021-07-06 | 四川译讯信息科技有限公司 | Image difference detection method |
CN110782459A (en) * | 2019-01-08 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Image processing method and device |
CN113189543A (en) * | 2021-04-27 | 2021-07-30 | 哈尔滨工程大学 | Interference suppression method based on motion compensation robust principal component analysis |
CN113393498A (en) * | 2021-05-26 | 2021-09-14 | 上海联影医疗科技股份有限公司 | Image registration method and device, computer equipment and storage medium |
CN113393498B (en) * | 2021-05-26 | 2023-07-25 | 上海联影医疗科技股份有限公司 | Image registration method, device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107705295B (en) | 2021-11-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106845374B (en) | Pedestrian detection method and detection device based on deep learning | |
Zhang et al. | A categorization of multiscale-decomposition-based image fusion schemes with a performance study for a digital camera application | |
CN106875415B (en) | Continuous and stable tracking method for small and weak moving targets in dynamic background | |
CN107705295A (en) | A kind of image difference detection method based on steadiness factor method | |
CN104899866B (en) | A kind of intelligentized infrared small target detection method | |
CN107301661A (en) | High-resolution remote sensing image method for registering based on edge point feature | |
CN108876723B (en) | Method for constructing color background of gray target image | |
CN111161313B (en) | Multi-target tracking method and device in video stream | |
CN107481315A (en) | A kind of monocular vision three-dimensional environment method for reconstructing based on Harris SIFT BRIEF algorithms | |
Li et al. | Automatic registration of panoramic image sequence and mobile laser scanning data using semantic features | |
CN109446917B (en) | Vanishing point detection method based on cascading Hough transform | |
CN109685732A (en) | A kind of depth image high-precision restorative procedure captured based on boundary | |
CN111738314A (en) | Deep learning method of multi-modal image visibility detection model based on shallow fusion | |
CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
CN110097617B (en) | Image fusion method based on convolutional neural network and significance weight | |
Lian et al. | A novel method on moving-objects detection based on background subtraction and three frames differencing | |
CN113065397B (en) | Pedestrian detection method and device | |
CN104616026A (en) | Monitor scene type identification method for intelligent video monitor | |
Zhao et al. | An adaptation of CNN for small target detection in the infrared | |
CN111611918A (en) | Traffic flow data set acquisition and construction method based on aerial photography data and deep learning | |
CN112329764A (en) | Infrared dim target detection method based on TV-L1 model | |
Liu et al. | HPN-SOE: Infrared small target detection and identification algorithm based on heterogeneous parallel networks with similarity object enhancement | |
CN111089586B (en) | All-day star sensor star point extraction method based on multi-frame accumulation algorithm | |
CN104700359A (en) | Super-resolution reconstruction method of image sequence in different polar axis directions of image plane | |
Balcilar et al. | Extracting vehicle density from background estimation using Kalman filter |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |