CN110097569A - Oil tank object detection method based on color Markov Chain conspicuousness model - Google Patents
Oil tank object detection method based on color Markov Chain conspicuousness model Download PDFInfo
- Publication number
- CN110097569A CN110097569A CN201910272945.1A CN201910272945A CN110097569A CN 110097569 A CN110097569 A CN 110097569A CN 201910272945 A CN201910272945 A CN 201910272945A CN 110097569 A CN110097569 A CN 110097569A
- Authority
- CN
- China
- Prior art keywords
- color
- oil tank
- super
- markov chain
- matrix
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 50
- 239000011159 matrix material Substances 0.000 claims abstract description 49
- 238000000034 method Methods 0.000 claims abstract description 42
- 230000004927 fusion Effects 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 13
- 238000010521 absorption reaction Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 101100295776 Drosophila melanogaster onecut gene Proteins 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 abstract description 5
- 238000005286 illumination Methods 0.000 abstract description 3
- 230000009466 transformation Effects 0.000 description 17
- 230000000694 effects Effects 0.000 description 11
- 238000012546 transfer Methods 0.000 description 7
- 230000001052 transient effect Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 238000009414 blockwork Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000013100 final test Methods 0.000 description 1
- 239000002803 fossil fuel Substances 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000005648 markovian process Effects 0.000 description 1
- 239000000155 melt Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of oil tank object detection methods based on color Markov Chain conspicuousness model, using contain the markovian unsupervised conspicuousness model of color, oil depot region and background area are distinguished by generating bottom-up potential notable figure, and color combining and location matrix, shade, the influence that imaging angle detects oil tank can be effectively reduced, merge the location and shape information of oil tank finally to weaken influence of the background area for oil tank target itself.By the oil tank mesh object detection method, it can sufficiently accurately detect that the oil tank target in the remote sensing images under different illumination conditions and target size, this method have stronger robustness.
Description
Technical field
The present invention relates to digital image processing techniques and computer vision field, more particularly to a kind of to be based on face
The oil tank object detection method of color Markov Chain conspicuousness model.
Background technique
Oil tank be one storage petroleum and fossil fuel place, its shape is generally circular, and be coated with white or
Other light color of person, with reflected sunlight as far as possible, reduce the absorption of heat, and then prevent oil tank because being chronically exposed under sunlight
And it sets off an explosion.The rule of oil tank arrangement is than more random, and its edge is more fuzzy, and the gray scale on surface is also uneven.Due to
The factors such as the influence of light, image quality, structure and position and satellite and the angle on ground, oil tank may on imaging surface
There can be a degree of geometry deformation.The case where this series of complex, brings very big to the detection and identification of oil tank target
It is difficult.In recent years, it with the raising of Optical remote satellite image resolution ratio, can be detected in remote sensing images more tiny
Target.Although high-resolution remote sensing image can provide the more detailed information about target, the background of image also will therefore
Become increasingly complex, this is but also the detection of typical remote sensing target becomes more difficult.The especially inspection of the targets such as oil tank
Survey becomes more difficult, it is easy to bring false-alarm and missing inspection.Along with oil tank is a distributed object, distribution position and
Distance is not fixed, and has become a great problem of object detection field.
The detection algorithm of oil tank target specifically includes that detection method based on geometries such as Hough transformations and is based at present
The method etc. of classifier, such as: a kind of remote sensing images circle oil tank detection method of view-based access control model conspicuousness and Hough transformation, it should
Method first significantly converts remote sensing image progress MHC vision and obtains notable figure, enhances using the morphology of mathematics, is enhanced
Vision significance figure.Loop truss, which is carried out, using Hough transformation on the basis of notable figure obtains doubtful oil tank region, while root
Another testing result is obtained according to turbo pixel and like circle feature.Finally two figures are classified to obtain most using SVM
Whole object detection results.This method has certain effect for brighter oil tank, but darker for color or in complexity
Oil tank target detection effect under background is poor.Another example is: a kind of oil tank detection side based on high-resolution optical remote sensing image
Method, this method chooses oil tank candidate target region first, and is classified to candidate target region by SVM, obtains oil tank
Suspicious region;Then oil tank suspicious region is further determined whether as oil tank target, if the length-width ratio in the region close to 1 and its
There are other white pixel connection regions in surrounding a certain range, be then judged as oil tank target area, is otherwise judged as non-oil tank mesh
Mark region.The geographical location of oil tank is finally mapped to by the position in image.Since the algorithm can not be done to like circle building
More accurately judge out, so will appear a large amount of erroneous detection as a result, detection effect is poor.
Therefore, how to provide a kind of oil tank object detection method that detection effect is good is those skilled in the art's urgent need to resolve
The problem of.
Summary of the invention
In view of this, the present invention provides a kind of oil tank target detection sides based on color Markov Chain conspicuousness model
Conspicuousness model is introduced into object detection method by method using the color characteristic of oil tank, while using shape feature as aobvious
The constraint condition of work property model, has stronger robustness in oil tank target detection, can be realized to the accurate of oil tank target
Detection and positioning, detection effect are good.
