CN102509119A - Method for processing image scene hierarchy and object occlusion based on classifier - Google Patents
Method for processing image scene hierarchy and object occlusion based on classifier Download PDFInfo
- Publication number
- CN102509119A CN102509119A CN2011103018997A CN201110301899A CN102509119A CN 102509119 A CN102509119 A CN 102509119A CN 2011103018997 A CN2011103018997 A CN 2011103018997A CN 201110301899 A CN201110301899 A CN 201110301899A CN 102509119 A CN102509119 A CN 102509119A
- Authority
- CN
- China
- Prior art keywords
- image
- hiding relation
- sorter
- scene
- clue
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a method for processing image scene hierarchy and object occlusion based on a classifier, which provides a new technical scheme for automatically judging the hierarchical relationship of image scenes and processing the occlusion relationship of objects. The method comprises the following steps: using features of five occlusion clues (i.e. semantics, location, contour, common boundary and adjacent point) on a training set, and training to obtain a classifier for judging the occlusion relationship; performing the occlusion relationship classifier on a test image to calculate an occlusion relationship predicted value, and constructing a weighted directional graph representing the occlusion relationship; by using a hierarchy ordering and reasoning algorithm, reasoning out the hierarchical structure of the image scenes on the weighted directional graph; and by using the obtained hierarchical structure of the scenes, processing the occlusion relationship between a newly-added object and the available scene objects to generate an image virtual scene. The method can be used for the occlusion processing of image or video virtual scenes, the generation of image or video virtual scenes and other applications.
Description
Technical field
The invention belongs to Flame Image Process and computer vision field, specifically a kind of image scene layering and object occlusion handling method based on sorter.
Background technology
Virtual scene generation technique based on video material is an important in Virtual Reality, also is the research focus of virtual reality, augmented reality, computer vision and the organic intersection of correlative study direction.It is the important component part that the video virtual scene generates that the image virtual scene generates, and wherein, it is the key issue that image virtual scene generation technique need solve that the blocking of image virtual scene handled.In the virtual scene generative process; Need to handle between the scenario objects, the various spaces hiding relation between scenario objects and the video scene; With the spatial relation of guaranteeing that the video virtual scene is correct, its key of problem is how to confirm the relative tertiary location of each scenario objects.With to block disposition in the past the augmented reality scene different; When virtual objects is joined another image scene; Must obtain the front and back position relation of each object in this image scene; Because image is two-dimentional, the front and back position relation of object just is converted into hiding relation between object and hierarchical relationship in the image.On the basis of the hierarchical relationship that obtains image scene, virtual objects is joined in this scene, according to the occlusion issue of known image scene hierarchical relationship correct handling virtual objects, thereby generate the image virtual scene that meets user's request.Therefore with to block disposition different in the past the augmented reality scene; Content of the present invention is front and back hiding relation and the hierarchical relationship between object in the correct handling image virtual scene; Realize stratification graphical representation; For initiate object correct handling hiding relation, generate new image virtual scene.
There have been at present many researchers to be devoted to the processing that image scene layering and object block, image scene layering and object are blocked processing combined research but also have no talent.2006, the people such as James H.Elder at Canadian York University vision research center improved profile polymerization algorithm, proposed to realize the extraction of image object profile by coming the overall situation of presentation video to limit to accurate multiple dimensioned Bayes's profile extraction algorithm roughly.But this method is not studied the level hiding relation of scenario objects.2007, the graduate height of the Central China University of Science and Technology and Lianhua Shan Mountain proposed based on the Bayesian inference algorithm inference graph that mixes markov random file as hierarchical structure like people such as new and Wu Tianfu.Utilize the hierarchical structure problem that solves true picture based on markov random file modeling and reasoning first.But this method does not have the disposal route of further research object hiding relation.Analyze the current various method of finding the solution the image scene lamination problem; These methods generally all are the various middle low layer information that study characterizes hiding relation from image; Then according to Bayesian inference theory and statistical computation; Set up the mathematical model of scene lamination problem, utilize reasoning algorithm in solution space, to search for possible layering and separate the result.These methods are the structure problems in pixel scale research hierarchical structure, do not infer the layering result that most probable in the image scene is represented its hierarchical structure.Other image scene layered approach of research object level of the present invention utilizes known high level of object and middle low layer block information characteristic, the sorter of training hiding relation, and the graph structure of structure object hiding relation is represented and the object hierarchy reasoning algorithm.After obtaining the hierarchical structure and virtual objects interested of image scene, the present invention just can join any position level in the image scene with virtual objects, and the situation of blocking between can the correct handling object, thereby generates image virtual scene true to nature.
Summary of the invention
In order to overcome the deficiency of prior art, the objective of the invention is to: a kind of image scene layering and object occlusion handling method based on sorter is provided.This method can obtain hiding relation between object and hierarchical relationship in the image scene according to the feature inference of existing image scene; Thereby can the initiate object of correct handling and existing object between the space hiding relation; Make the image virtual scene that generates to meet objective reality and customer requirements simultaneously, effective technical support is provided for blocking in the image virtual scene to handle problems.
For accomplishing goal of the invention; Technical solution of the present invention is: at first; On image scene semantic marker result's basis, use semanteme, position, profile, public boundary and point of interface to block clue for five kinds, and above-mentioned five kinds of detection training dataset image block clue; At the positive sample and the negative sample of training dataset up-sampling hiding relation, utilize five kinds to block the sextuple proper vector that clue makes up each sample, training obtains the sorter of hiding relation thus; For any input picture; Structure belongs to the hiding relation proper vector set of this image; Utilize the sorter trained, obtain belonging to the predicted value of all hiding relations of this image, make up the complete digraph of cum rights of performance hiding relation with this; Reasoning solves the hierarchical structure of image scene on the complete digraph of cum rights; At last, the object of appointment is put into the appointment level of layering scene, correct handling is blocked phenomenon, generates the image virtual scene.
In the image scene semantic marker, exist a plurality of in low layers and high-rise information interregional hiding relation and the hierarchical relationship of reflection object to a certain extent.The present invention's multiclass information selected of comforming goes out object semantic clues, object's position clue, object outline clue, public boundary clue and five kinds of clues that can characterize hiding relation between subject area of point of interface clue, for the structure of the sorter of the hiding relation of back provides proper vector.
The semantic marker figure that given image is corresponding can obtain the hiding relation between the different semantic objects according to people's experience, therefore obtains the hiding relation response between the different semantic objects in the training stage.Regional location clue, region contour compactedness clue, regional public boundary clue and regional point of interface clue can both be expressed interregional hiding relation and scene hierarchical structure to a certain extent.Block the eigenwert construction feature vector that clue obtains, the sorter of training hiding relation according to these.With hiding relation < A; B>be example; Its 6 DOF blocks that the characteristic response value of clue is blocked in the position that semanteme that proper vector is meant A and B blocks characteristic response value, A and the B of clue, the profile of A blocks the characteristic response value of clue; The profile of B blocks the characteristic response value that public boundary between characteristic response value, A and the B of clue blocks clue, and the public point of interface between A and the B blocks the characteristic response value of clue.Come the preference score value of hiding relation in the predicted picture through sorter, the preference score value of a hierarchical sequence is the preference score value sum of all hiding relations in this sequence.When the preference score value reached maximal value, it was the hierarchical structure that most probable occurs in the image scene that corresponding image level is represented to separate.Therefore, only require to such an extent that maximum preference score value is separated, just can realize this target of image scene layering.
Publish picture as the optimum solution of layering for rapid solving, the present invention proposes a kind of reasoning algorithm of finding the solution hierarchical sequence, transforms the problem of on the complete digraph of cum rights of expression hiding relation, finding the solution best ordering with finding the solution reasoning process.
The present invention's beneficial features compared with prior art is: 1, the present invention is on the basis of picture material structure analysis and image scene semantic marker; Handle hiding relation and hierarchical relationship between image object from the angle of stratification graphical representation, the method for processing image hiding relation different from the past.2, the present invention is blocked the proper vector that clue makes up the hiding relation sample according to five kinds, the preference score value that the positive negative sample of sampling hiding relation is predicted hiding relation and provided hiding relation with training classifier.3, the present invention converts two dimensional image into the cum rights digraph and representes, and has proposed a kind of level reasoning algorithm of arranging entirely fast, on the cum rights digraph, finds the solution the image level structure.4, the present invention can specify interested object and level by the user, object is joined on the appointment level of layering scene, and the space hiding relation between the correct handling object generates syncretizing effect virtual scene preferably.
Description of drawings
Fig. 1 is a hiding relation transitivity synoptic diagram of the present invention;
Fig. 2 is an overall construction drawing of the present invention;
Fig. 3 is the feature extraction schematic flow sheet that blocks of the present invention;
Fig. 4 is a cum rights digraph synoptic diagram of the present invention;
Fig. 5 is a hierarchical reasoning process flow synoptic diagram of the present invention;
Fig. 6 is an overall process synoptic diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing the present invention is elaborated.
The present invention proposes a kind of image scene layering and object occlusion handling method based on sorter, on existing image scene semantic marker basis, and the sorter of training hiding relation; For any input picture; Utilize the sorter that has trained; Obtain the predicted value of all hiding relations of test pattern; With this complete digraph of cum rights that makes up the performance hiding relation, rapid solving goes out the hierarchical structure of image scene and generates new virtual scene on the complete digraph of cum rights.
Key step of the present invention is as shown in Figure 2: various clue and the characteristic response values thereof of blocking that at first detect the training dataset image; By the proper vector of these characteristic response values structure hiding relation samples, utilize the training of Adaboost method to obtain the sorter of hiding relation then; For the image of an input, detect and block clue, extract wherein hiding relation and proper vector, use the hiding relation sorter that trains to obtain the prediction character and the predicted value of each hiding relation, structure belongs to the cum rights digraph of the hiding relation of this figure; The last hierarchical ranking reasoning algorithm that on cum rights digraph structure, utilizes the present invention to propose is found the solution the hierarchical structure of image scene.At last, the object of appointment is put into the appointment level of layering scene, correct handling is blocked phenomenon, generates the image virtual scene.
The present invention has defined five kinds of clues of blocking that characterize hiding relation on the basis of image scene semantic marker, detect the various of data set image and block clue and calculate its characteristic response value.
Five kinds of alternative semantic clues of the present invention, object's position clue, object outline clue, public boundary clue and point of interface clues can characterize the clue of hiding relation and hierarchical relationship between subject area.In the natural image scene, the semantic information between object can provide relative block information, as horse block the meadow, automobile obstructs the road, so semantic information is one of characteristic of scene layering.Generally speaking, be positioned at the downside of image near the object of camera lens, generally below, sky generally is in the top such as ground.This shows, the position of subject area on image, hiding relation and hierarchical relationship between reflection object also are one of characteristics of scene layering to a certain extent.When object was not blocked by other objects near the place ahead, its contour shape generally was a rule and compact.Therefore region contour is more regular, and it is more little by the possibility that other objects block, and this shows, the subject area profile is the situation that is blocked of reflecting regional self to a certain extent, is one of characteristic of scene layering.The interregional public boundary of adjacent object is the hiding relation between reflecting regional to a certain extent, is one of important clue that solves the image scene lamination problem.In the natural image scene, when three objects occurred blocking phenomenon, one of them object sheltered from other two objects usually, showed to be on the two dimensional image point of interface phenomenon to occur.Therefore, the interregional point of interface of contiguous object also is the important clue of hiding relation.The curve that the present invention will constitute point of interface is reduced to vector form, in the subrange of point of interface, asks the direction vector of the mean change of curve in limited pixel according to the trend of each curve, counterclockwise to select the angle between two vectors to represent point of interface J
t
Work of the present invention at first need detect blocks clue and calculates its characteristic response value, as shown in Figure 3, detects that to block the clue step following: at first, the semantic marker image is read in, therefrom extract the semantic marker of each object, the objects of statistics number; Then, initialization subject area information, position, zoning; Then, according to subject area information extraction region contour, and the length and the area of calculating profile; Then, each contours of objects is split into the interregional public boundary of contiguous object, calculates the curvature and the length of public boundary; At last, detect at the end points place of public boundary and to judge whether to be point of interface, the position of initialization point of interface, separated region, three boundary curve sections, describe the information such as angle of point of interface shape.After various in obtaining image were blocked hint information, the present invention designed and realizes various mathematical models definition and response computation processes of blocking clue.
The response of object semantic relation is calculated, and at first two object R of computing semantic relation are wanted in input
iAnd R
j, because different colours is represented the different semantics object in semantic marker figure, so discern two object R
iAnd R
jColor, promptly discern the semantic information of two objects, import the layering result of artificial division then, obtain the hiding relation histogram of different semantic objects, horizontal ordinate and ordinate all are all semantic classess of data centralization.(x y) is example, the value S of correspondence on this coordinate (x, y) frequency values of expression semantic classes x occlusion semantics classification y with coordinate.Block the characteristic response value of clue as semanteme with this histogrammic statistics.
The mathematical model of position clue is:
For two section object R
iAnd R
j,, therefrom select Y direction height value according to the regional center position of blocking each object that the clue detection computations goes out
With
H is a picture altitude.Utilize position clue calculated with mathematical model two objects to measure the hiding relation value P between the two from the relative position angle
Pos, at last this response is returned.
The present invention is according to the area and the length information of each region contour that in blocking the clue detection, obtains; Utilize the compactedness of the single region contour of profile compactedness calculated with mathematical model this object of tolerance and the response of the relation of being blocked; Then the response of these all profiles of object is averaged, at last this mean value is returned as the response of subject area profile compactedness.Region R compactedness mathematical model is following, and wherein, L is a profile length, and A is a region area, the number of pixels that n is comprised for the zone, and α is a weighting coefficient:
In order to describe the relation between interregional public boundary and the hiding relation; The present invention utilizes curvature definition public boundary modulus of convexity type function to be:
wherein; κ is the curvature of public boundary any point;
is anticlockwise public boundary curve, and L is the public boundary curve arc long.Utilize public boundary convexity Model Calculation two objects to judge the response p of hiding relation from the public boundary clue
ConvAnd it is returned.In given two region R
i, R
jHiding relation R
i＜R
j(be R
iBlock R
j) condition under, the present invention definition and the mathematical model that quantizes curved transition reflection hiding relation do
Promptly
Wherein, possibly have many public boundaries between two objects, N is a region R
i, R
jBetween the public boundary number.
Definition point of interface shape mathematical model is: p
Ang(J|R
i＜R
j, R
i＜R
k) ∝ p
J(θ
1, θ
2), wherein, R
i, R
j, R
kThere are three zones of point of interface in expression, and its hiding relation is R
i＜R
j, R
i＜R
k, i.e. R
iBlock R
j, R
iBlock R
kp
J(θ
1, θ
2) be the vector angle of under current hiding relation, confirming, θ
1Be region R
iThe angle of region within the jurisdiction, θ
2Be the vector angle that continues to confirm along counterclockwise.Point of interface shape Statistics amount is calculated, and at first input will be calculated the point of interface J of point of interface shape reflection hiding relation
t,, utilize the interregional hiding relation response of point of interface shape calculated with mathematical model three contiguous object of above-mentioned definition and return then according at the angle information that blocks the description point of interface shape that clue calculates in detecting.
At the positive sample and the negative sample of training dataset up-sampling hiding relation, utilize five kinds to block the hiding relation proper vector that clue makes up each sample, training obtains the sorter of hiding relation thus.
After obtaining the response of respectively blocking clue, we next step what will do is exactly to set up the cum rights digraph structure that image blocks level, thereby carry out the hierarchical sequence reasoning.At first we will obtain hiding relation prediction and predicted value thereof between each semantic object.We have defined a kind of preference function PREF, and it is the associating of a two-value indicator function.The PREF function converts the hiding relation proper vector into the predicted value that can express hiding relation, i.e. preference score value.The definition of PREF function is following:
Wherein A＜B represent A before the B or A block B, A＞B represent B before A or B block A, the preference score value is high more to show that this hiding relation possibility is big more.According to the characteristics of PREF function, we selected the Adaboost method to train to obtain we by the linear hiding relation strong classifier of forming of some Weak Classifiers.Finally, our PREF function is found the solution like this:
h
t(x)＝1[f
t(x)＞θ
t]
Wherein x is a proper vector, and h (x) is the predicted value about proper vector, h
t(x) be Weak Classifier (the present invention selects decision tree for use), θ
tBe Weak Classifier h
t(x) threshold value, α
tBe the weighted value of each Weak Classifier, T is the number of Weak Classifier, f
t(x) be binaryzation function about proper vector x.When satisfying f
t(x)＞θ
tThe time, h
t(x)=1; When not satisfying f
t(x)＞θ
tThe time, h
t(x)=0.If PREF (A, B) and PREF (B, A) all be positive number and PREF (A, (when B, A) absolute value, then A blocks B and more likely takes place absolute value B) greater than PREF; When PREF (A, B) and PREF (B is that (A, (when B, A) absolute value, then B blocks A and more likely takes place absolute value B) greater than PREF for negative and PREF A).
For hiding relation<a, B>, its proper vector FV<objectA, objectB>Be used for measuring the hiding relation between A and the B, it be by the corresponding semanteme of A and B block clue characteristic response value S (A, B), the position of A and B blocks the characteristic response value P of clue
Pos<a, B>, A profile block the characteristic response value P of clue
Com(A), the profile of B blocks the characteristic response value P of clue
Com(B), the public boundary between A and the B blocks the characteristic response value P of clue
Conv<a, B>, the public point of interface between A and the B blocks the characteristic response value P of clue
Ang<a, B>Constitute.
Positive sample and negative sample at training dataset up-sampling hiding relation; Hiding relation has transitivity; For example A blocks B, and B blocks C, and A has also blocked C (we suppose that each semantic object has only a hierarchical information here so; Do not exist an object situation of many levels to be arranged, the situation that does not yet exist object to block each other).So for hiding relation < A, B>and < B, A>arbitrarily, their represented hiding relation is relative, its characteristic of correspondence vector is also inequality.Therefore, on whole data set, the positive sample size of hiding relation is consistent with negative sample quantity, and through sampling, we have obtained the sample data of training hiding relation sorter, and then training hiding relation sorter.
For any input picture, detect the various clues of blocking of test data set image, described in testing process such as the step 1 then.According to semantic classes with block hint information; Structure belongs to the hiding relation proper vector set of this image; Utilize the sorter trained, obtain belonging to the predicted value of all hiding relations of this image, make up the complete digraph of cum rights of performance hiding relation with this.
The semantic marker figure that given image is corresponding; The present invention is blocked clue and response thereof according to all that obtain this image described in the step 1; Extract of the input of hiding relation proper vector then, just can obtain the PREF predicted value of all hiding relations of image as the sorter that obtains with training.Having obtained in the image can making up the complete digraph G=of cum rights < V, E>of hiding relation after the PREF value of all hiding relations, as shown in Figure 4.In the graph structure, the semantic object in each node correspondence image, the directed edge between the node is represented hiding relation, the weight on limit is the PREF value.Node R
1It is that 1.64778 directed edge points to node R that weight is arranged
2, can know R so
1Block R
2The preference score value be 1.64778.Here can exist a kind of situation of erroneous judgement, such as,<r
4, R
5>Be predicted as positive hiding relation by sorter, its preference score value is 0.315956,<r
5, R
4>Also sorter is predicted to be positive hiding relation, and its preference score value is 1.02053.Inconsistent situation of this hiding relation and the object that we set have only the hypothesis of a level not to be inconsistent.When hiding relation is inconsistent; Need to judge that which hiding relation is believable; In other words, need find that a near-optimization level is separated
and make it possible to farthest to satisfy hiding relations all in the image and the preference score value the highest.Target of the present invention is at W
LSolution space
In, seek and represent W at given image 2D
2DCondition under have the approximate optimal solution of maximum preference score value
Promptly
Wherein ρ is a hierarchical sequence of all objects of image, and P is the set of all possible hierarchical sequence, ρ
*Be optimum hierarchical sequence, in sequence ρ, object A has ρ (A)＜ρ (B) when the front of object B or A when blocking B.To any hierarchical sequence, (ρ PREF) satisfies ρ (A)＜ρ (B) for all to its preference score value AGREE.The PREF preference score value sum of the hiding relation of character.How we will find the solution optimum hierarchical structure and convert the hierarchical sequence of seeking maximum preference score value into like this.
Utilize the hierarchical ranking reasoning algorithm, on the complete digraph of cum rights, solve the hierarchical structure of image scene.Analyze the characteristics that the present invention dealt with problems, because the object in the image scene can be too not numerous and diverse, so the present invention has adopted a kind of full permutation algorithm of simply and fast and effectively enumerating.Its process is following: given graph structure G=<v, E>, wherein V is a vertex set, E is the limit set, obtains the full arrangement set P of all objects among the V; For each sequence among the P, its preference score value π (ρ)=∑
(A, B ∈ V) ∩ (A ≠ B)PREF (A, B); Make π (ρ
*)=argmax
ρ ∈ P(π (ρ)), ρ
*Be exactly the optimum hierarchical structure W that we will find the solution
L *The process flow diagram of reasoning process is as shown in Figure 5.
