CN104361613B - Scene video object method for extracting region in baking box based on level set - Google Patents
Scene video object method for extracting region in baking box based on level set Download PDFInfo
- Publication number
- CN104361613B CN104361613B CN201410669923.6A CN201410669923A CN104361613B CN 104361613 B CN104361613 B CN 104361613B CN 201410669923 A CN201410669923 A CN 201410669923A CN 104361613 B CN104361613 B CN 104361613B
- Authority
- CN
- China
- Prior art keywords
- mrow
- level set
- region
- subgraph
- baking box
- 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.)
- Expired - Fee Related
Links
Classifications
-
- 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
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses scene video object method for extracting region in the baking box based on level set, it is by the more consistent foreground area of target color and luster in scene in mask technology, improved Level Set Models and region threshold technology rapid extraction burner hearth, comprises the following steps that：(1) the culinary art video in baking box burner hearth is absorbed, and set acquisition interval to obtain to need video frame images to be processed；(2) circle selects interest region from first two field picture, spanning subgraph segmentation mask, obtains subgraph；(3) cut and form subgraph；(4) initial profile curve is set；(5) level set movements；(6) region threshold is handled；(7) Objective extraction；(8) latter two field picture, repeat step (3)-(7) are read in, until all two field pictures complete Objective extraction.This method can both obtain clearly object boundary, and it only needs iteration 6 12 times to a width hardwood image, implements the ultrahigh in efficiency of on-line checking.
Description
Technical field
The invention belongs to scene field of video processing in baking box, and in particular to target area extraction side in video frame images
Method, available for the roasting Objective extraction for processing food in omnipotent steaming and baking box.
Background technology
, it is necessary to identify the food region in baking box in scene during omnipotent steaming and baking box is scorched to food, with
The automatic detection of Color Quality is convenient for, ensures the quality scorched, still, automatically extracts in baking box target area in scene
Technology is difficult to.Although level set movements curve possesses target area function [the Li C et for automatically extracting free boundary shape
al.Distance regularized level set evolution and its application to image
segmentation[J].Image Processing,IEEE Transactions on.2010,19(12):3243-3254],
But when meeting with the reduction object boundary in baking box in the video image of scene, easily evolution is caused to be crossed the border；And each hardwood image
The iterations of level set movements needs more than 200 times unexpectedly, if implementing on-line checking, efficiency is very low.
The content of the invention
Problem to be solved by this invention is to provide one kind, and more accurately and rapidly scene video object region carries in baking box
Method is taken, it can both obtain clearly object boundary, and it only needs iteration 6-12 times to a width hardwood image, implement online inspection
The ultrahigh in efficiency of survey.
Scene video object method for extracting region in baking box of the invention based on level set, it is by mask technology, improves
Level Set Models and region threshold technology rapid extraction burner hearth in the more consistent foreground area of target color and luster in scene,
Comprise the following steps that：
(1) the culinary art video in baking box burner hearth is absorbed, and set acquisition interval to obtain to need video frame images to be processed；
(2) circle selects interest region from first two field picture, spanning subgraph segmentation mask, obtains subgraph：In frame sequence, read
Enter first two field picture, and the interest region of target is manually included in circle choosing thereon, the maximum boundary rectangle in the region is solved, according to this
Rectangle builds the dividing sub-picture mask of test image；
(3) cut and form subgraph：Two field picture is tentatively cut using dividing sub-picture mask, obtains subgraph；
(4) initial profile curve is set：By computer in the target area of subgraph one is set according to default ratio
Individual rectangle frame is used for level set movements as initial profile curve；
(5) level set movements：Developed using the partial differential equation guiding initial curve of improved Level Set Models, drilled
Change obtains target area profile after terminating；Partial differential equation are
(6) region threshold is handled：The small closed curve of noise that level set movements are formed is handled using region threshold, filtered off
Noise closed curve, so as to obtain more complete target area profile；
(7) Objective extraction：According to the contour curve obtained in step 6, curvilinear inner target is extracted, that is, obtains target prospect
Image；
(8) latter two field picture, repeat step (3)-(7) are read in, until all two field pictures complete Objective extraction.
In view of the characteristic of experimental image, based on the present invention is using the mean information and contrast information of image, structure
New level set movements model, i.e. adaptive equalization level set movements model (ABLSE), below to improved level set movements
Model is described further.
The energy function of the model is as follows：
WhereinIt is as follows for the local energy item after improvement, its calculation formula：
Kσ(u) it is gaussian kernel function, is represented by：
θ (x) is window global contrast, and it takes neighborhood of the window of r × r sizes as point x centered on current point x
Scope, r value is relevant with the resolution ratio of image, and image resolution ratio is higher, and r values are bigger, this paper r=5, θ (x) expression formulas
For：
(4) M in formulamaxAnd MminThe maximum and minimum value of image intensity value in window area are represented respectively.θi(x) it is envelope
Closed curve interior contrast degree, θo(x) it is closed curve external contrast, the two construction is identical.Expression formula is：
θ in formulai(x) it is closed curve interior contrast degree, Mi_maxAnd Mi_minThe curvilinear inner image intensity value is represented respectively
Maximum and minimum value；θo(x) it is closed curve external contrast, Mo_maxAnd Mo_minCurved exterior image ash is represented respectively
The maximum and minimum value of angle value.
It is as follows for global energy item, its calculation formula：
ω is adaptation coefficient, and it can be according to the position of current curves, two energy term proportions of dynamic equilibrium.Ask
Solving expression formula is：
ω=abs (α+arctan (β)/π) (7)
Wherein α is constant coefficient scope (α=0.3 herein) between 0 to 0.5, and β is local variance average, its expression formula
For：
β=average (var (Mr×r(φ))) (8)
Wherein Mr×r(φ) is the neighborhood of r × r sizes centered on the point on evolution curve, var (Mr×r(φ)) it is neighborhood
Variance yields.Formula (7) and (8) show, coefficient of balanceωSpan between 0 to 1, and can follower's evolution curve move
State updates, and so can preferably balance the ratio shared by two kinds of energy models.
L (C) and P (φ) is respectively curve C length energy term and punishment energy term, is defined as follows：
The minimum value of energy equation is solved using gradient descent method and the calculus of variations, obtains following partial differential equation：
Wherein, F1And F2It is by-local forces and global active force respectively：
The background and prospect f of local fit can be tried to achieve using the calculus of variations1,f2For：
c1,c2For evolution inside or outside of curve portion gray average, it is represented by：
H in formulaεFor Heaviside formula, it is defined as：
Its derived function represents as follows:
Wherein e1,e2For contrast energy term：
Using the partial differential equation tried to achieve, guiding initial profile curve approaches object boundary, completes to develop.
Beneficial effects of the present invention are：Method involved in the present invention can be selected according to operating personnel in first two field picture circle
Food target area interested, automatically generate cutting mask, complete comprising the subgraphs sequence including target, and can utilize and improve
Level Set Models be automatically performed extraction subimage sequence target prospect image, be advantageous to as in automatic detection target area
Color Quality lays the foundation.The another present invention is based on improved Level Set Models, only needs iteration 6-12 times just to a width hardwood image
Clearly object boundary can be obtained, will not both evolution is crossed the border, and implements the ultrahigh in efficiency of on-line checking.
Brief description of the drawings
Fig. 1 is the flow chart of the inventive method；
Fig. 2 is the mask schematic diagram of the present invention；
Fig. 3 is tiny area threshold value eradicating efficacy figure；
Fig. 4 is the target prospect image finally extracted.
Embodiment
In order that the object, technical solutions and advantages of the present invention are clearer, below in conjunction with specific embodiment, and with reference to attached
Figure, the invention will be further described.
It is to be based on burner hearth video figure that the present invention, which provides scene video object method for extracting region in the baking box based on level set,
As collection and computing device are implemented：Burner hearth video image acquisition is with computing device by camera, omnipotent steaming and baking box and computer
Equipment is connected, and calculates processing software by corresponding IMAQ, transmission and video image and formed；Burner hearth video image is adopted
Collect equipment for the collection of video image, transmission and store into computer, then processing software is calculated by video image and carries out phase
The algorithm process answered, and the target prospect image extracted by computer export.
The application environment of this embodiment is:Dell inspiron 5447PC,Intel(R)i5-4210U CPU,RAM:
6GB, tested using matlab2012b.The parameter lambda of model1、λ2、λ3、λ41 is all set to, parameter lambda5、λ610 are all set to, parameter
ν=0.001 × 2552, time step τ=0.1 of μ=1.0, σ=3.0, target is roast chicken.
The method flow diagram that the present invention is embodied is as shown in figure 1, step is as follows：
1st, roast chicken culinary art video acquisition and image sampling
Camera is fixed on the ad-hoc location of omnipotent steaming and baking box glass door, by baking box with 200 DEG C preheating 2 minutes it is laggard
Row culinary art, while roast chicken culinary art video is obtained, transmit and be stored in notebook computer.The audio-visual-materials of acquisition are adopted
Sample, sampling method are to obtain a video frame images at interval of 3 seconds.This example has selected 8 images to be tested altogether, by not
Each stage with culinary art enters line label to two field picture.
2nd, first two field picture is read in, and circle selects interest region thereon
First two field picture is read in using image processing algorithm, and manually thereon, circle selects roast chicken target as interest region, its
Operation is as shown in Fig. 2 (a).
3rd, spanning subgraph segmentation mask
The one piece of interest region for including roast chicken target selected according to technosphere in step 2, solves the maximum external of the region
Rectangle, as shown in Fig. 2 (b), the dividing sub-picture mask of this group of test image is built according to this rectangle.
4th, preliminary cutting and spanning subgraph
The subgraph as shown in Fig. 2 (c) is obtained according to segmentation mask in step 3, for level set movements.
The 5th, initial profile curve is set
By computer according to default ratio (area is the size of picture 1/8, and is located at picture centre) in the roasting of subgraph
One rectangle frame is set in chicken region as initial profile curve.
6th, level set movements
According to the partial differential equation of improved Level Set Models, guiding initial profile curve develops toward roast chicken border, this
Test iterations is 8 times (it is 6-12 times generally to set iterations).In the starting stage of developing, due to depositing for global energy item
The information of its meeting thorough search image, forming multiple zonules；In subsequent evolution, gradually merge " similar " region.When
When evolution curve is close to roast chicken border, local energy item and contrast energy term can guiding curve develop and edge details and weak
Borderline region, so as to obtain more preferable evolution result, the extraction result of evolution curve is as shown in Figure 3.
7th, region threshold is handled
There can be the small closed curve of partial noise due to partial differential equation, in evolution result, therefore in step 6
Obtained final evolution curve carries out region threshold processing, with default threshold value (threshold value that this example is set is 200) with making an uproar
The pixel summation in sound area domain is compared, and filters the noise region less than threshold value, reduces the error that zonule is brought, makes simultaneously
The image of extraction is more complete.
8th, Objective extraction
According to the result of gained in step 7, roast chicken foreground image is obtained, as shown in Figure 4.Repeat step 4-8, until this reality
All image zooming-outs of example are complete.
We select accuracy OIRemove evaluation and test extraction result, accuracy OIRepresent the target image that is extracted of context of methods and
The coincidence degree of the benchmark image taken by hand.Its calculation formula is as follows, and its value is higher, represents
The accuracy of Objective extraction is higher.
OI=| IF∩IG|/|IF∪IG| 100%
I in formulaFFor the real goal image taken by hand, IGFor segmentation gained target image.
It is as shown in table 1 to obtain extraction accuracy and the iterations of this example roast chicken image, average be 85.05 ±
1.15%, the needs of Color Quality quantizating index in follow-up zoning can be met.
Table 1
And the DRLSE models in background technology are to extraction accuracy and the iterations such as institute of table 2 of this example roast chicken image
Show：
Table 2
As can be seen from Table 1 and Table 2, the ABLSE models that the present invention is carried, its accuracy are higher than DRLSE models, and iteration
Number is far fewer than the model.
Therefore, extracting method of the invention can both ensure accurately to extract target, will not be because of the object boundary of reduction
And cause evolution to be crossed the border, iterations can be greatly reduced again, make On-line testing target efficiency high.
