CN107610126A - A kind of interactive image segmentation method based on local prior distribution - Google Patents
A kind of interactive image segmentation method based on local prior distribution Download PDFInfo
- Publication number
- CN107610126A CN107610126A CN201710768732.9A CN201710768732A CN107610126A CN 107610126 A CN107610126 A CN 107610126A CN 201710768732 A CN201710768732 A CN 201710768732A CN 107610126 A CN107610126 A CN 107610126A
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- segmentation
- segmentation result
- distribution
- 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.)
- Pending
Links
Abstract
A kind of interactive image segmentation method based on local prior distribution, comprises the following steps:1) image to be split is smoothed, removes noise;2) part for choosing cutting object is used as distribution priori to calculate the distributed intelligence of target and background, it is assumed that original image pixel set is I;3) graph model is established according to conventional images, the minimum value of energy function is adjusted using max-flow/minimal cut;4) the gradation of image figure good to primary segmentation carries out rim detection, removes extraneous region, the segmentation result finally given is further smoothed to obtain final segmentation result.The invention provides a kind of interactive image segmentation method based on local prior distribution, can reduce the time required for image segmentation, and improve the precision of complex scene segmentation result.
Description
Technical field
The invention belongs to technical field of image segmentation, more particularly to a kind of image partition method.
Background technology
Common, image is before carrying out registration and rebuilding, it is necessary to carry out Accurate Segmentation to image.But existing image point
A system has not yet been formed in segmentation method, causes various image segmentation algorithms various.Existing conventional image segmentation side
Method has Threshold segmentation, and region increases, rim detection, based on graph theory segmentation etc..A kind of and side based on graph theory of GrabCut methods
Method, this method are in the basic improved method of GraphCut partitioning algorithms, and the overall situation is tried to achieve most using the mode of iteration
Excellent result, but the shortcomings that sliced time length, complicated image segmentation effect difference be present in this algorithm.
Technological deficiency is existing for GrabCut:Sliced time is grown, and complicated image segmentation effect is poor.
The content of the invention
In order to overcome the shortcomings of that the sliced time of existing GrabCut methods is long, complicated image segmentation effect is poor, the present invention
A kind of interactive image segmentation method based on local prior distribution is provided, the time required for image segmentation can be reduced, and
Improve the precision of complex scene segmentation result.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of interactive image segmentation method based on local prior distribution, comprises the following steps:
1) image to be split is smoothed, removes noise;
2) part for choosing cutting object is used as distribution priori to calculate the distributed intelligence of target and background, it is assumed that original image
Pixel set is I, and it is as follows to choose calculating process:
2.1) cutting object topography S is chosen, situation is divided into two kinds:
2.1.1) for simple scenario, the part of cutting object is selected;
2.1.2) for complex scene, cutting object scope frame is first increased to reduce segmentation scope, this scope is U, then is selected
The part of cutting object is selected, arbitrary close-shaped selection can be used.
In summary two kinds of situations, the pixel in image I is divided into four class set:{ F, B, PF, PB }, wherein F are prospects
Pixel, B are background pixel points, and PF is possible foreground pixel point, and PB is possible background pixel point;Wherein for simple field
Two kinds of pixel PB and F, p in scape be present is any pixel;
For complex scene,
2.2) distribution of foreground and background is calculated using corresponding local message
Pp(Vp| O)=| | Vp-μO||Φ
Pp(Vp| B)=| | Vp-μB||Φ
Wherein, Pp(Vp| O) and Pp(Vp| B) distribution probability of target and background, V are belonging respectively to for any pixel ppIt is to work as
Preceding display p pixel value, μOIt is the average value of foreground pixel value sum, μBIt is the average value of background pixel value sum, Φ is selected
The distributed model selected;
3) graph model is established according to conventional images, the minimum value of energy function is adjusted using max-flow/minimal cut, process is such as
Under:
3.1) figure structure is first carried out, local message is tried to achieve to build figure using above-mentioned.Two kinds of side n- are included wherein in figure
Links and t-links.N-links is the side being connected between pixel and pixel, t-links be pixel respectively with target
The side of point S and T-phase even, local prior information obtained above determine the assignment on t-links sides;
3.2) minimum value of energy function is solved using max flow/min cut;
4) the gradation of image figure good to primary segmentation carries out rim detection, removes extraneous region, the segmentation to finally giving
As a result further smooth to obtain final segmentation result.