The user's interest object is joined the appointment level of layering scene, block, finally generate the image virtual scene according to the front and back level correct handling between object.
On the basis as a result of scene layering, the appointment level that the present invention realizes joining with interested object layering scene blocks according to the front and back level correct handling between object, finally generates the image virtual scene.As shown in Figure 6, the user utilizes the focus object method for distilling to extract objects-Niu, utilizes scene layered approach of the present invention to obtain the hierarchical structure (horse/meadow/trees) of another image scene; When ox was joined this scene, it was the 1st layer that the user specifies the level of ox, like this in the image virtual scene that generates; 0 layer of Ma Weidi; Ox is the 1st layer, and the meadow is the 2nd layer, and trees are the 3rd layer; And block according to the hierarchical relationship correct handling between object, finally obtain to merge the better image virtual scene.
The above is merely basic explanations more of the present invention, and any equivalent transformation according to technical scheme of the present invention is done all should belong to protection scope of the present invention.
Claims (8)
1. image scene layering and object occlusion handling method based on a sorter is characterized in that the method includes the steps of:
(1) on image scene semantic marker result's basis; Utilize object semanteme, object's position, object outline, public boundary and point of interface to block clue for five kinds and characterize hiding relation and hierarchical relationship between subject area; And above-mentioned five kinds of detecting the training dataset image block clue, calculate its characteristic response value;
(2) at training dataset up-sampling hiding relation sample, training obtains the sorter of hiding relation;
(3) for any input picture; Given its scene semantic marker result; Detect above-mentioned five kinds and block hint information, make up the set of hiding relation proper vector, utilize the sorter that has trained; Obtain belonging to the predicted value of all hiding relations of this image, make up the complete digraph of cum rights of performance hiding relation with this;
(4) hierarchical structure that obtains image scene is found the solution in reasoning on the complete digraph of the cum rights of hiding relation;
(5) the user's interest object is joined the appointment level of layering scene, block, finally generate the image virtual scene according to the front and back level correct handling between object.
2. a kind of image scene layering and object occlusion handling method according to claim 1 based on sorter, it is characterized in that: the sorter of the hiding relation in the step (2) obtains according to following steps:
(2.1) the positive sample and the negative sample of structure hiding relation;
(2.2) utilize five kinds of characteristic response values of blocking clue to make up the sextuple proper vector that shows hiding relation;
(2.3) in a plurality of positive sample of training dataset up-sampling and the proper vector of negative sample, the sorter of training hiding relation.
3. a kind of image scene layering and object occlusion handling method according to claim 2 based on sorter; It is characterized in that: the positive sample of hiding relation and the definition of negative sample are following; Semantic object A and B in the image; If A before the B or A block B, hiding relation < A, B>is positive sample so; Otherwise hiding relation < A, B>is a negative sample.
4. a kind of image scene layering and object occlusion handling method according to claim 2 based on sorter; It is characterized in that: for hiding relation < A; B >; The sextuple proper vector of described performance hiding relation; Be that profile that the profile of position that semanteme by A and B blocks characteristic response value, A and the B of the clue characteristic response value of blocking clue, A blocks the characteristic response value of clue, B blocks the characteristic response value that public boundary between characteristic response value, A and the B of clue blocks clue, the characteristic response value that the public point of interface between A and the B blocks clue constitutes.
5. a kind of image scene layering and object occlusion handling method according to claim 2 based on sorter; It is characterized in that: the sorter of described hiding relation, be meant a kind of preference function of having gathered a plurality of two-value indicator functions, be a kind of Adaboost sorter; To each hiding relation < A; B >, can both make up its sextuple proper vector, as the input of this sorter.