To achieve the goals above, the present invention adopts the following technical scheme:
A kind of oil tank object detection method based on color Markov Chain conspicuousness model, comprising:
Gray proces and super-pixel segmentation are carried out to collected remote sensing images;
Radial extraction is carried out to the remote sensing images by gray proces, obtains candidate center location;
The weighted value for calculating each candidate center of circle obtains round feature using the weighted value in the center of circle as entirely round gray value
Distribution map;
Circle density matrix is calculated based on super-pixel segmentation result and the round characteristic profile;
The remote sensing images Jing Guo super-pixel segmentation are handled based on color Markov Chain conspicuousness model, and respectively
Extract location matrix and color matrix;
Fusion Features are carried out based on the round density matrix, the location matrix and the color matrix, feature is obtained and melts
Close figure;
Index constraint, interpolation and segmentation are carried out to the Fusion Features figure using color Markov chain model processing result
Optimization processing obtains potential Saliency maps;
Circle characteristic profile and potential Saliency maps are subjected to Bayesian Fusion, obtain fusion notable feature figure;
Binary conversion treatment and denoising are carried out to the fusion notable feature figure, obtain conspicuousness testing result.
Preferably, the specific steps radially extracted include:
Gradient is calculated to the remote sensing images after gray proces, obtains durection component and range weight;
Candidate center location is calculated based on the durection component and the range weight.
Preferably, the remote sensing images Jing Guo super-pixel segmentation are handled based on color Markov Chain conspicuousness model
Specific steps include:
Non-directed graph is calculated based on the weight relationship between neighbouring super pixels block, the boundary of selected digital image is as absorption node;
For each super-pixel block, the absorbing probability of the super-pixel block on all boundaries in image boundary is calculated, and is dropped
Sequence arrangement;
H+1 to k-th absorbing probability value after descending arranges is extracted, the significance value of each super-pixel block is calculated, obtains
Color Markov Chain processing result;Wherein, k indicates the number of absorbing state super-pixel block, and h indicates for the background dot selected
Number, and
Preferably, the method for extracting location matrix includes: based on non-directed graph obtained in the processing of color Markov Chain, meter
The weight for calculating the Euclidean distance and Euclidean distance between each super-pixel block central point, obtains location matrix;
The method for extracting color matrix includes: to be calculated each based on non-directed graph obtained in the processing of color Markov Chain
In the Euclidean distance of CIE Lab color space between super-pixel block vertex, color matrix is obtained.
Preferably, binary conversion treatment is realized using OneCut for GrabCut method.
Preferably, the specific steps of denoising include:
The area information and shape information of noise, and oil are extracted to the fusion significant characteristics figure by binary conversion treatment
The area information and shape information of tank target are simultaneously compared, and determine the area threshold of connected region pixel;
The region that connected domain area is less than connected region elemental area threshold value is removed, remaining connected region is carried out round
The solution of degree finally obtains the testing result of oil tank target;Wherein, the formula of circularity is as follows:
Wherein, Square (domain) indicates the area of connected domain domain, Dp2(domain) perimeter of connected domain is indicated
Square.
It can be seen via above technical scheme that compared with prior art, the present disclosure provides one kind to be based on color horse
The oil tank object detection method of Er Kefu chain conspicuousness model, using contain the markovian unsupervised conspicuousness mould of color
Type distinguishes oil tank region and background area, and color combining and position square by generating bottom-up potential notable figure
Battle array, can be effectively reduced shade, the influence that imaging angle detects oil tank, merge the location and shape information of oil tank finally to subtract
Influence of the weak background area for oil tank target itself.By the oil tank mesh object detection method, can sufficiently accurately detect
Oil tank target in remote sensing images under different illumination conditions and target size out, this method have stronger robustness.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the realization of the oil tank object detection method provided by the invention based on color Markov Chain conspicuousness model
Flow diagram;
Fig. 2 is oil tank object detection method provided by the invention and existing method Contrast on effect result figure one;
Fig. 3 is oil tank object detection method provided by the invention and existing method Contrast on effect result figure two;
Fig. 4 is oil tank object detection method provided by the invention and existing method Contrast on effect result figure three.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to attached drawing 1, for the difficulties of oil tank target detection in remote sensing images, the invention proposes one kind to be based on face
The markovian unsupervised conspicuousness detection model of color, the model pass through building color Markov Chain and color, position
Matrix is set, fusion generates conspicuousness model, and shade, imaging angle and non-oil tank border circular areas can be effectively reduced and detect to oil tank
Influence.
Oil tank mesh calibration method was different, and the present invention converts round radial symmetric feature from utilizing circular feature to detect in the past
For circular gradient figure, and trust evaluation is carried out to border circular areas by circle density matrix, so as to improve the robustness of algorithm.
In addition, the present invention combines bottom-up color characteristic with top-down circular feature figure, one is constructed
A unsupervised oil tank object detection method, not only avoids the extra time cost of large sample training, can also overcome approximation
The interference of circular target and imaging angle detected oil tank region more completely, accurately.
Combination particular technique implementation detail of the embodiment of the present invention is elaborated further technical solution of the present invention.