The process of the step 4) is as follows:
4.1) the girth C for calculating each closed curve judges whether the enclosed region retains;
if C>T then retain the region
Else deletes the region
Wherein T is threshold curve length;
4.2) segmentation result is carried out smoothly, making the image border of segmentation result smooth.Thus final segmentation knot is obtained
Fruit.
The present invention technical though be:This method is slow for GrabCut splitting speeds, and complex scene segmentation result difference is asked
It is proposed method is inscribed to improve.Noise information is removed by filtering first, then chooses and retains the important local message of cutting object, so
Selection data distribution model determines n-links value afterwards.Using max flow/min cut try to achieve energy function minimum value so as to
Preliminary segmentation result is obtained, while removes extraneous region further to improve the precision of segmentation result.Finally to final result
It is smoothed, obtains the segmentation result of edge smoothing.
Beneficial effects of the present invention are mainly manifested in:1. directly retain important local message 2. by choosing local message
Simplify data distribution model, reduce amount of calculation, improve splitting speed 3. and be directed to complex scene, segmentation result precision improves, simultaneously
Segmentation result is stable.
Brief description of the drawings
Fig. 1 is the flow chart of the interactive image segmentation method based on local prior distribution.
Fig. 2 is the idiographic flow and the schematic diagram of operation of the dividing method of simple scenario.
Fig. 3 is the idiographic flow and the schematic diagram of operation of the dividing method of complex scene.
Embodiment
The present invention is further described below in conjunction with the accompanying drawings.
Reference picture 1- Fig. 3, a kind of interactive image segmentation method based on local prior distribution, comprises the following steps:
1) image to be split to Fig. 2 (a), Fig. 3 (a) is smoothed, and removes noise.Medium filtering is utilized in this example
Carry out the smoothing processing of image.
2) part for choosing cutting object is used as distribution priori to calculate the distributed intelligence of target and background.Assuming that original image
Pixel set is I, and it is as follows to choose calculating process:
2.1) it can be a region to choose cutting object topography S, S, or multiple regions, use Subgraph
As choosing, in this example by taking rectangle frame as an example.Situation can be divided into two kinds by this situation:
2.1.1) for such as Fig. 2 (a) simple scenario, the part of cutting object is selected, such as Fig. 2 (b)
2.1.2) for such as Fig. 3 (a) complex scene, such as Fig. 3 (b), the scope U of cutting object is chosen, then choose segmentation
The part of object.
Pixel in image I can be now divided into four class set:{ F, B, PF, PB }, wherein F are foreground pixel points, and B is
Background pixel point, PF are possible foreground pixel points, and PB is possible background pixel point.Wherein for having two in simple scenario
Kind pixel PB and F.P is any pixel.
For complex scene,
2.2) distribution of foreground and background is calculated using corresponding local message
Pp(Vp| O)=| | Vp-μO||Φ
Pp(Vp| B)=| | Vp-μB||Φ
Wherein Pp(Vp| O) and Pp(Vp| B) distribution probability of target and background is belonging respectively to for any pixel p.VpIt is to work as
Preceding display p pixel value, μOIt is the average value of foreground pixel value sum, μBIt is the average value of background pixel value sum.Φ is selected
The distributed model selected, Gaussian Profile is selected to calculate distributed model.
3) graph model is established according to conventional images, the minimum value of energy function is adjusted using max-flow/minimal cut
3.1) figure structure is first carried out, local message is tried to achieve to build figure using above-mentioned.
3.2) minimum value of energy function is solved using max flow/min cut.
4) the gradation of image figure good to primary segmentation carries out rim detection, removes extraneous region, the segmentation to finally giving
As a result further smooth to obtain final segmentation result.