6. a kind of image scene layering and object occlusion handling method based on sorter according to claim 1 is characterized in that: the complete digraph of cum rights of the described hiding relation of step (3), and its construction step is following:
(3.1) proper vector of all hiding relations in the extraction image is utilized the sorter that trains, and predicts the character and the preference score value of each hiding relation;
(3.2) with the object in the image as the node in the graph structure, have directed edge, the hiding relation between the expressive object between the corresponding node of graph of object; Hiding relation has transitivity, if A blocks B, B blocks C, also exists hiding relation so between A and the C, and showing in the graph structure has a cum rights directed edge that points to C for A;
(3.3) weight on the digraph limit is the preference score value that sorter provides.
7. a kind of image scene layering and object occlusion handling method based on sorter according to claim 1 is characterized in that: the described reasoning of step (4) is meant the approximate optimal solution of on the cum rights digraph, seeking hierarchical sequence; Each hierarchical sequence all calculates a score, and the score soprano in all in the drawings hierarchical sequences is exactly the hierarchical structure optimum solution that we will find the solution.
8. a kind of image scene layering and object occlusion handling method according to claim 1 based on sorter; It is characterized in that: the described image scene of step (5) is blocked processing procedure; Be meant according to the image scene hierarchical relationship that obtains; When new object is put into image scene, the hiding relation between the correct handling object, thus obtain a correct image virtual scene.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110301899.7A CN102509119B (en) | 2011-09-30 | 2011-09-30 | Method for processing image scene hierarchy and object occlusion based on classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110301899.7A CN102509119B (en) | 2011-09-30 | 2011-09-30 | Method for processing image scene hierarchy and object occlusion based on classifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102509119A true CN102509119A (en) | 2012-06-20 |
| CN102509119B CN102509119B (en) | 2014-03-05 |
Family
ID=46221200
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110301899.7A Active CN102509119B (en) | 2011-09-30 | 2011-09-30 | Method for processing image scene hierarchy and object occlusion based on classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102509119B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930536A (en) * | 2012-10-16 | 2013-02-13 | 深圳先进技术研究院 | Indoor scene motility analysis and detection method based on hierarchical structure |
| US10089762B2 (en) | 2014-07-04 | 2018-10-02 | Mapillary Ab | Methods for navigating through a set of images |
| CN108764186A (en) * | 2018-06-01 | 2018-11-06 | 合肥工业大学 | Personage based on rotation deep learning blocks profile testing method |
| CN109002746A (en) * | 2017-06-06 | 2018-12-14 | 刘禹岐 | 3D solid fire identification method and system |
| CN109754009A (en) * | 2018-12-29 | 2019-05-14 | 北京沃东天骏信息技术有限公司 | Item identification method, device, vending system and storage medium |
| CN112288669A (en) * | 2020-11-08 | 2021-01-29 | 西北工业大学 | Point cloud map acquisition method based on light field imaging |
| CN112446246A (en) * | 2019-08-30 | 2021-03-05 | 初速度（苏州）科技有限公司 | Image occlusion detection method and vehicle-mounted terminal |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030058237A1 (en) * | 2001-09-27 | 2003-03-27 | Koninklijke Philips Electronics N.V. | Multi-layered background models for improved background-foreground segmentation |
| US20110044537A1 (en) * | 2009-08-18 | 2011-02-24 | Wesley Kenneth Cobb | Background model for complex and dynamic scenes |
| CN101986717A (en) * | 2010-11-11 | 2011-03-16 | 昆山龙腾光电有限公司 | Image data generating system for stereo display |
| CN102197415A (en) * | 2008-10-21 | 2011-09-21 | 皇家飞利浦电子股份有限公司 | Method and device for providing a layered depth model of a scene |
-
2011
- 2011-09-30 CN CN201110301899.7A patent/CN102509119B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030058237A1 (en) * | 2001-09-27 | 2003-03-27 | Koninklijke Philips Electronics N.V. | Multi-layered background models for improved background-foreground segmentation |
| CN102197415A (en) * | 2008-10-21 | 2011-09-21 | 皇家飞利浦电子股份有限公司 | Method and device for providing a layered depth model of a scene |