1, first step absorption is carried out to image using color Markov Chain, obtains rough potential Saliency maps Ma Erke
Husband's chain is a kind of more commonly used familiar random process.Markov Chain containing absorbing state, which is referred to as, absorbs Markov
Chain.Given volume of data Simage={ s1, s2..., sl, what l was indicated is the number of volume of data, each data can be with
A referred to as state.The process is since one of these states and from a state continuous moving to another state.If y
It is two different states with z, if Markov Chain is currently at state sy, then by the pro of referred to as transition probabilityyzIt indicates
Probability is moved to state sz.Therefore, Markov Chain can be determined by transfer matrix P.For with k absorbing state and m
Any absorption Markov Chain of a transient state, obtains the canonical form of above-mentioned transfer matrix P:
Wherein, Q is the probability transfer matrix of momentary status, and R is one and is turned by the probability that momentary status turns to absorbing state
Matrix is moved, element therein proves that the transition probability between transfer and absorbing state, I are then the unit matrixs of k × k, and k is to absorb
State number.Another matrix is obtained according to matrix Q:
N=(I-Q)-1 (2)
N is transfer number matrix.Element N thereinyzIf giving process in transient behaviour syWink is changed into when beginning
State state szAnticipated number.Final absorbing probability matrix is as follows:
B=NR (3)
Color Markov Chain is a kind of random walk model, for the marking area in detection image.Different from absorbing
Markov Chain determines the conspicuousness of image by soak time, and what color Markov Chain utilized is the super-pixel block and edge
The otherness of super-pixel block determines significance value.Since the segmentation of super-pixel indicates, image identification can be become into G (V, E)
Non-directed graph.Wherein what vertex V was indicated is each super-pixel in image, and E is then comprising internuncial one between two super-pixel
Group nonoriented edge.The side for connecting two super-pixel i and j is represented as eij, and wijFor weighted value, expression is based in super-pixel i
Similarity between the feature defined on super-pixel j distributes to side eijWeight.It is defined using CIE Lab color space every
The color characteristic of a super-pixel node, be because Lab color model describe the display mode of color, be based on the mankind for
The feeling of color.Weight relationship expression formula between neighbouring super pixels block i and j are as follows:
Characteristic quantity of the color average of each pixel CIE Lab in super-pixel block as two super-pixel of i and j is taken, c is usedi
And cjIt indicates.And σ is then strength constant, for controlling the intensity of weight, generally by σ2It is set as 0.05, then can be utilized
wijObtain the vertex incidence matrix W=[w of undirected graph model G (V, E)ij]mm.Then it is saved the edge of selected digital image as absorption
Point carries out the markovian process of color.The transient state established between transient state super-pixel block i and absorbing state super-pixel block j simultaneously absorbs
Incidence matrix T=[tio]mk.Wherein m is transient state number, and k is absorbing state number.tioWhat is indicated is the transient state super-pixel in image
The degree of correlation between block i and absorbing state super-pixel block o.Final incidence matrix D={ d is found out according to two above matrix11,
d22..., dmm, whereinTherefore, the probability transfer matrix Q=D of momentary status is enabled-1W, by
The probability transfer matrix R=D of momentary status steering absorbing state-1T.Final absorbing probability matrix B is obtained according to above formula.Then
The conspicuousness of each node is defined as the similitude with image boundary.For the transient state s in color Markov Chaini, it is absorbed
State sjThe probability b of absorptionijWhat is actually indicated is exactly relationship between the two.It, will for each super-pixel block i in image
The absorbing probability value b of the super-pixel block j (j ∈ { 1,2 ..., k }) on all boundaries in image boundaryijIt is arranged according to descending:
bi1≥bi2≥...≥bik (5)
H is arranged (wherein before taking), what h was indicated is the number for selecting background dot.Bs (i) is between super-pixel block
Similar value, i.e.,For indicating the similitude of super-pixel block i and super-pixel block j.F (i) is super-pixel block
Between dissimilar value, useWhat is indicated is the dissimilarity of super-pixel block i and super-pixel block j, is obtained most
The significance value of each super-pixel block eventually are as follows:
Smarkov(i)=e(1-f(i))·bs(i) (6)
The rough Saliency maps picture obtained based on color Markov Chain, that is, prospect probability point are thus obtained
Butut.
2, significant using the color position matrix between super-pixel block, and circle density matrix refinement color Markov Chain
Property model generate as a result, being used to protrude the part of oil tank target.
After image passes through super-pixel segmentation, image is expressed as Simage={ s1, s2..., sl, wherein l is indicated
It is the number of super-pixel, slWhat is indicated is first of super-pixel block.Nothing has been formd during color Markov Chain before
It, thus can be in the hope of the Euclidean distance between each super-pixel block in CIE Lab color space as two to figure G (V, E)
The weighted value on the connected side of super-pixel, referred to as Colorij(si, sj), as color matrix, then seek super-pixel block siAnd sjIn
Euclidean distance between heart point, referred to as dist (si, sj), the weight equation (7) of Euclidean distance is to get arriving location matrix:
Wherein σdistFor the constant constraint value of weight equation, by σdistIt is set as 0.25 according to convention.Last available two
Contrast between a super-pixel block determines the fusion between formula namely location matrix, color matrix and circle density matrix
Are as follows:
Ctr(si, sj)=Colorij(si, sj)wdist(si, sj)dcir(si, sj) (8)
Formula (8) is the fundamental formular to the optimization of color Markov Chain degree of comparing.Ctr(si, sj) what is indicated is super
The size of contrast between block of pixels, referred to as contrast matrix.In the present invention, it is contemplated that the spy of Remote Sensing Target detection
Point and oil tank shape feature, to color Markov Chain degree of comparing optimize while also to incorporate shape feature into
Row processing.For shape feature, circular feature is extracted using by the method for calculating gradient radial symmetric in image.About ladder
The method of radial symmetric is spent, can be explained in detail again in the next section.Wherein dcir(si, sj) it is referred to as circle density comparison matrix, i
Two different super-pixel block in image are respectively corresponded with j, are super-pixel s in the circle density feature for calculate the remote sensing imagesj
And sjBetween average distance.Contrast is determined that the result of formula is constrained by index, is dissolved into color Markov
In chain, formula is as follows:
Sal(si, sj)=Ctr (si, sj)*(1-exp(-6*dmarkov(si, sj))) (9)
Sal(si, sj) what is indicated is the potential significance value of each super-pixel block.Here dmarkov(si, sj) it is known as color
Markov Chain compares matrix, and expression is Smarkov(i) the distance between each super-pixel block in.Then by Sal (si, sj) ask
Just the significance value of final each super-pixel block itself is obtained:
FinalSal (s) is exactly the significance value of final each super-pixel block.
3, the circular shape feature extraction based on radial symmetric
Circle is a symmetrical figure, and point diametrically is symmetrical about the center of circle, if it is possible to which effective use is round
The characteristic of this radial symmetric, then the feature of the related circle extracted also can be more stable and efficient.The process radially extracted is such as
Under: the gradient information of image is obtained using Sobel transformation first, a radius section Radius is then given, enables oil tank target
Radius r ' ∈ Radius.Belong to the radius r ' of Radius for each, it is believed that assuming that there are the circle that radius is r ', that
The gradient of the upper each point of circle can all contribute the radial symmetry at the u of the center of circle, and the pixel of positive contribution is u+(u), and
The pixel of passiveness contribution is u-(u), this cumulative contribution margin is calculated using the durection component of gradient and range weight, is determined
The candidate center of circle.
Wherein,Refer to that the gradient units direction vector at point u, round () indicate the operation that rounds up.It gives
After the positive influence point of all pixels point and negative sense influence point out, according to the u of all pixels in image+(u) and u-(u) it calculates
The positive and negative gradient direction of current radius r ' Gradient and positive and negative gradient amplitude out.
Wherein | g (u) | give directions the range value of gradient at u, the initial value of gradient direction Transformation Graphs and gradient amplitude Transformation Graphs
It is all set as 0, their calculation is exactly according to formula (12) and formula (13), and for all positive negatively influencing points, positive gradient direction becomes
Change the u in figure+(u) place adds 1, the u in positive gradient amplitude Transformation Graphs+(u) place adds | g (u) |;And in negative gradient direction transformation figure
u-(u) place subtracts 1, the u in negative gradient amplitude Transformation Graphs-(u) place subtracts | g (u) |, thus obtained positive and negative gradient direction Transformation Graphs
With positive and negative gradient amplitude Transformation Graphs WithHere four width gradient Transformation Graphs illustrate figure
Cumulative contribution of the gradient of all pixels as in for the radial nature of its surrounding pixel.Rapidly radially symmetry transformation is one
The effect of active influence point and negatively influencing point is integrated in secondary gradient direction Transformation Graphs or gradient amplitude Transformation Graphs.In the present invention, respectively
The effect of active influence point and negatively influencing point is showed, with two width gradient direction Transformation Graphs and gradient amplitude Transformation Graphs in order to strengthen gradient
The effect in direction and amplitude, final gradient direction and gradient amplitude calculate as follows:
And then calculate the radial projected forms of the lower image gradient of current radius r ':
Ur(u)=Mr(u)·Or(u) (15)
Ur(u) be each candidate center of circle weighted value, the weighted value in the center of circle is set to entirely round gray value, so as to
To obtain round characteristic profile.I.e. the higher circle of gray value, its weighted value are higher.Super-pixel segmentation result is then based on to ask
Obtain super-pixel block s in the circle density feature of remote sensing imagesiAnd sjBetween average distance dcir(si, sj)。
4, significant result interpolation and subsection optimization processing
Since the background of original input picture is sufficiently complex, there are part backgrounds not to be filled in some Saliency maps
It the shortcomings that inhibition divided, needs to optimize conspicuousness algorithm.Firstly, being calculated by the K- mean cluster in Lab color space
The super-pixel block of input picture is grouped into Y cluster by method, and Y is cluster number.Then the conspicuousness of each super-pixel block can be with
It is modified by the simple interpolations in same cluster.About super-pixel block siConspicuousness optimization can pass through following public affairs
Formula is realized:
Sf(si) it is super-pixel block s after interpolationiSignificance value, among these β be each cluster in contained node number,
DlijFor interpolation factor, for comparing difference of the super-pixel block i and j in CIE Lab space.ciAnd cj
It is channel value of the super-pixel block i and j in CIE Lab space respectively.And σIIt is in each characteristic dimension of CIE Lab space
The summation of variance.The item in left side indicates node s in above formulaiThe conspicuousness tentatively optimized, and the FinalSal (s of right side of the formulai)
With FinalSal (sj) it is respectively node siAnd sjOriginal significant result.α in above formula is balance parameters, is rule of thumb set
It is set to 0.5.
Although above-mentioned interpolation method effectively highlights some foreground parts, the background parts of some images are not
Inhibited well.In order to further solve this problem, introduce a simple piecewise function remove it is not significant or
The conspicuousness of error large area.Function is defined as:
Wherein θ is the threshold value defined significantly with not significant range, is empirically set as 0.6.X is the aobvious of the super-pixel block
Work property value.Significance value after inhibiting for segmentation.The final image of conspicuousness part is thus obtained.
5, the Bayesian Fusion of potential Saliency maps and shape feature figure
After carrying out shape extraction and color feature extracted respectively to target image, need both different methods
Integrate the result that can reach ideal.Therefore, it introduces Bayes and integrates function and carry out knot to above two method
Fruit is merged.
The potential Saliency maps final result obtained after color Markov Chain conspicuousness model optimization is set as S1,
The circle characteristic profile that the oil tank object detection method of radial symmetric generates is set as S2.If p and q are respectively the angle of image
Mark.First by piece image S thereinp(p={ 1,2, }) is set as foreground image, by other piece image Sq(q ≠ p, q=
{ 1,2 }) it is set as background image, come a possibility that both calculating, it can integrate more letters from different Saliency maps in this way
Breath.First by image SpThreshold segmentation is carried out, uses the average gray value of image as threshold value.Image is partitioned into F respectivelyp
And Bp.Wherein FpFor foreground area, and BpFor background area, z is pixel.In each area, by comparing SqAnd SpIn picture
Calculability is carried out in foreground and background region at plain z, and formula is as follows:
Sq(z) characteristic value for being pixel z.Wherein p (Sq(z)|Fp) and p (Sq(z)|Bp) respectively indicate be and pixel z
Pixel with same color feature vector respectively appears in target collection FpWith background set BpProbability.Indicate be
The number of pixels of foreground area, andWhat is indicated is comprising color characteristic Sq(z) foreground areaPixel
Number.SimilarlyWhat is indicated is the number of pixels of background area,What is indicated is comprising color characteristic Sq(z) before
Scene areaNumber of pixels.Use SpObtain the calculation formula of posterior probability are as follows:
p(Fp|SqIt (z)) is foreground area FpPosterior probability values.Similarly, also by by p and q exchange, p (F is obtainedq|Sp
(z)) result.Next final Saliency maps are calculated using the two posterior probability, formula is as follows:
Smap(S1(z), S2(z))=p (F1|S2(z))+p(F2|S1(z)) (21)
The result of potential Saliency maps is wherein set as S1, F1For S1Foreground area examines the oil tank target of radial symmetric
The result of survey method is set as S2, F2For S2Foreground area.Smap(S1(z), S2It (z)) is the notable figure after Bayes's integration
As a result, its result can uniformly protrude the marking area in image.
6, the post-processing of algorithm
The part of post-processing is introduced, the reason is that needing obtained final result is a binary map.Introduce OneCut for
The binarization of GrabCut method realization notable figure.It is the modified version of GrabCut, and an energy can be set most in it
Smallization function calculates the threshold value in every picture, while being also conceivable to the integrality of segmentation result.Minimizing energy can be with
It is acquired by following formula:
Wherein, SmapS inside the Saliency maps for being, that is, formula (21)map(S1(z), S2(z)), smap(v) refer to
Be pixel v significance value.And spWhat is indicated is whether this pixel belongs to prospect.Belonging to prospect is then 1, belongs to background then
It is 0.βlBe then background and prospect overlapping penalty term, andIt is then a smooth item.Available each using OneCut
The binaryzation form of figure.But this method can also bring some small noise problems simultaneously, the noisy face by extracting
Product information and shape information, while being compared with the area in oil tank region and shape, determine the pixel of removal connected region
Then threshold value removes noise according to threshold value.Connected domain area less than 40 pixels (i.e. above-mentioned threshold value, the threshold values are removed first
Be determined according to the actual situation, be not a fixed threshold value) region, then remaining all connected domains are justified
The solution of shape degree.Circularity formula is as follows:
Wherein Square (domain) is the area of connected domain domain, and Dp2(domain) what is indicated is connected domain
Square of perimeter.Then the part for not meeting similar round is all removed, remaining connected domain is considered as final tank farm
Domain obtains final testing result.
