CN106446870A - Human body contour feature extracting method and device - Google Patents
Human body contour feature extracting method and device Download PDFInfo
- Publication number
- CN106446870A CN106446870A CN201610916300.3A CN201610916300A CN106446870A CN 106446870 A CN106446870 A CN 106446870A CN 201610916300 A CN201610916300 A CN 201610916300A CN 106446870 A CN106446870 A CN 106446870A
- Authority
- CN
- China
- Prior art keywords
- human body
- intrinsic mode
- body contour
- mode function
- function image
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The invention discloses a human body contour feature extracting method and device, and the method comprises the following steps: 100, obtaining a source image, and decomposing the source image into a plurality of two-dimensional intrinsic mode function images through a two-dimensional empirical mode decomposition method; 200, building a two-dimensional intrinsic mode decomposition-multi-scale tree structure model to extract the human body contour features of the two-dimensional intrinsic mode function images. The method can accurately extract the human body contour features, improves the recognition capability of human body contour features, and achieves the accurate description, so as to meet the requirements of no-supervision body posture detection and behavior recognition of a reality scene and achieve the adaptive recognition. In the above analysis, the two-dimensional intrinsic mode decomposition method for decomposition for a plurality of times, and the two-dimensional intrinsic mode decomposition-multi-scale tree structure model is employed for extracting the features of a bottom layer. The features are accurate and have no noise pollution, and the method can meet the requirements of body contour feature extraction and recognition, simplifies the operation steps, and reduces the calculation time consumption.
Description
Technical field
The present invention relates to technical field of image processing, more particularly, to a kind of human body contour outline feature extracting method and device.
Background technology
Human body behavior analysiss and technology of identification are one of focuses in artificial intelligence study, promote the man-machine friendship of video in reality
Mutually, the development of the application such as intelligent video monitoring.From the point of view of correlational study achievement in recent years, the human body behavior self adaptation of view-based access control model
Identification, because the reason such as individual variation and external environment condition produces substantial amounts of variable, makes this problem of research become more difficult.Behavior
Recognition methodss, or directly or indirectly use bottom human body contour feature, it is direct mode that space-time characteristic extracts, the statistics of feature
Learning method is indirect mode.Or the application of feature is partial body's shape information is overall body shape information, wheel
The great importance of wide feature extraction.
Human bodys' response technology is divided into by two classes with low-level image feature combining classification method：The identification skill of template matching classification
Art and the technology of identification of statistical learning classification.Either still statistical learning method is required for for human body for template matching classification
The feature of study inward nature, that is, need to detect or follow the tracks of human body global or local information.
Therefore how accurately to extract human body contour outline feature, thus improving the identification ability of human body contour outline feature, accurately
Description image, to meet the detection of unsupervised human body attitude and the Activity recognition of reality scene, to reach self-adapting estimation, become
The problem of those skilled in the art's urgent need to resolve.
Content of the invention
Based on above-mentioned technical barrier, the present invention provides a kind of human body contour outline feature extracting method, can accurately extract human body
Contour feature, thus improving the identification ability of human body contour outline feature, accurately describes image, to meet the unsupervised of reality scene
Human body attitude detection and Activity recognition, to reach self-adapting estimation, the method is different from the technical scheme disclosed in prior art.
The human body contour outline feature extracting method that the present invention provides, the method comprising the steps of：
Step 100：Obtain source images, if source images are decomposed into by dried layer two-dimensional solid by two-dimensional empirical mode decomposition method have mould
State function image；
Step 200：Set up two-dimensional empirical mode decomposition-multi-resolution tree structural model to extract two-dimentional intrinsic mode function image
Human body contour outline feature.
Preferably, described step 100 is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；
For i-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
Preferably, described step 200 is specially：
Step 201：Take, extract n-th two-dimentional intrinsic mode function imageIndividual stabilizing area MSERs, counts respectively
Calculate stabilizing area MSERsThe oval barycenter of provincial characteristicss parameterCoordinate, oval length
AxleWith transverse orientation angle,；
Step 202：Take, whenThen enter step 203；Otherwise then enter step 206；
Step 203：Obtain theThe region of stabilizing area MSERs of individual two dimension intrinsic mode function imageFeature
Parameter, extracts stabilizing area MSERs of i-th two-dimentional intrinsic mode function image；
Step 204：In corresponding centroid positionStabilizing area MSERsThe oval matter of zoning characteristic parameter
The heartCoordinate, oval major and minor axisWith transverse orientation angle；
Step 205：In corresponding centroid position, by i-th two-dimentional j-th region of intrinsic mode function imageFeature
Parameter and theIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, if meeting condition, remembers
RecordAnd its provincial characteristicss parameter, enter step 202；If being unsatisfactory for condition, delete, enter step 202；
Step 206：ObtainAs the characteristic area extracting.
Preferably, all two-dimentional intrinsic mode function figure is realized by the method in watershed in described step 201 and step 203
The extraction of stabilizing area MSERs of picture.
Preferably, the oval barycenter of provincial characteristicss parameter in described step 201 and step 204Coordinate, oval major and minor axisWith transverse orientation angle, specially：
（6）
（7）
（8）
（9）
（10）
（11）
（12）
（13）
（14）
（15）
（16）
Wherein,ForRank two-dimensional geometry square,,Representative is intended to aspiring for stability region MSERs image-region
The set of pixel,Representative is intended to aspiring for stability region MSERs image,It is respectively second-order moment around mean matrixTwo eigenvalues.
Preferably, by i-th two-dimentional j-th region of intrinsic mode function image in described step 205Characteristic parameter
WithIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, and meets condition and is specially：
, and（17）
（18）
Wherein, d, the q range of choice according to the illumination of image and or dimensional properties setting.
Preferably, d takesRadian, q takes 50%.
Present invention also offers a kind of human body contour outline feature deriving means, including picture breakdown module and feature extraction mould
Block, wherein：
Picture breakdown module, for obtaining source images, if be decomposed into dried layer by two-dimensional empirical mode decomposition method by source images
Two-dimentional intrinsic mode function image, and if have mode function image to be sent to characteristic extracting module dried layer two-dimensional solid；
Described characteristic extracting module, for obtain two-dimentional intrinsic mode function image and by set up two-dimensional empirical mode decomposition-
Multi-resolution tree structural model is come the human body contour outline feature to extract.
Preferably, described image decomposing module is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；
For i-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
Preferably, described characteristic extracting module includes fitting unit, characteristic parameter comparing unit, characteristic area determination list
Unit, wherein：
Fitting unit, for extracting stabilizing area MSERs of two-dimentional intrinsic mode function image, calculates stabilizing area respectively
The oval barycenter of the provincial characteristicss parameter of MSERsCoordinate, oval major and minor axisAnd ellipse
Long axis direction angle,, m is the number of stabilizing area MSERs of n-th two-dimentional intrinsic mode function image, and
Stabilizing area MSERs and its provincial characteristicss parameter are sent to characteristic parameter comparing unit；
Characteristic parameter comparing unit, in corresponding centroid position, by i-th two-dimentional intrinsic mode function image jth
Individual regionCharacteristic parameter andIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts；
Characteristic area determining unit, for for determining according to characteristic parameter comparing unit comparative resultAs the spy extracting
Levy region.
The method of the present invention can accurately extract human body contour outline feature and device, thus improving the identification of human body contour outline feature
Ability, accurately describes image, to meet unsupervised human body attitude detection and the Activity recognition of reality scene, to reach self adaptation
Identification.Above analysis is decomposed using some sublevel two-dimensional empirical mode decomposition methods, application two-dimensional empirical mode decomposition-multiple dimensioned
Tree-structure Model extracts low-level image feature, the accurate noise-less pollution of feature, can reach human body contour outline feature extraction and the requirement of identification,
Simplify calculation step, decrease computing and take.
Brief description
Fig. 1 provides the FB(flow block) of the first human body contour outline feature extracting method embodiment for the present invention；
Fig. 2 is that 4 layers of two-dimensional empirical mode decomposition method decompose source images logic relation picture；
Fig. 3 is source images and 1 to 4 layer of two-dimentional intrinsic mode function image；
Fig. 4 provides the FB(flow block) of second human body contour outline feature extracting method embodiment for the present invention；
Fig. 5 is the process instance figure extracting human body contour outline feature by two-dimensional empirical mode decomposition-multi-resolution tree structural model；
Fig. 6 provides second human body contour outline feature extracting method to extract the figure of human body contour outline feature for source images and by the present invention
Picture；
Fig. 7 provides a kind of structured flowchart of human body contour outline feature deriving means for the present invention.