| US20110044537A1 (en) * | 2009-08-18 | 2011-02-24 | Wesley Kenneth Cobb | Background model for complex and dynamic scenes |
| CN101986717A (en) * | 2010-11-11 | 2011-03-16 | 昆山龙腾光电有限公司 | Image data generating system for stereo display |
Non-Patent Citations (1)
| Title |
|---|
| 王正光等: "基于层次图变换的多Agent组织结构动态重组机制", 《软件学报》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930536A (en) * | 2012-10-16 | 2013-02-13 | 深圳先进技术研究院 | Indoor scene motility analysis and detection method based on hierarchical structure |
| US10089762B2 (en) | 2014-07-04 | 2018-10-02 | Mapillary Ab | Methods for navigating through a set of images |
| CN109002746A (en) * | 2017-06-06 | 2018-12-14 | 刘禹岐 | 3D solid fire identification method and system |
| CN109002746B (en) * | 2017-06-06 | 2021-12-17 | 刘禹岐 | 3D (three-dimensional) fire identification method and system |
| CN108764186A (en) * | 2018-06-01 | 2018-11-06 | 合肥工业大学 | Personage based on rotation deep learning blocks profile testing method |
| CN108764186B (en) * | 2018-06-01 | 2021-10-26 | 合肥工业大学 | Figure occlusion contour detection method based on rotation deep learning |
| CN109754009A (en) * | 2018-12-29 | 2019-05-14 | 北京沃东天骏信息技术有限公司 | Item identification method, device, vending system and storage medium |
| CN109754009B (en) * | 2018-12-29 | 2021-07-13 | 北京沃东天骏信息技术有限公司 | Article identification method, article identification device, vending system and storage medium |
| CN112446246A (en) * | 2019-08-30 | 2021-03-05 | 初速度（苏州）科技有限公司 | Image occlusion detection method and vehicle-mounted terminal |
| CN112446246B (en) * | 2019-08-30 | 2022-06-21 | 魔门塔(苏州)科技有限公司 | Image occlusion detection method and vehicle-mounted terminal |
| CN112288669A (en) * | 2020-11-08 | 2021-01-29 | 西北工业大学 | Point cloud map acquisition method based on light field imaging |
| CN112288669B (en) * | 2020-11-08 | 2024-01-19 | 西北工业大学 | Point cloud map acquisition method based on light field imaging |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102509119B (en) | 2014-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110837836B (en) | Semi-supervised semantic segmentation method based on maximized confidence | |
| CN110322446B (en) | Domain self-adaptive semantic segmentation method based on similarity space alignment | |
| CN102509119B (en) | Method for processing image scene hierarchy and object occlusion based on classifier | |
| CN102509105B (en) | Hierarchical processing method of image scene based on Bayesian inference | |
| Guo et al. | Classification of airborne laser scanning data using JointBoost | |
| US20170124415A1 (en) | Subcategory-aware convolutional neural networks for object detection | |
| CN105760886A (en) | Image scene multi-object segmentation method based on target identification and saliency detection | |
| Wang et al. | YOLOv3-MT: A YOLOv3 using multi-target tracking for vehicle visual detection | |
| CN108830171B (en) | Intelligent logistics warehouse guide line visual detection method based on deep learning | |
| CN111126202A (en) | Optical remote sensing image target detection method based on void feature pyramid network | |
| CN105528575B (en) | Sky detection method based on Context Reasoning | |
| CN103810299A (en) | Image retrieval method on basis of multi-feature fusion | |
| Zhong et al. | Class-specific object proposals re-ranking for object detection in automatic driving | |
| CN103577875A (en) | CAD (computer-aided design) people counting method based on FAST (features from accelerated segment test) | |
| JP5180922B2 (en) | Image search system and image search method | |
| Saleh et al. | Traffic signs recognition and distance estimation using a monocular camera | |
| Shang et al. | Robust unstructured road detection: the importance of contextual information | |
| CN102222234A (en) | Image object extraction method based on mean shift and K-means clustering technology | |
| CN102902976A (en) | Image scene classification method based on target and space relationship characteristics | |
| CN107392254A (en) | A kind of semantic segmentation method by combining the embedded structural map picture from pixel | |
| Payet et al. | Scene shape from texture of objects | |
| CN110930413A (en) | Image segmentation method based on weak supervision multi-core classification optimization merging | |
| Tao et al. | An adaptive frame selection network with enhanced dilated convolution for video smoke recognition | |
| Huang et al. | An object detection algorithm combining semantic and geometric information of the 3D point cloud | |
| CN116958873A (en) | Pedestrian tracking method, device, electronic equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |