CN110533593A - A kind of method of the accurate trimap of quick creation - Google Patents
A kind of method of the accurate trimap of quick creation Download PDFInfo
- Publication number
- CN110533593A CN110533593A CN201910801239.1A CN201910801239A CN110533593A CN 110533593 A CN110533593 A CN 110533593A CN 201910801239 A CN201910801239 A CN 201910801239A CN 110533593 A CN110533593 A CN 110533593A
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- node
- trimap
- considered
- 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
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000011218 segmentation Effects 0.000 claims abstract description 8
- 230000003993 interaction Effects 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000003706 image smoothing Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000000463 material Substances 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 239000003550 marker Substances 0.000 claims 1
- 230000000694 effects Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 241000219109 Citrullus Species 0.000 description 1
- 235000012828 Citrullus lanatus var citroides Nutrition 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20156—Automatic seed setting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention proposes the methods that one kind quickly creates accurate trimap, method includes the following steps: being smoothed using gradient sparse prior to original input picture, remove unnecessary grain details, the difference between prominent foreground and background;Divide smoothed image using SLIC super-pixel segmentation algorithm, each super-pixel is considered as node;Robust interaction, user is scribbled and is integrated with super-pixel, the super-pixel comprising mark information is considered as seed node, remaining node is considered as unlabelled data;The score that each super-pixel is calculated based on popular sequence carries out ranking to super-pixel according to score, similar super-pixel can have as close possible to score, according to the result given threshold of prevalence sequence to create trimap.The trimap of high quality can be generated in the present invention, to ensure that the accuracy of the alpha mask of matting algorithm estimation.
Description
Technical field
The present invention relates to a kind of image processing method more particularly to image matting methods.
Background technique
Natural image matting is the process for accurately estimating user-defined foreground object opacity.Matting technology
It has a wide range of applications in image and video editing field, has obtained the extensive concern of scholars.In image and video editing,
People accurately extract foreground object and merge it to render new scene with new background using alpha mask.With image
The development of processing technique, to matting technology, more stringent requirements are proposed.Need the more superior matting algorithm of performance with
And more accurate trimap.And manual creation trimap is undoubtedly a time-consuming and laborious work.Each pixel in image
It can be seen as the linear combination of foreground pixel and background pixel.Therefore, the mathematical model of matting problem can indicate are as follows:
Wherein, indicate pixeliThe opacity at place.F,BRespectively foreground and background pixel.If, then should
Pixel fully belongs to prospect;If, then the pixel fully belongs to background;Otherwise, which is mixed pixel.Due to unknown
The foreground pixel and background pixel in region are unknown, available 3 known quantities and 7 unknown quantitys, therefore matting problem
The problem of being a serious underconstrained.Matting algorithm is intended to estimate unknown alpha value using known pixel color.
The key task of matting problem is the alpha value of accurate estimation mixed pixel.General matting problem requires user and mentions
The complexity of matting problem is reduced for prior information, and prior information is generally provided in a manner of trimap.Trimap can
Input picture is labeled as prospect, background and zone of ignorance.Using trimap provide prior information, can preferably by
Pixel Information is transmitted to zone of ignorance from known region.In matting algorithm, it is generally recognized that the trimap of input picture is
Through what is provided.But the trimap of user's manual creation is generally relatively rough, and create a large amount of accurate trimap and be undoubtedly
The work of one complexity and time-consuming.
When estimating the alpha value of pixel, most of pixel can be divided simply matting algorithm in input picture
For foreground pixel or background pixel, and the mixed pixel estimated is needed only to occupy the sub-fraction of input picture.Matting algorithm
It is exactly the alpha value using known foreground/background Pixel Information estimation mixed pixel.Trimap is in estimation mixed pixel
It is played an important role when alpha value.Accurate trimap can reduce the degree of difficulty of matting problem and improve matting
The performance of algorithm.Trimap is more accurate, and the unknown pixel of required estimation is fewer, can be believed using more foreground/backgrounds
Breath estimates less unknown pixel.Ideal trimap only includes the mixed pixel of foreground object edges.But in real work
In, the almost impossible creation trimap accurate in this way of user, this needs a large amount of man-machine interactively, and result usually can not
It takes.When the edge of foreground object includes a large amount of tiny hair or foreground color and background color is closer to, manually
Creating accurate trimap is a job that is uninteresting and expending energy.Most of high performance matting algorithm requires
Trimap as prior information, the precision of trimap be influence matting result accuracy an important factor for one of.Levin etc.
People proposes spectrum matting algorithm, which does not need user and provide trimap as prior information, can oneself
The dynamic alpha mask that foreground object is estimated from input picture.But the algorithm may be generated in the complicated image of processing
The result of mistake.The present invention proposes a kind of method for quickly creating accurate trimap, to reduce the workload of user.
In order to reduce the workload of user, researcher proposes to use strokes as prior information.Based on strokes
Mode provide the prior information of foreground/background by simple handmarking's mode.This method is simple in structure, background color
Good result can be obtained in single image.But ideal result cannot be obtained in the image of distribution of color complexity.
The accuracy of the alpha mask of matting algorithm estimation based on strokes is estimated lower than the matting algorithm based on trimap
The alpha mask of meter.And the quality that the trimap of different degree of roughness will lead to matting result is not quite similar.Use SAD pairs
The trimap of different degree of roughness is quantitatively evaluated, and the assessment result for the alpha mask estimated using coarse trimap is to make
With twice of the assessment result of the alpha mask of accurate trimap estimation.Coarse trimap can reduce the estimation of matting algorithm
Alpha mask quality.
Matting problem can be considered as the process of a soft segmentation.Image segmentation is that input picture is divided into binary
Image, each pixel or belongs to prospect or belongs to background.And the soft cutting procedure of matting problem then defines user
Foreground and background between pixel be considered as mixed pixel, the then opacity of each mixed pixel of explication.
Summary of the invention
In order to create accurate trimap, the workload of user is reduced, the present invention provides a kind of quickly creation is accurate
The method of trimap, it can be improved in image matting estimation alpha mask accuracy, and can be applied to image and
In terms of the Digital Image Processing such as video editing.This method is based on popular sequence, assigns each super-pixel different score values, according to point
Value sequence determines the zone of ignorance of trimap.The interaction of user's robust can provide better label information for prevalence sequence.
Specific steps of the invention are as follows:
Step 1: utilizing gradient sparse prior, original input picture is smoothed, unnecessary grain details are removed, dash forward
Difference between foreground and background out;
Step 2: dividing smoothed image using SLIC super-pixel segmentation algorithm, each super-pixel is considered as node;
Step 3: user is scribbled and is integrated with super-pixel, the super-pixel comprising mark information is considered as seed by robust interaction
Node, remaining node are considered as unlabelled data;
Step 4: the score of each super-pixel is calculated based on popular sequence, and ranking is carried out to super-pixel according to score, it is similar super
Pixel can have as close possible to score, according to prevalence sequence result given threshold to create trimap.
In the step 1, input picture is smoothed first, it is smooth after image reduce popular sequence
Complexity is conducive to generate accurate trimap.
In the step 2, it is based on sharpening result, super-pixel segmentation is carried out to smoothed out image using SLIC algorithm, with
Super-pixel is that basic element obtains foreground and background query node.
In the step 3, based on super-pixel segmentation as a result, interactive strokes result is integrated with super-pixel.
Using popular sort algorithm to foreground node and background node sequencing, the prospect label or background label of each super-pixel are obtained
And prospect score ranking and background score ranking.The present invention uses the seed labeled as prospect as query node, remaining section
Point is considered as unlabelled data.Available instruction vector accordingly.
In the step 4, using the method for threshold value comparison, the foreground information for merging each super-pixel and background information are with life
At final trimap.For graph structure, the present invention is more likely to belong to same target in view of the adjacent super-pixel in part.Each
Node not only directly adjacent adjacent node connection, but also the node with those nodes sharing public boundaries adjacent thereto
Connection.It is connected with each other labeled as the node of prospect, is also connected with each other labeled as the node of background.The foreground node and label of label
Background node will not be connected with each other.The present invention is by strokes information integration into graph structure.Divided using SLIC algorithm and is inputted
Each super-pixel is considered as node by image.The graph structure that the present invention constructs considers the spatial relationship of super-pixel, what user provided
Information and picture material.
Detailed description of the invention
Fig. 1 is overall procedure schematic diagram of the invention;
Fig. 2 is that the present invention is based on the results after the sparse smooth input picture of gradient;
Fig. 3 is the popular ranking results based on scribble information and super-pixel in the present invention;
Fig. 4 is the effect contrast figure using the method for the present invention trimap created and the trimap of manual creation;
Fig. 5 is the comparison result of the trimap of manual creation and the trimap created using inventive method in test image.
Specific embodiment
The present invention provides the methods that one kind quickly creates accurate trimap.Fig. 1 is that a kind of quickly creation of the invention is accurate
The flow chart of trimap.Input picture is smoothed by the present invention first, it is smooth after image reduce popular sequence
Complexity, be conducive to generate accurate trimap.Then super-pixel segmentation is carried out to smoothed out image using SLIC algorithm,
Foreground and background query node is obtained by basic element of super-pixel, interactive strokes result and super-pixel are integrated in one
It rises.Secondly, using popular sort algorithm to foreground node and background node sequencing, obtain each super-pixel prospect label or
Background label and prospect score ranking and background score ranking.Finally, merging each super-pixel using the method for threshold value comparison
Foreground information and background information to generate final trimap.
Fig. 2 illustrates the result after image smoothing.(a) (c) (d) is original input picture, (b) (d) in Fig. 2
(f) be it is smooth after image.The present invention is based on the details of gradient sparse prior smoothed image.After image smoothing, it can subtract
Few a large amount of unnecessary detail textures, and foreground object can be protruded, so that the difference between foreground and background is bigger, be conducive to
Popular sort algorithm assigns the boundary that foreground and background is more clear.
In specific implementation, the present invention is smoothed image using gradient, filters out the unessential details in part, and increase
The contrast of strong foreground and background.Objective function can be with is defined as:
WhereinLImage after indicating smooth,IFor input picture.Indicate sparse punishment,xIt is gradient value,
ParameterkIt is a fixed constant,。iIt is pixel index,,Indicate different filters.This
Invention uses first derivative filter and second order Laplace filter.First derivative filter can restoreLIn important side
Edge, and Laplace filter can encode smooth change therein.ParameterControlLSmoothness.
Due to objective function be it is non-convex, the present invention is based on half secondary separating methods to solve the non-of not inequality constraints
Convex problem.And in each iteration order, the present invention, which executes normalization step, forces solution to fall into restriction range.This hair
It is bright that auxiliary variable is introduced at each pixel, to establish new cost function:
WhereinIt is present invention weight increased during optimization.WithIncrease, solution is closer to equation (2).Pass through
It calculatesLAnd updateAlternately, it minimizes and fixesCost function.It updatesAnd it keepsLIt is constant, find each pixel
Closing form solution to minimize cost function.It can indicate are as follows:
It calculatesLAnd it willIt is fixed.The present invention uses 2D FFTHBy convolution matrixDiagonalization can be found optimalL。
Wherein * indicates complex conjugate, parameterThe stability of algorithm, τ=10 can be increased-16.Finally, standardizationL, will minimize
Objective function:
WhereinIt is indicator function, makesOnly existWhen be equal to 1, andOnly exist
When be equal to 1, otherwise they are equal to 0.Simple gradient decline can be applied to this step.Accordingly, the present invention can obtain compared with
Good smoothed image, the contrast between foreground and background become larger.
Fig. 3 illustrates the result that popular sequence is carried out using smoothed image.After picture smooth treatment, the knot of prevalence sequence
Fruit is obviously improved.There is significantly more difference before foreground and background.Fig. 3 (c) illustrates the sequence knot based on foreground information
Fruit, Fig. 3 (d) illustrate the ranking results based on background information.For adjacent super-pixel, the present invention can assign as far as possible to be compared
Close score.It is sorted based on prevalence, in conjunction with prospect ranking results and background ranking results, the present invention can be created accurately
trimap.The present invention assigns super-pixel element different label informations using strokes.When generating trimap, present invention design
The ordering rule of node.Based on the rule that the present invention designs, super-pixel information and strokes information can be made full use of with life
At ideal trimap.
In specific implementation, the basis of popular sort algorithm is structure figures as a result, then utilizing the popular immanent structure of data
Figure is marked.Give a data set X={ x1, x2..., xq, xq+1..., xn}∈Rm×n, which includes marked
Node and unlabelled node.Unmarked node is ranked up according to the similarity relationships of marked node.Each node
Ranking value fiBy ranking functions f:X → RnIt provides.Enable vector y=[y1, y2... yn]TIndicate the label situation of node.If yi
=1, then xiIt is query object;If yi=0, then xiIt is ordering joint.Graph structure can be expressed as G=(V, E), and wherein V is by counting
According to the node composition in collection X.E is by incidence matrix W=[wI, j]n×nIt determines.D=diag { d11, d12..., dnnBe figure degree square
Battle array, whereinThe ranking value of each node can be obtained by solving following optimization problem:
Wherein, the fitting constraint condition of the smoothness constraint of parameter μ control first item and Section 2.It is sorted by prevalence, so that adjacent
Close, the sequence score between similar node is as close as possible.Meanwhile to make the score of sequence score and seed point as far as possible
It is close.The optimal solution of formula (2) can be by obtaining being equal to 0 after objective function derivation.Finally, available popular sequence
Function are as follows:
f*=(D- α W)(-1)y (8)
The effect of popular sort algorithm is affected by query node, and query node setting is more accurate, and effect is better.In this hair
In bright, pass through simple strokes tag query node.It is marked based on strokes, available accurate query node, with
Generate accurate trimap.
The basis of popular sort algorithm is building graph structure, using super-pixel as figureGNode, between any two node
Weight are as follows:
WhereinIndicate Euclidean distance.Indicate the feature descriptor of node.The feature descriptor of each super-pixel by
Average color composition in average intensive SIFT descriptor and the space RGB, HSV and LAB.Indicate adjacent node.For control
Side right weight processed.Usually willIt is defined as fixed value.However, the optimal value of the parameter of different images is not necessarily identical, joined using fixed
The effect that numerical value obtains may be unsatisfactory.It willIt is defined as an auto-adaptive parameter:
Fig. 4 illustrate using the method for the present invention create trimap(Fig. 4 (a)) and manual creation trimap(Fig. 4 (b)).It can
Clearly to obtain, the trimap created using the method for the present invention is more accurate, and zone of ignorance almost only includes foreground object
The mixed pixel of marginal portion.And the trimap of manual creation is more coarse, the area of zone of ignorance is larger.What the present invention created
Trimap can accurately be bonded the edge of foreground object.
In specific implementation, the present invention uses the seed labeled as prospect, and as query node, remaining node is considered as unmarked
Data.Accordingly, available instruction vector.Node ranking score is calculated by cost function.These rank scores form N-dimensional
Vector.Wherein, N indicates the quantity of super-pixel.The similitude of corresponding node and prospect inquiry is saved in this vector.Before being based on
The rank score of scape label can indicate are as follows:
WhereiniIt is indexed for super-pixel,It isNormalized form.
It is inquired using the seed node labeled as background.It is similar with prospect inquiry, available expression background node
The vector of score.The rank score of respective nodes is calculated according to cost function, and is standardized.The rank score of background label
Is defined as:
After obtaining prospect rank score and background rank score, integrated in next stage.The present invention uses a kind of non-
It is often simple tactful:
WhereinIt indicates to press pixel product.Indicate the threshold processing operations based on average value.
WhereinIt isAverage value.WhenWhen, indicate that the part is prospect.WhenWhen, table
Show that the part is background.Remainder indicates zone of ignorance.Method according to the invention it is possible to generate the trimap of high quality.
Fig. 5 illustrates the partial test image that the present invention uses.The first row is test image in Fig. 5, and the second row is manual
The trimap of creation, last line are the trimap created using the method for the present invention.It is available from Fig. 5, use the present invention
The trimap of method creation is more more accurate than the trimap of manual creation.In Fig. 5 (d), at the scarf of bear, manual creation
Trimap is more coarse, and the edge of scarf is considered as an entirety, and can will be enclosed using the trimap that the method for the present invention creates
Towel is handled respectively, and the zone of ignorance of generation can preferably include the details of image.It can be with using the trimap that creates of the present invention
The alpha mask for estimating matting algorithm is more accurate.In image relatively simple for structure (Fig. 5 (e)), before watermelon
Scape and background structure are relatively simple, but its foreground color and background color are closer to.Although the trimap of manual creation
It is more accurate, but the trimap that the present invention creates is almost only comprising the edge of foreground object.And manual creation trimap is one
The extremely cumbersome work of item.
Claims (5)
1. the method that one kind quickly creates accurate trimap, which is characterized in that including the following steps:
Step 1: utilizing gradient sparse prior, original input picture is smoothed, unnecessary grain details are removed, dash forward
Difference between foreground and background out;
Step 2: dividing smoothed image using SLIC super-pixel segmentation algorithm, each super-pixel is considered as node;
Step 3: user is scribbled and is integrated with super-pixel, the super-pixel comprising mark information is considered as seed by robust interaction
Node, remaining node are considered as unlabelled data;
Step 4: the score of each super-pixel is calculated based on popular sequence, and ranking is carried out to super-pixel according to score, it is similar super
Pixel can have as close possible to score, according to prevalence sequence result given threshold to create trimap.
2. the method that one kind as described in claim 1 quickly creates accurate trimap, which is characterized in that the step (1) is flat
Sliding processing input picture, the present invention are smoothed image using gradient, are drawn using first derivative filter and second order general
Lars filter, first derivative filter can restore edge important in image, and Laplace filter can encode it
In smooth change, minimize objective function realize image smoothing:
WhereinIt is indicator function,Image after indicating smooth,For input picture;
Due to objective function be it is non-convex, the present invention is based on half secondary separating methods to ask to solve the non-convex of not inequality constraints
Topic, and in each iteration order, it executes normalization step and solution is forced to fall into restriction range, it will be under simple gradient
Drop is applied to this step, available preferable smoothed image.
3. the method that one kind as described in claim 1 quickly creates accurate trimap, which is characterized in that the step (2) is real
The super-pixel segmentation of existing smoothed image, the present invention is based on the super-pixel that SLIC algorithm obtains smoothed image, the present invention will each surpass
Pixel is considered as node, is the spatial relationship for considering super-pixel in the graph structure of the popular sequence of building, information that user provides and
Picture material.
4. the method that one kind as described in claim 1 quickly creates accurate trimap, which is characterized in that step (3) Shandong
Stick interaction, user is scribbled and is integrated with super-pixel, first by simply scribble label foreground and background, then will be applied
Crow integrates with super-pixel, and the prospect mark information that can be used as super-pixel of scribbling and context marker information, the present invention make
Using the seed labeled as prospect, remaining node is considered as unlabelled data as query node.
5. the method that one kind as described in claim 1 quickly creates accurate trimap, which is characterized in that step (4) base
It sorts in prevalence and creates trimap, give a data set, the number
According to comprising marked node and unlabelled node, unmarked node is arranged according to the similarity relationships of marked node
Sequence, the ranking value of each nodeBy ranking functionsIt provides, enables vectorIndicate node
Label situation, if, thenIt is query object;If, thenIt is ordering joint, graph structure can be expressed as, whereinBy data setIn node composition,EBy incidence matrixIt determines,It is the degree matrix of figure, wherein, asked by the optimization for solving following
Topic can obtain the ranking value of each node:
It is the degree matrix of figure, wherein, parameterControl first item
The fitting constraint condition of smoothness constraint and Section 2 uses the super-pixel for the prospect of being labeled as query node, remaining super-pixel
It is considered as unlabelled data, merges prospect Rank scores and background Rank scores create accurate trimap.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910801239.1A CN110533593B (en) | 2019-09-27 | 2019-09-27 | Method for quickly creating accurate trimap |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910801239.1A CN110533593B (en) | 2019-09-27 | 2019-09-27 | Method for quickly creating accurate trimap |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110533593A true CN110533593A (en) | 2019-12-03 |
CN110533593B CN110533593B (en) | 2023-04-11 |
Family
ID=68664772
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910801239.1A Active CN110533593B (en) | 2019-09-27 | 2019-09-27 | Method for quickly creating accurate trimap |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110533593B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177446A (en) * | 2013-03-13 | 2013-06-26 | 北京航空航天大学 | Image foreground matting method based on neighbourhood and non-neighbourhood smoothness prior |
JP2015041254A (en) * | 2013-08-22 | 2015-03-02 | 大日本印刷株式会社 | Trimap generation device, trimap generation method and program |
US20150117779A1 (en) * | 2013-10-30 | 2015-04-30 | Thomson Licensing | Method and apparatus for alpha matting |
CN104899877A (en) * | 2015-05-20 | 2015-09-09 | 中国科学院西安光学精密机械研究所 | Image foreground extraction method based on super-pixels and fast three-division graph |
CN107730528A (en) * | 2017-10-28 | 2018-02-23 | 天津大学 | A kind of interactive image segmentation and fusion method based on grabcut algorithms |
CN108846404A (en) * | 2018-06-25 | 2018-11-20 | 安徽大学 | A kind of image significance detection method and device based on the sequence of related constraint figure |
CN109087330A (en) * | 2018-06-08 | 2018-12-25 | 中国人民解放军军事科学院国防科技创新研究院 | It is a kind of based on by slightly to the moving target detecting method of smart image segmentation |
-
2019
- 2019-09-27 CN CN201910801239.1A patent/CN110533593B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177446A (en) * | 2013-03-13 | 2013-06-26 | 北京航空航天大学 | Image foreground matting method based on neighbourhood and non-neighbourhood smoothness prior |
JP2015041254A (en) * | 2013-08-22 | 2015-03-02 | 大日本印刷株式会社 | Trimap generation device, trimap generation method and program |
US20150117779A1 (en) * | 2013-10-30 | 2015-04-30 | Thomson Licensing | Method and apparatus for alpha matting |
CN104899877A (en) * | 2015-05-20 | 2015-09-09 | 中国科学院西安光学精密机械研究所 | Image foreground extraction method based on super-pixels and fast three-division graph |
CN107730528A (en) * | 2017-10-28 | 2018-02-23 | 天津大学 | A kind of interactive image segmentation and fusion method based on grabcut algorithms |
CN109087330A (en) * | 2018-06-08 | 2018-12-25 | 中国人民解放军军事科学院国防科技创新研究院 | It is a kind of based on by slightly to the moving target detecting method of smart image segmentation |
CN108846404A (en) * | 2018-06-25 | 2018-11-20 | 安徽大学 | A kind of image significance detection method and device based on the sequence of related constraint figure |
Non-Patent Citations (5)
Title |
---|
JINJIANG LI 等: "Generating Trimap for Image Matting Using Color Co-Fusion", 《IEEE ACCESS》 * |
JINJIANG LI 等: "Robust trimap generation based on manifold ranking", 《INFORMATION SCIENCES》 * |
向娅玲 等: "一种改进的基于KNN颜色线性模型抠图算法", 《电视技术》 * |
苑根基: "基于先验知识的matting算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
高智勇 等: "矩阵的低秩稀疏表达在视频目标分割中的研究", 《中南民族大学学报(自然科学版)》 * |
Also Published As
Publication number | Publication date |
---|---|
CN110533593B (en) | 2023-04-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Alameda-Pineda et al. | Recognizing emotions from abstract paintings using non-linear matrix completion | |
CN104732506B (en) | A kind of portrait photographs' Color Style conversion method based on face semantic analysis | |
CN107123088B (en) | A kind of method of automatic replacement photo background color | |
CN106778604B (en) | Pedestrian re-identification method based on matching convolutional neural network | |
CN104599275B (en) | The RGB-D scene understanding methods of imparametrization based on probability graph model | |
Choutas et al. | Accurate 3D body shape regression using metric and semantic attributes | |
CN102332034B (en) | Portrait picture retrieval method and device | |
TW201123081A (en) | Method and system for picture segmentation and method for image matting of a picture | |
WO2017181892A1 (en) | Foreground segmentation method and device | |
CN110569859B (en) | Color feature extraction method for clothing image | |
CN112365471B (en) | Cervical cancer cell intelligent detection method based on deep learning | |
CN108846404A (en) | A kind of image significance detection method and device based on the sequence of related constraint figure | |
CN115641583B (en) | Point cloud detection method, system and medium based on self-supervision and active learning | |
CN102542285B (en) | Image collection scene sorting method and image collection scene sorting device based on spectrogram analysis | |
Guo | Digital anti-aging in face images | |
CN110533593A (en) | A kind of method of the accurate trimap of quick creation | |
CN108765384B (en) | Significance detection method for joint manifold sequencing and improved convex hull | |
CN111161282A (en) | Target scale selection method for image multi-level segmentation based on depth seeds | |
Ju et al. | Stereo grabcut: Interactive and consistent object extraction for stereo images | |
CN115018729A (en) | White box image enhancement method for content | |
CN113239867A (en) | Mask region self-adaptive enhancement-based illumination change face recognition method | |
Kong et al. | SimLocator: robust locator of similar objects in images | |
Mary et al. | Content based image retrieval using colour, multi-dimensional texture and edge orientation | |
CN117392693B (en) | Method and equipment for removing handwriting of pathological image | |
Singh et al. | Automatic generation of trimap for image matting |
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 |