Claims (4)
1. scene video object method for extracting region in the baking box based on level set, pass through mask technology, improved level set mould
The more consistent foreground area of target color and luster in scene in type and region threshold technology rapid extraction burner hearth, specific steps are such as
Under：
(1) the culinary art video in baking box burner hearth is absorbed, and sets acquisition interval to obtain to need video frame images to be processed, between collection
It is divided into the 3-10 seconds；
(2) circle selects interest region from first two field picture, spanning subgraph segmentation mask, obtains subgraph：In frame sequence, read in first
Two field picture, and the interest region of target is manually included in circle choosing thereon, the maximum boundary rectangle in the region is solved, according to this rectangle
Build the dividing sub-picture mask of test image；
(3) cut and form subgraph：Two field picture is tentatively cut using dividing sub-picture mask, obtains subgraph；
(4) initial profile curve is set：One square is set in the target area of subgraph according to default ratio by computer
Shape frame is used for level set movements as initial profile curve；
(5) level set movements：Developed using the partial differential equation guiding initial curve of improved Level Set Models, develop knot
Target area profile is obtained after beam；Partial differential equation are
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>&phi;</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<mi>t</mi>
</mrow>
</mfrac>
<mo>=</mo>
<msub>
<mi>&delta;</mi>
<mi>&epsiv;</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>&phi;</mi>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<mi>&omega;</mi>
<mo>&CenterDot;</mo>
<msub>
<mi>F</mi>
<mn>1</mn>
</msub>
<mo>+</mo>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mi>&omega;</mi>
</mrow>
<mo>)</mo>
<mo>&CenterDot;</mo>
<msub>
<mi>F</mi>
<mn>2</mn>
</msub>
<mo>+</mo>
<mi>&mu;</mi>
<mo>&CenterDot;</mo>
<mi>d</mi>
<mi>i</mi>
<mi>v</mi>
<mo>(</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>&dtri;</mo>
<mi>&phi;</mi>
<mo>|</mo>
</mrow>
<mi>&phi;</mi>
</mfrac>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>v</mi>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<msup>
<mo>&dtri;</mo>
<mn>2</mn>
</msup>
<mi>&phi;</mi>
<mo>-</mo>
<mi>d</mi>
<mi>i</mi>
<mi>v</mi>
<mo>(</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>&dtri;</mo>
<mi>&phi;</mi>
<mo>|</mo>
</mrow>
<mi>&phi;</mi>
</mfrac>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Contain F in the partial differential equation1Energy term be local energy item, contain F2Energy term be global energy item, by from
Accommodation coefficient ω adjustments,For level set function φ gradient, ν is length energy term coefficient, and μ is punishment energy term system
Number：
Wherein, ω is adaptation coefficient, F1And F2It is by-local forces and global active force respectively：
(6) region threshold is handled：The small closed curve of noise that level set movements are formed is handled using region threshold, filters off noise
Closed curve, so as to obtain more complete target area profile；
(7) Objective extraction：According to the contour curve obtained in step (6), curvilinear inner target is extracted, that is, obtains target prospect figure
Picture；
(8) latter two field picture, repeat step (3)-(7) are read in, until all two field pictures complete Objective extraction.
2. extracting method according to claim 1, it is characterized in that：In step (4), due to the food in cooking stage baking box
It will not be moved, therefore, all subgraphs set the initial profile of the same level set movements, and the profile is arranged to all
Subgraph in a rectangle frame inside target area.
3. extracting method according to claim 1, it is characterized in that：When step (4) sets initial profile curve, by computer
It is that area is the size of picture 1/8 according to default ratio, and is located at picture centre.
4. extracting method according to claim 1, it is characterized in that：Step (6) is specially：With default threshold value and noise range
The pixel summation in domain is compared, and filters the noise region less than threshold value, reduces the error that zonule is brought, while make extraction
Image it is more complete.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410669923.6A CN104361613B (en) | 2014-11-20 | 2014-11-20 | Scene video object method for extracting region in baking box based on level set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410669923.6A CN104361613B (en) | 2014-11-20 | 2014-11-20 | Scene video object method for extracting region in baking box based on level set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104361613A CN104361613A (en) | 2015-02-18 |
| CN104361613B true CN104361613B (en) | 2017-11-21 |
Family
ID=52528871
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410669923.6A Expired - Fee Related CN104361613B (en) | 2014-11-20 | 2014-11-20 | Scene video object method for extracting region in baking box based on level set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104361613B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105574528B (en) * | 2015-12-15 | 2019-01-22 | 安徽工业大学 | It is a kind of that cell image segmentation method is adhered based on multiphase mutual exclusion level set |
| CN108171722B (en) * | 2017-12-26 | 2020-12-22 | 广东美的厨房电器制造有限公司 | Image extraction method and device and cooking utensil |
| CN108038863B (en) * | 2018-01-29 | 2021-02-19 | 歌尔科技有限公司 | Image segmentation method and device |
| CN110222720A (en) * | 2019-05-10 | 2019-09-10 | 九阳股份有限公司 | A kind of cooking equipment with short video acquisition function |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102651130A (en) * | 2012-03-30 | 2012-08-29 | 宋怡 | Level set image processing method |
| CN103093473A (en) * | 2013-01-25 | 2013-05-08 | 北京理工大学 | Multi-target picture segmentation based on level set |
| CN103403761A (en) * | 2011-02-10 | 2013-11-20 | 诺华美亚有限责任公司 | Level set function based image processing |
-
2014
- 2014-11-20 CN CN201410669923.6A patent/CN104361613B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103403761A (en) * | 2011-02-10 | 2013-11-20 | 诺华美亚有限责任公司 | Level set function based image processing |
| CN102651130A (en) * | 2012-03-30 | 2012-08-29 | 宋怡 | Level set image processing method |
| CN103093473A (en) * | 2013-01-25 | 2013-05-08 | 北京理工大学 | Multi-target picture segmentation based on level set |
Non-Patent Citations (2)
| Title |
|---|
| 万能蒸烤箱烤鸡的色泽一致性检测方法与分析;纪滨 等;《现代食品科技》;20140315;第30卷(第3期);第2.2节第1段 * |
| 基于变分水平集方法的图像分割和目标轮廓跟踪研究;史娜;《中国博土学位论文全文数据库 信息科技辑》;20140815(第8期);第32页第1段,第3.2-3.4节、图3.5,3.10,3.11 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104361613A (en) | 2015-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106910176B (en) | A kind of facial image based on deep learning removes occlusion method | |
| CN104361613B (en) | Scene video object method for extracting region in baking box based on level set | |
| CN103871053B (en) | Vision conspicuousness-based cloth flaw detection method | |
| CN104899866B (en) | A kind of intelligentized infrared small target detection method | |
| CN110414362A (en) | Electric power image data augmentation method based on production confrontation network | |
| CN105678767B (en) | A kind of cloth surface flaw detection method based on SoC Hardware/Software Collaborative Design | |
| CN104574353B (en) | The surface defect decision method of view-based access control model conspicuousness | |
| CN111724372A (en) | Method, terminal and storage medium for detecting cloth defects based on antagonistic neural network | |
| CN107644418B (en) | Optic disk detection method and system based on convolutional neural networks | |
| CN108960255A (en) | Conspicuousness fabric defect detection method based on color similarity and position aggregation | |
| CN109191421B (en) | Visual detection method for pits on circumferential surface of cylindrical lithium battery | |
| CN106370403A (en) | Instant frequency estimation method based on edge detection | |
| CN104199823B (en) | A kind of fabric defects dynamic testing method of view-based access control model data-driven | |
| CN107463954A (en) | A kind of template matches recognition methods for obscuring different spectrogram picture | |
| CN114399480A (en) | Method and device for detecting severity of vegetable leaf disease | |
| CN111080574A (en) | Fabric defect detection method based on information entropy and visual attention mechanism | |
| CN103472031A (en) | Navel orange sugar degree detection method based on hyper-spectral imaging technology | |
| CN109711389A (en) | A kind of milking sow posture conversion identification method based on Faster R-CNN and HMM | |
| CN105954202A (en) | Hyperspectral model transfer method of citrus canker | |
| CN109613023A (en) | A kind of fruit surface defect rapid detection method of regional luminance adaptively correcting | |
| CN107886539B (en) | High-precision gear visual detection method in industrial scene | |
| CN103942786B (en) | The self adaptation block objects detection method of unmanned plane visible ray and infrared image | |
| CN111161295A (en) | Background stripping method for dish image | |
| CN107341807A (en) | A kind of method for extracting tobacco leaf color digital expression characteristic value | |
| CN111178405A (en) | Similar object identification method fusing multiple neural networks |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201225 Address after: 210046 room 1302, building 16, shangchengjingjing Beiyuan, No.7, Yaojia Road, yaohuamen street, Qixia District, Nanjing City, Jiangsu Province Patentee after: Nanjing Kaiyou Intelligent Technology Co.,Ltd. Address before: 243000 No. 59 East Lake Road, Anhui, Ma'anshan Patentee before: ANHUI University OF TECHNOLOGY |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171121 Termination date: 20211120 |