Detection method provided by the invention, which uses, is based on the markovian conspicuousness model of color, is utilizing face first
Color Markov chain model is treated in journey input picture, due to shadow color and oil tank main body itself otherness with
And the similitude with background, cause it to be easier to be absorbed.So the significance value of shade is generally lower, and oil tank ontology is aobvious
Work property value is relatively high.Using further more matrix restraints, it can be seen that shade is substantially eliminated, and institute is in this way
For shade there are the phenomenon that show to obtain more robust.Meanwhile when oil tank at an angle when, by color interpolation, oil
The body part of tank can also be detected well.So even if oil tank exists due to terrestrial latitude with shooting camera
Certain angle, oil tank still are able to more completely be detected, are to fully achieve just with round this single features
Not.Finally, the method based on SHAPE DETECTION can generally be regarded as being oil tank when there is round jamming target, and
Method provided by the invention passes through the comparison in the region and background color, can be determined as background.It is obvious that relying solely on shape
Shape detection, in oil tank context of detection, there is significant limitations.Method provided by the invention can well solve this herein
The limitation of sample improves the accuracy rate entirely detected.
Below with reference to experimental result, technical scheme is described further.
The present invention is tested with figure as shown in figs. 2 to 4, and the resolution ratio of image is 400 × 400, the radius of oil tank target
Between 7 meters to 40 meters, the advantages of showing this method under three kinds of complex situations, the first situation is that tool is imaged in oil tank itself
There is certain visual angle, show three-dimensional structure, referring to attached drawing 2;Second situation is that similar round target jamming occurred, referring to attached drawing
3;The third situation is since the case where shadow interference occurs in the influence of illumination, referring to attached drawing 4.Wherein figure a is to be originally inputted
Image, figure b are the arithmetic result figure that oil tank detection is carried out according to shape, and figure c is the oil tank detection obtained based on the method for the present invention
Result figure.The part correctly detected is represented in green among these, red represents the part not detected, and blue represents mistake
The part of detection.It can be seen that method provided by the invention all has when interfering in face of shadow interference, shape interference and angle
Good detection performance.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For device disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (6)
1. a kind of oil tank object detection method based on color Markov Chain conspicuousness model characterized by comprising
Gray proces and super-pixel segmentation are carried out to collected remote sensing images;
Radial extraction is carried out to the remote sensing images by gray proces, obtains the candidate center of circle;
The weighted value for calculating each candidate center of circle obtains round feature distribution using the weighted value in the center of circle as entirely round gray value
Figure;
Circle density matrix is calculated based on super-pixel segmentation result and the round characteristic profile;
The remote sensing images Jing Guo super-pixel segmentation are handled based on color Markov Chain conspicuousness model, and are extracted respectively
Location matrix and color matrix;
Fusion Features are carried out based on the round density matrix, the location matrix and the color matrix, obtain Fusion Features figure;
Index constraint, interpolation and subsection optimization are carried out to the Fusion Features figure using color Markov chain model processing result
Processing, obtains potential Saliency maps;
Circle characteristic profile and potential Saliency maps are subjected to Bayesian Fusion, obtain fusion notable feature figure;
Binary conversion treatment and denoising are carried out to the fusion notable feature figure, obtain conspicuousness testing result.
2. the oil tank object detection method according to claim 1 based on color Markov Chain conspicuousness model, special
Sign is that the specific steps radially extracted include:
Gradient is calculated to the remote sensing images after gray proces, obtains durection component and range weight;
Candidate center location is calculated based on the durection component and the range weight.
3. the oil tank object detection method according to claim 1 based on color Markov Chain conspicuousness model, special
Sign is, the specific step handled based on color Markov Chain conspicuousness model the remote sensing images Jing Guo super-pixel segmentation
Suddenly include:
Non-directed graph is calculated based on the weight relationship between neighbouring super pixels block, the boundary of selected digital image is as absorption node;
For each super-pixel block, the absorbing probability of the super-pixel block on all boundaries in image boundary is calculated, and carries out descending row
Column;
H+1 to k-th absorbing probability value after descending arranges is extracted, the significance value of each super-pixel block is calculated, obtains color
Markov Chain processing result;Wherein, k indicates the number of absorbing state super-pixel block, and h indicates the number for the background dot selected, and
4. the oil tank object detection method according to claim 1 based on color Markov Chain conspicuousness model, special
Sign is,
The method for extracting location matrix includes: to calculate each super picture based on non-directed graph obtained in the processing of color Markov Chain
The weight of Euclidean distance and Euclidean distance between plain block central point, obtains location matrix;
The method for extracting color matrix includes: to calculate each super picture based on non-directed graph obtained in the processing of color Markov Chain
In the Euclidean distance of CIE Lab color space between plain block vertex, color matrix is obtained.
5. the oil tank object detection method according to claim 1 based on color Markov Chain conspicuousness model, special
Sign is, realizes binary conversion treatment using OneCut for GrabCut method.
6. the oil tank object detection method according to claim 5 based on color Markov Chain conspicuousness model, special
Sign is that the specific steps of denoising include:
To the area information for merging significant characteristics figure extraction noise and shape information and oil tank for passing through binary conversion treatment
Area information and shape information are simultaneously compared, and determine the area threshold of connected region pixel;
The region that connected domain area is less than connected region elemental area threshold value is removed, circularity is carried out to remaining connected region
It solves, finally obtains the testing result of oil tank target;Wherein, the formula of circularity is as follows:
Wherein, Square (domain) indicates the area of connected domain domain, Dp2(domain) the flat of the perimeter of connected domain is indicated
Side.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910272945.1A CN110097569B (en) | 2019-04-04 | 2019-04-04 | Oil tank target detection method based on color Markov chain significance model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910272945.1A CN110097569B (en) | 2019-04-04 | 2019-04-04 | Oil tank target detection method based on color Markov chain significance model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110097569A true CN110097569A (en) | 2019-08-06 |
CN110097569B CN110097569B (en) | 2020-12-22 |
Family
ID=67444406
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910272945.1A Active CN110097569B (en) | 2019-04-04 | 2019-04-04 | Oil tank target detection method based on color Markov chain significance model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110097569B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110765875A (en) * | 2019-09-20 | 2020-02-07 | 浙江大华技术股份有限公司 | Method, equipment and device for detecting boundary of traffic target |
CN111179334A (en) * | 2019-11-14 | 2020-05-19 | 青岛理工大学 | Sea surface small-area oil spilling area detection system and detection method based on multi-sensor fusion |
Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001046707A2 (en) * | 1999-12-22 | 2001-06-28 | Siemens Corporate Research, Inc. | Method for learning-based object detection in cardiac magnetic resonance images |
WO2006029227A2 (en) * | 2004-09-07 | 2006-03-16 | Siemens Medical Solutions Usa, Inc. | System and method for anatomical structure parsing and detection |
CN101726498A (en) * | 2009-12-04 | 2010-06-09 | 河海大学常州校区 | Intelligent detector and method of copper strip surface quality on basis of vision bionics |
CN102867313A (en) * | 2012-08-29 | 2013-01-09 | 杭州电子科技大学 | Visual saliency detection method with fusion of region color and HoG (histogram of oriented gradient) features |
CN104408712A (en) * | 2014-10-30 | 2015-03-11 | 西北工业大学 | Information fusion-based hidden Markov salient region detection method |
CN104574375A (en) * | 2014-12-23 | 2015-04-29 | 浙江大学 | Image significance detection method combining color and depth information |
CN105426895A (en) * | 2015-11-10 | 2016-03-23 | 河海大学 | Prominence detection method based on Markov model |
CN105957054A (en) * | 2016-04-20 | 2016-09-21 | 北京航空航天大学 | Image change detecting method |
CN105976378A (en) * | 2016-05-10 | 2016-09-28 | 西北工业大学 | Graph model based saliency target detection method |
CN106909902A (en) * | 2017-03-01 | 2017-06-30 | 北京航空航天大学 | A kind of remote sensing target detection method based on the notable model of improved stratification |
US20170289434A1 (en) * | 2016-03-29 | 2017-10-05 | Sony Corporation | Method and system for image processing to detect salient objects in image |
CN107609552A (en) * | 2017-08-23 | 2018-01-19 | 西安电子科技大学 | Salient region detection method based on markov absorbing model |
US20180165541A1 (en) * | 2016-12-12 | 2018-06-14 | Jack Cooper Logistics, LLC | System, method, and apparatus for detection of damages on surfaces |
CN108427919A (en) * | 2018-02-22 | 2018-08-21 | 北京航空航天大学 | A kind of unsupervised oil tank object detection method guiding conspicuousness model based on shape |
CN108629286A (en) * | 2018-04-03 | 2018-10-09 | 北京航空航天大学 | A kind of remote sensing airport target detection method based on the notable model of subjective perception |
CN108921873A (en) * | 2018-05-29 | 2018-11-30 | 福州大学 | The online multi-object tracking method of Markovian decision of filtering optimization is closed based on nuclear phase |
CN108921833A (en) * | 2018-06-26 | 2018-11-30 | 中国科学院合肥物质科学研究院 | A kind of the markov conspicuousness object detection method and device of two-way absorption |
CN109471081A (en) * | 2018-11-07 | 2019-03-15 | 中国人民解放军国防科技大学 | Single pulse radar weak and small target combined detection and state estimation method |
-
2019
- 2019-04-04 CN CN201910272945.1A patent/CN110097569B/en active Active
Patent Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001046707A2 (en) * | 1999-12-22 | 2001-06-28 | Siemens Corporate Research, Inc. | Method for learning-based object detection in cardiac magnetic resonance images |
WO2006029227A2 (en) * | 2004-09-07 | 2006-03-16 | Siemens Medical Solutions Usa, Inc. | System and method for anatomical structure parsing and detection |
CN101726498A (en) * | 2009-12-04 | 2010-06-09 | 河海大学常州校区 | Intelligent detector and method of copper strip surface quality on basis of vision bionics |
CN102867313A (en) * | 2012-08-29 | 2013-01-09 | 杭州电子科技大学 | Visual saliency detection method with fusion of region color and HoG (histogram of oriented gradient) features |
CN104408712A (en) * | 2014-10-30 | 2015-03-11 | 西北工业大学 | Information fusion-based hidden Markov salient region detection method |
CN104574375A (en) * | 2014-12-23 | 2015-04-29 | 浙江大学 | Image significance detection method combining color and depth information |
CN105426895A (en) * | 2015-11-10 | 2016-03-23 | 河海大学 | Prominence detection method based on Markov model |
US20170289434A1 (en) * | 2016-03-29 | 2017-10-05 | Sony Corporation | Method and system for image processing to detect salient objects in image |
CN105957054A (en) * | 2016-04-20 | 2016-09-21 | 北京航空航天大学 | Image change detecting method |
CN105976378A (en) * | 2016-05-10 | 2016-09-28 | 西北工业大学 | Graph model based saliency target detection method |
US20180165541A1 (en) * | 2016-12-12 | 2018-06-14 | Jack Cooper Logistics, LLC | System, method, and apparatus for detection of damages on surfaces |
CN106909902A (en) * | 2017-03-01 | 2017-06-30 | 北京航空航天大学 | A kind of remote sensing target detection method based on the notable model of improved stratification |
CN107609552A (en) * | 2017-08-23 | 2018-01-19 | 西安电子科技大学 | Salient region detection method based on markov absorbing model |
CN108427919A (en) * | 2018-02-22 | 2018-08-21 | 北京航空航天大学 | A kind of unsupervised oil tank object detection method guiding conspicuousness model based on shape |
CN108629286A (en) * | 2018-04-03 | 2018-10-09 | 北京航空航天大学 | A kind of remote sensing airport target detection method based on the notable model of subjective perception |
CN108921873A (en) * | 2018-05-29 | 2018-11-30 | 福州大学 | The online multi-object tracking method of Markovian decision of filtering optimization is closed based on nuclear phase |
CN108921833A (en) * | 2018-06-26 | 2018-11-30 | 中国科学院合肥物质科学研究院 | A kind of the markov conspicuousness object detection method and device of two-way absorption |
CN109471081A (en) * | 2018-11-07 | 2019-03-15 | 中国人民解放军国防科技大学 | Single pulse radar weak and small target combined detection and state estimation method |
Non-Patent Citations (4)
Title |
---|
JINGANG SUN 等: "Saliency Region Detection Based on Markov Absorption Probabilities", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
LIBAO ZHANG 等: "Saliency-Driven Oil Tank Detection Based on Multidimensional Feature Vector Clustering for SAR Images", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
姜博文: "基于马尔可夫链的显著性检测", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
韩现伟: "大幅面可见光遥感图像典型目标识别关键技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110765875A (en) * | 2019-09-20 | 2020-02-07 | 浙江大华技术股份有限公司 | Method, equipment and device for detecting boundary of traffic target |
CN110765875B (en) * | 2019-09-20 | 2022-04-19 | 浙江大华技术股份有限公司 | Method, equipment and device for detecting boundary of traffic target |
CN111179334A (en) * | 2019-11-14 | 2020-05-19 | 青岛理工大学 | Sea surface small-area oil spilling area detection system and detection method based on multi-sensor fusion |
CN111179334B (en) * | 2019-11-14 | 2024-03-19 | 青岛理工大学 | Sea surface small-area oil spill area detection system and detection method based on multi-sensor fusion |
Also Published As
Publication number | Publication date |
---|---|
CN110097569B (en) | 2020-12-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109583425B (en) | Remote sensing image ship integrated recognition method based on deep learning | |
Bai et al. | Deep watershed transform for instance segmentation | |
Brandtberg et al. | Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis | |
Li et al. | Multiscale features supported DeepLabV3+ optimization scheme for accurate water semantic segmentation | |
CN102360421B (en) | Face identification method and system based on video streaming | |
Chen et al. | A double-threshold image binarization method based on edge detector | |
CN107464252A (en) | A kind of visible ray based on composite character and infrared heterologous image-recognizing method | |
CN107633226B (en) | Human body motion tracking feature processing method | |
CN111079739B (en) | Multi-scale attention feature detection method | |
CN109299720A (en) | A kind of target identification method based on profile segment spatial relationship | |
CN110309808B (en) | Self-adaptive smoke root node detection method in large-scale space | |
CN108510504A (en) | Image partition method and device | |
Cao et al. | Infrared small target detection based on derivative dissimilarity measure | |
CN103870818A (en) | Smog detection method and device | |
CN110097569A (en) | Oil tank object detection method based on color Markov Chain conspicuousness model | |
CN112418165A (en) | Small-size target detection method and device based on improved cascade neural network | |
Michael et al. | A general framework for human-machine digitization of geographic regions from remotely sensed imagery | |
Zhu et al. | AOPDet: Automatic organized points detector for precisely localizing objects in aerial imagery | |
Zardoua et al. | A survey on horizon detection algorithms for maritime video surveillance: advances and future techniques | |
Chen et al. | Change detection algorithm for multi-temporal remote sensing images based on adaptive parameter estimation | |
Gustafsson et al. | Automotive 3D object detection without target domain annotations | |
Chen et al. | Shape similarity intersection-over-union loss hybrid model for detection of synthetic aperture radar small ship objects in complex scenes | |
CN114140484A (en) | High-robustness sea-sky-line extraction method based on photoelectric sensor | |
CN116310832A (en) | Remote sensing image processing method, device, equipment, medium and product | |
CN110211106A (en) | Average drifting SAR image coastline Detection Method method based on segmentation Sigmoid bandwidth |
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 |