4.1) the girth C for calculating each closed curve judges whether the enclosed region retains.
if C>T then retain the region
Else deletes the region
Wherein T is threshold curve length.It is a value that can be adjusted that T is set into 50, T in this example.Fig. 2 (d) and Fig. 3
(d) edge detection graph after treating is shown.
4.2) medium filtering is carried out to segmentation result so that the image border of segmentation result is smooth.Thus obtain final
Segmentation result such as Fig. 2 (c) and Fig. 3 (c).The binary image of segmentation result is shown in Fig. 2 (e) and Fig. 3 (e).
Claims (2)
1. a kind of interactive image segmentation method based on local prior distribution, it is characterised in that comprise the following steps:
1) image to be split is smoothed, removes noise;
2) part for choosing cutting object is used as distribution priori to calculate the distributed intelligence of target and background, it is assumed that original image pixel
Collection is combined into I, and it is as follows to choose calculating process:
2.1) cutting object topography S is chosen, situation is divided into two kinds:
2.1.1) for simple scenario, the part of cutting object is selected;
2.1.2) for complex scene, cutting object scope frame is first increased to reduce segmentation scope, this scope is U, and reselection is divided
The part of object is cut, arbitrary close-shaped selection can be used;
In summary two kinds of situations, the pixel in image I is divided into four class set:{ F, B, PF, PB }, wherein F are foreground pixels
Point, B are background pixel points, and PF is possible foreground pixel point, and PB is possible background pixel point;Wherein in simple scenario
It is any pixel in the presence of two kinds of pixel PB and F, p;
For complex scene,
2.2) distribution of foreground and background is calculated using corresponding local message
Pp(Vp| O)=| | Vp-μO||Φ
Pp(Vp| B)=| | Vp-μB||Φ
Wherein, Pp(Vp| O) and Pp(Vp| B) distribution probability of target and background, V are belonging respectively to for any pixel ppIt is current aobvious
Show p pixel value, μOIt is the average value of foreground pixel value sum, μBIt is the average value of background pixel value sum, Φ is selected
Distributed model;
3) graph model is established according to conventional images, the minimum value of energy function is adjusted using max-flow/minimal cut, process is as follows:
3.1) first carry out figure structure, local message tried to achieve to build figure using above-mentioned, wherein in figure comprising two kinds of side n-links and
T-links, n-links are the sides being connected between pixel and pixel, t-links be pixel respectively with target point S and T-phase
Side even, local prior information obtained above determine the assignment on t-links sides;
3.2) minimum value of energy function is solved using max flow/min cut;
4) the gradation of image figure good to primary segmentation carries out rim detection, extraneous region is removed, to the segmentation result finally given
Further smooth to obtain final segmentation result.
2. the interactive image segmentation method as claimed in claim 1 based on local prior distribution, it is characterised in that the step
Rapid process 4) is as follows:
4.1) the girth C for calculating each closed curve judges whether the enclosed region retains;
if C>T then retain the region
Else deletes the region
Wherein T is threshold curve length;
4.2) segmentation result is carried out smoothly, to make the image border of segmentation result smooth, thus obtain final segmentation result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710768732.9A CN107610126A (en) | 2017-08-31 | 2017-08-31 | A kind of interactive image segmentation method based on local prior distribution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710768732.9A CN107610126A (en) | 2017-08-31 | 2017-08-31 | A kind of interactive image segmentation method based on local prior distribution |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107610126A true CN107610126A (en) | 2018-01-19 |
Family
ID=61056696
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710768732.9A Pending CN107610126A (en) | 2017-08-31 | 2017-08-31 | A kind of interactive image segmentation method based on local prior distribution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107610126A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109559316A (en) * | 2018-10-09 | 2019-04-02 | 浙江工业大学 | A kind of improved graph theory dividing method based on super-pixel |
CN110378359A (en) * | 2018-07-06 | 2019-10-25 | 北京京东尚科信息技术有限公司 | A kind of image-recognizing method and device |
WO2023030182A1 (en) * | 2021-08-30 | 2023-03-09 | 华为技术有限公司 | Image generation method and apparatus |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592268A (en) * | 2012-01-06 | 2012-07-18 | 清华大学深圳研究生院 | Method for segmenting foreground image |
CN103632361A (en) * | 2012-08-20 | 2014-03-12 | 阿里巴巴集团控股有限公司 | An image segmentation method and a system |
CN104899851A (en) * | 2014-03-03 | 2015-09-09 | 天津医科大学 | Lung nodule image segmentation method |
-
2017
- 2017-08-31 CN CN201710768732.9A patent/CN107610126A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592268A (en) * | 2012-01-06 | 2012-07-18 | 清华大学深圳研究生院 | Method for segmenting foreground image |
CN103632361A (en) * | 2012-08-20 | 2014-03-12 | 阿里巴巴集团控股有限公司 | An image segmentation method and a system |
CN104899851A (en) * | 2014-03-03 | 2015-09-09 | 天津医科大学 | Lung nodule image segmentation method |
Non-Patent Citations (2)
Title |
---|
QIU GUAN等: "A MODIFIED GRABCUT APPROACH FOR IMAGE SEGMENTATION BASED ON LOCAL PRIOR DISTRIBUTION", 《2017 INTERNATIONAL CONFERENCE ON WAVELET ANALYSIS AND PATTERN RECOGNITION(ICWAPR)》 * |
辛月兰: "基于Grabcut的图像目标提取", 《青海师范大学学报(自然科学版)》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110378359A (en) * | 2018-07-06 | 2019-10-25 | 北京京东尚科信息技术有限公司 | A kind of image-recognizing method and device |
CN110378359B (en) * | 2018-07-06 | 2021-11-05 | 北京京东尚科信息技术有限公司 | Image identification method and device |
CN109559316A (en) * | 2018-10-09 | 2019-04-02 | 浙江工业大学 | A kind of improved graph theory dividing method based on super-pixel |
WO2023030182A1 (en) * | 2021-08-30 | 2023-03-09 | 华为技术有限公司 | Image generation method and apparatus |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7609888B2 (en) | Separating a video object from a background of a video sequence | |
CN103559719B (en) | A kind of interactive image segmentation method | |
CN111709420B (en) | Text detection method, electronic device and computer readable medium | |
CN102968782B (en) | In a kind of coloured image, remarkable object takes method automatically | |
CN108537239B (en) | Method for detecting image saliency target | |
CN105184763B (en) | Image processing method and device | |
CN100539698C (en) | The video of interactive time-space unanimity is scratched drawing method in a kind of Digital Video Processing | |
CN103699900B (en) | Building horizontal vector profile automatic batch extracting method in satellite image | |
CN106846339A (en) | A kind of image detecting method and device | |
CN104715451B (en) | A kind of image seamless fusion method unanimously optimized based on color and transparency | |
CN108447068B (en) | Ternary diagram automatic generation method and foreground extraction method using ternary diagram | |
CN107610126A (en) | A kind of interactive image segmentation method based on local prior distribution | |
CN103295013A (en) | Pared area based single-image shadow detection method | |
CN108986185B (en) | Image data amplification method based on deep learning | |
CN104809709A (en) | Single-image self-adaptation defogging method based on domain transformation and weighted quadtree decomposition | |
CN111507334A (en) | Example segmentation method based on key points | |
CN104021566A (en) | GrabCut algorithm-based automatic segmentation method of tongue diagnosis image | |
CN109934843A (en) | A kind of real-time profile, which refines, scratches image space method and storage medium | |
CN105719275A (en) | Parallel combination image defect segmentation method | |
CN111414860A (en) | Real-time portrait tracking and segmenting method | |
CN103020953A (en) | Segmenting method of fingerprint image | |
CN107909597B (en) | Multi-scale MRF model image segmentation method with edge preservation function | |
Vu et al. | Graph cut segmentation of neuronal structures from transmission electron micrographs | |
CN104268845A (en) | Self-adaptive double local reinforcement method of extreme-value temperature difference short wave infrared image | |
CN112561919A (en) | Image segmentation method, device and computer readable storage medium |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180119 |
|
RJ01 | Rejection of invention patent application after publication |