Specific embodiment
In order that those skilled in the art more fully understand technical scheme, below in conjunction with the accompanying drawings to the present invention
It is described in further detail.
Referring to Fig. 1, Fig. 1 provides the FB(flow block) of the first human body contour outline feature extracting method embodiment for the present invention.
The human body contour outline feature extracting method that the present invention provides, the method comprising the steps of：
Step 100：Obtain source images, if source images are decomposed into by dried layer two-dimensional solid by two-dimensional empirical mode decomposition method have mould
State function image；
Step 200：Set up two-dimensional empirical mode decomposition-multi-resolution tree structural model to extract two-dimentional intrinsic mode function image
Human body contour outline feature.
The method of the present invention can accurately extract human body contour outline feature, thus improving the identification ability of human body contour outline feature more
Accurately description image, to meet unsupervised human body attitude detection and the Activity recognition of reality scene, to reach self-adapting estimation.
Above analysis is decomposed using some sublevel two-dimensional empirical mode decomposition methods, application two-dimensional empirical mode decomposition-multi-resolution tree knot
Structure model extraction low-level image feature, the accurate noise-less pollution of feature, human body contour outline feature extraction and the requirement of identification can be reached, simplify
Calculation step, decreases computing and takes.
Referring to Fig. 2 and Fig. 3, Fig. 2 is that 4 layers of two-dimensional empirical mode decomposition method decompose source images logic relation picture, and Fig. 3 is source
Image and 1 to 4 layer of two-dimentional intrinsic mode function image.
Described step 100 is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；
For i-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
After i layer two-dimensional empirical mode decomposition method is decomposed, final source imagesScreening process can represent
For
(2)
Wherein,It is i-th two-dimentional intrinsic mode function image,It is to decompose through i layer
Trendgram picture afterwards, L is the maximum decomposition level number once sieving, and N is natural number.We only use two dimension natural mode of vibration therein
Function, because these functions describe the characteristic information of source images multiple instantaneous frequency point well.
Wherein derivation in detail is：
（3）
（4）
（5）.
Referring to Fig. 2, taking 4 layers of two-dimensional empirical mode decomposition method as a example, to be decomposed by two-dimensional empirical mode decomposition method
When.What each two-dimentional intrinsic mode function image represented is different time one scale feature compositions in primary signal, and residual value
What signal represented is the trend amount information in initial data, and the screening process of two-dimentional intrinsic mode function image is exactly constantly to use
Last screening surpluses deduct the local mean value obtaining according to this surplus, until reaching stop condition.
Referring to Fig. 3, with the increase of Decomposition order, the surface of image is more and more flat.It means that with Decomposition order
Increase, most radio-frequency component is stripped, and low-frequency component is retained, thus in this sense, two-dimensional empirical mould
The process that state decomposition method decomposes, that is, high frequency is removed, the process that low frequency is retained.Ensemble empirical mode decomposition method can make
Obtain image to be decomposed according to the height difference of frequency, and being incremented by with Decomposition order, the information of image can gradually decrease.
If investigation characteristic point, distant characteristic point can fade away, and the characteristic point with high intensity identification can be left
Next layer, the high region of degree of being contrasted, there is strong edge.The ground floor that the decomposition of two-dimensional empirical mode decomposition method obtainsHuman body contour outline and background have obvious separated region, andSeparated region ratioBecome apparent from,,Reach the strongest, human body contour outline is kept completely separate with background, has strong edge.Analysis reason is although source figure
As whole resolution is not high, but human body is larger with background difference.Sieved by two-dimensional empirical mode decomposition method frequency, successively put
Big this diversity.
Referring to Fig. 4 to Fig. 6, Fig. 4 provides the flow chart element of second human body contour outline feature extracting method embodiment for the present invention
Figure, Fig. 5 is the process instance figure extracting human body contour outline feature by two-dimensional empirical mode decomposition-multi-resolution tree structural model,
Fig. 6 provides second human body contour outline feature extracting method to extract the image of human body contour outline feature for source images and by the present invention.
The difference of the present embodiment and embodiment one be to teach in detail pass through in step 200 two-dimensional empirical mode decomposition-
The method to extract human body contour outline feature for the multi-resolution tree structural model.
Step 201：Take, extract n-th two-dimentional intrinsic mode function imageIndividual stabilizing area MSERs, point
Ji Suan not stabilizing area MSERsThe oval barycenter of provincial characteristicss parameterCoordinate, oval
Major and minor axisWith transverse orientation angle,；
Step 202：Take, whenThen enter step 203；Otherwise then enter step 206；
Step 203：Obtain theThe region of stabilizing area MSERs of individual two dimension intrinsic mode function imageFeature
Parameter, extracts stabilizing area MSERs of i-th two-dimentional intrinsic mode function image；
Step 204：In corresponding centroid positionStabilizing area MSERsThe oval matter of zoning characteristic parameter
The heartCoordinate, oval major and minor axisWith transverse orientation angle；
Step 205：In corresponding centroid position, by i-th two-dimentional j-th region of intrinsic mode function imageFeature
Parameter and theIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, if meeting condition, remembers
RecordAnd its provincial characteristicss parameter, enter step 202；If being unsatisfactory for condition, delete, enter step 202；
Step 206：ObtainAs the characteristic area extracting.
Preferably, all two-dimentional intrinsic mode function figure is realized by the method in watershed in described step 201 and step 203
The extraction of stabilizing area MSERs of picture.
Preferably, the oval barycenter of provincial characteristicss parameter in described step 201 and step 204Coordinate, oval major and minor axisWith transverse orientation angle, specially：
（6）
（7）
（8）
（9）
（10）
ForRank two-dimensional geometry square,,Represent and be intended to aspiring for stability region MSERs image-region pixel
Set, image MSERs is discrete binaryzation region, and in region, pixel value is 1, is 0 outside region.Therefore image-region
Rank two-dimensional geometry square can be calculated by following formula,Representative is intended to aspiring for stability region MSERs image
（11）
（12）
（13）
（14）
It is respectively second-order moment around mean matrixTwo eigenvalues：
（15）
（16）.
Preferably, by i-th two-dimentional j-th region of intrinsic mode function image in described step 205Characteristic parameter
WithIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, and meets condition and is specially：
, and（17）
（18）
Wherein, d, the q range of choice according to the illumination of image and or dimensional properties setting.
Preferably, d takesRadian, q takes 50%.
To extract 4 layers of two-dimentional intrinsic mode function image to set up two-dimensional empirical mode decomposition-multi-resolution tree structural model
Human body contour outline feature as a example carry out feature extraction.The 4 layers of decomposition of two-dimensional empirical mode decomposition method obtain two-dimentional natural mode of vibration letter
Number imageArrive, yardstick changes from coarse to fine.In Fig. 4, in initial gauges spaceDetection is stable
9, region：Arrive；Calculate the moment characteristics parameter in 9 regions respectively,Coordinate, oval length
Short axleWith transverse orientation angle；.Moment characteristics parameter is delivered toMiddle detection corresponds to
The stability region of position simultaneously calculates moment characteristics, withParameter compares stability region direction, major and minor axis ratio, meets condition
Region be,.It is similar,Parameter is delivered to, calculate and compare, the region meeting condition is.Parameter is delivered to, calculate and compare, the region meeting condition is.Feature extraction completes.
Referring to Fig. 6, human body contour outline feature extraction accurately, noise-less pollution, and details enriches it will therefore be readily appreciated that human body fork leg
Stand, substantially, handss vertically put lower limb side to bifurcation site, and body slightly leans forward, waist portions are every obvious, characteristic area indexing height.
Referring to Fig. 7, Fig. 7 provides a kind of structured flowchart of human body contour outline feature deriving means for the present invention.
The present invention provides a kind of human body contour outline feature deriving means, including picture breakdown module 10 and characteristic extracting module
20, wherein：
Source images, for obtaining source images, are decomposed into some by picture breakdown module 10 by two-dimensional empirical mode decomposition method
The two-dimentional intrinsic mode function image of layer, and if have mode function image to be sent to characteristic extracting module dried layer two-dimensional solid；
Described characteristic extracting module 20, for obtaining two-dimentional intrinsic mode function image and being divided by setting up two-dimensional empirical modal
Solution-multi-resolution tree structural model is come the human body contour outline feature to extract.
The human body contour outline feature that above-mentioned human body contour outline feature deriving means can be used for implementing above-described embodiment one offer carries
Take method.
Assembly of the invention can accurately extract human body contour outline feature, thus improving the identification ability of human body contour outline feature more
Accurately description image, to meet unsupervised human body attitude detection and the Activity recognition of reality scene, to reach self-adapting estimation.
Above analysis is decomposed using some sublevel two-dimensional empirical mode decomposition methods, application two-dimensional empirical mode decomposition-multi-resolution tree knot
Structure model extraction low-level image feature, the accurate noise-less pollution of feature, human body contour outline feature extraction and the requirement of identification can be reached, simplify
Calculation step, decreases computing and takes.
In further scheme, described image decomposing module 10 is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；
For i-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
Taking 4 layers of two-dimensional empirical mode decomposition method as a example, with the increase of Decomposition order, the surface of image is more and more flat
Smooth.It means that with the increase of Decomposition order, most radio-frequency component is stripped, and low-frequency component is retained, thus from
Say in this meaning, the process that two-dimensional empirical mode decomposition method is decomposed, that is, high frequency is removed, the mistake that low frequency is retained
Journey.Ensemble empirical mode decomposition method so that image is decomposed according to the height difference of frequency, and with Decomposition order
It is incremented by, the information of image can gradually decrease.If investigation characteristic point, distant characteristic point can fade away, and has height
The characteristic point of intensity identification can leave next layer, the high region of degree of being contrasted, and has strong edge.
In further scheme, described characteristic extracting module 20 includes fitting unit 21, characteristic parameter comparing unit
22nd, characteristic area determining unit 23, wherein：
Fitting unit 21, for extracting stabilizing area MSERs of two-dimentional intrinsic mode function image by the method in watershed,
Calculate the oval barycenter of the provincial characteristicss parameter of stabilizing area MSERs respectivelyCoordinate, oval length
Short axleWith transverse orientation angle,, m is the stabilisation of n-th two-dimentional intrinsic mode function image
The number of region MSERs, and stabilizing area MSERs and its provincial characteristicss parameter are sent to characteristic parameter comparing unit 22；
Characteristic parameter comparing unit 22, in corresponding centroid position, by i-th two-dimentional intrinsic mode function image the
J regionCharacteristic parameter andIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts；
Characteristic area determining unit 23, for determining according to characteristic parameter comparing unit 22 comparative resultAs the spy extracting
Levy region.
Wherein,（6）
（7）
（8）
（9）
（10）
ForRank two-dimensional geometry square,,Represent and be intended to aspiring for stability region MSERs image-region pixel
Set, image MSERs is discrete binaryzation region, and in region, pixel value is 1, is 0 outside region.Therefore image-region
Rank two-dimensional geometry square can be calculated by following formula,Representative is intended to aspiring for stability region MSERs image
（11）
（12）
（13）
（14）
It is respectively second-order moment around mean matrixTwo eigenvalues：
（15）
（16）
By i-th two-dimentional j-th region of intrinsic mode function image in characteristic parameter comparing unit 22Characteristic parameter andIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, and meets condition and is specially：
, and（17）
（18）
Above-mentioned human body contour outline feature deriving means can be used for implementing the human body contour outline feature extraction side that above-described embodiment two provides
Method.
To extract 4 layers of two-dimentional intrinsic mode function image to set up two-dimensional empirical mode decomposition-multi-resolution tree structural model
Human body contour outline feature as a example carry out feature extraction.Human body contour outline feature extraction is accurate, noise-less pollution, and details is enriched, special
Levy discrimination high.
Above a kind of human body contour outline feature extracting method provided by the present invention and device are described in detail.Herein
In apply specific case the principle of the present invention and embodiment be set forth, the explanation of above example is only intended to help
Assistant solves the core concept of the present invention.It should be pointed out that for those skilled in the art, without departing from this
On the premise of bright principle, the present invention can also be carried out with some improvement and modify, these improve and modification also falls into present invention power
In the protection domain that profit requires.
Claims (10)
1. a kind of human body contour outline feature extracting method is it is characterised in that the method comprising the steps of：
Step 100：Obtain source images, if source images are decomposed into by dried layer two-dimensional solid by two-dimensional empirical mode decomposition method have mould
State function image；
Step 200：Set up two-dimensional empirical mode decomposition-multi-resolution tree structural model to extract two-dimentional intrinsic mode function image
Human body contour outline feature.
2. human body contour outline feature extracting method according to claim 1 is it is characterised in that described step 100 is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；For
I-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
3. human body contour outline feature extracting method according to claim 2 is it is characterised in that described step 200 is specially：
Step 201：Take, extract n-th two-dimentional intrinsic mode function imageIndividual stabilizing area MSERs, counts respectively
Calculate stabilizing area MSERsThe oval barycenter of provincial characteristicss parameterCoordinate, oval length
AxleWith transverse orientation angle,；
Step 202：Take, whenThen enter step 203；Otherwise then enter step 206；
Step 203：Obtain theThe region of stabilizing area MSERs of individual two dimension intrinsic mode function imageFeature is joined
Number, extracts stabilizing area MSERs of i-th two-dimentional intrinsic mode function image；
Step 204：In corresponding centroid positionStabilizing area MSERsThe oval matter of zoning characteristic parameter
The heartCoordinate, oval major and minor axisWith transverse orientation angle；
Step 205：In corresponding centroid position, by i-th two-dimentional j-th region of intrinsic mode function imageFeature
Parameter and theIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts, if meeting condition, remembers
RecordAnd its provincial characteristicss parameter, enter step 202；If being unsatisfactory for condition, delete, enter step 202；
Step 206：ObtainAs the characteristic area extracting.
4. human body contour outline feature extracting method according to claim 3 is it is characterised in that described step 201 and step 203
In all realize the extraction of stabilizing area MSERs of two-dimentional intrinsic mode function image by the method in watershed.
5. the human body contour outline feature extracting method according to claim 3 or 4 is it is characterised in that described step 201 and step
The oval barycenter of provincial characteristicss parameter in 204Coordinate, oval major and minor axisLong with ellipse
Direction of principal axis angle, specially：
（6）
（7）
（8）
（9）
（10）
（11）
（12）
（13）
（14）
（15）
（16）
Wherein,ForRank two-dimensional geometry square,,Representative is intended to aspiring for stability region MSERs image-region
The set of pixel,Representative is intended to aspiring for stability region MSERs image,It is respectively second-order moment around mean matrixTwo eigenvalues.
6. human body contour outline feature extracting method according to claim 5 is it is characterised in that by i-th in described step 205
Two-dimentional j-th region of intrinsic mode function imageCharacteristic parameter andIndividual j-th of intrinsic mode function image of two dimension
RegionCharacteristic parameter contrasts, and meets condition and is specially：
, and（17）
（18）
Wherein, d, the q range of choice according to the illumination of image and or dimensional properties setting.
7. human body contour outline feature extracting method according to claim 6 is it is characterised in that d takesRadian, q takes 50%.
8. a kind of human body contour outline feature deriving means are it is characterised in that include picture breakdown module and characteristic extracting module, its
In：
Picture breakdown module, for obtaining source images, if be decomposed into dried layer by two-dimensional empirical mode decomposition method by source images
Two-dimentional intrinsic mode function image, and if have mode function image to be sent to characteristic extracting module dried layer two-dimensional solid；
Described characteristic extracting module, for obtain two-dimentional intrinsic mode function image and by set up two-dimensional empirical mode decomposition-
Multi-resolution tree structural model is come the human body contour outline feature to extract.
9. the human body contour outline feature deriving means according to claims 8 are it is characterised in that described image decomposing module
It is specially：
（1）
Wherein,For source images；It is to source imagesAverage for i time；For
I-th two-dimentional intrinsic mode function image,, n is predetermined Decomposition order.
10. the human body contour outline feature deriving means according to claims 9 are it is characterised in that described feature extraction mould
Block includes fitting unit, characteristic parameter comparing unit, characteristic area determining unit, wherein：
Fitting unit, for extracting stabilizing area MSERs of two-dimentional intrinsic mode function image, calculates stabilizing area respectively
The oval barycenter of the provincial characteristicss parameter of MSERsCoordinate, oval major and minor axisAnd ellipse
Long axis direction angle,, m is the number of stabilizing area MSERs of n-th two-dimentional intrinsic mode function image, and
Stabilizing area MSERs and its provincial characteristicss parameter are sent to characteristic parameter comparing unit,；
Characteristic parameter comparing unit, in corresponding centroid position, by j-th of i-th two-dimentional intrinsic mode function image
RegionCharacteristic parameter andIndividual two dimension j-th region of intrinsic mode function imageCharacteristic parameter contrasts；
Characteristic area determining unit, for determining according to characteristic parameter comparing unit comparative resultAs the characteristic area extracted
Domain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610916300.3A CN106446870B (en) | 2016-10-21 | 2016-10-21 | A kind of human body contour outline feature extracting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610916300.3A CN106446870B (en) | 2016-10-21 | 2016-10-21 | A kind of human body contour outline feature extracting method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106446870A true CN106446870A (en) | 2017-02-22 |
| CN106446870B CN106446870B (en) | 2019-11-26 |
Family
ID=58175606
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610916300.3A Active CN106446870B (en) | 2016-10-21 | 2016-10-21 | A kind of human body contour outline feature extracting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106446870B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107274395A (en) * | 2017-06-13 | 2017-10-20 | 电子科技大学 | A kind of bus gateway head of passenger detection method based on empirical mode decomposition |
| CN109657534A (en) * | 2018-10-30 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | The method, apparatus and electronic equipment analyzed human body in image |
| CN109934835A (en) * | 2019-01-25 | 2019-06-25 | 广西科技大学 | Profile testing method based on the adjacent connection of deeply network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102322822A (en) * | 2011-08-08 | 2012-01-18 | 西安交通大学 | Three-dimensional measurement method for triple-frequency color fringe projection |
| CN102542296A (en) * | 2012-01-10 | 2012-07-04 | 哈尔滨工业大学 | Method for extracting image characteristics by multivariate gray model-based bi-dimensional empirical mode decomposition |
| CN103065299A (en) * | 2012-12-22 | 2013-04-24 | 深圳先进技术研究院 | Ultrasonic image edge extraction method and ultrasonic image edge extraction device |
| CN103400114A (en) * | 2013-07-18 | 2013-11-20 | 上海交通大学 | Illumination normalization processing system aiming at face recognition |
-
2016
- 2016-10-21 CN CN201610916300.3A patent/CN106446870B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102322822A (en) * | 2011-08-08 | 2012-01-18 | 西安交通大学 | Three-dimensional measurement method for triple-frequency color fringe projection |
| CN102542296A (en) * | 2012-01-10 | 2012-07-04 | 哈尔滨工业大学 | Method for extracting image characteristics by multivariate gray model-based bi-dimensional empirical mode decomposition |
| CN103065299A (en) * | 2012-12-22 | 2013-04-24 | 深圳先进技术研究院 | Ultrasonic image edge extraction method and ultrasonic image edge extraction device |
| CN103400114A (en) * | 2013-07-18 | 2013-11-20 | 上海交通大学 | Illumination normalization processing system aiming at face recognition |
Non-Patent Citations (2)
| Title |
|---|
| 巨西诺等: "《一种改进的基于MSER算法的不变量提取算法》", 《贵州工业大学学报（自然科学版）》 * |
| 梁胤程: "《图像局部不变量特征描述方法研究》", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107274395A (en) * | 2017-06-13 | 2017-10-20 | 电子科技大学 | A kind of bus gateway head of passenger detection method based on empirical mode decomposition |
| CN107274395B (en) * | 2017-06-13 | 2020-12-29 | 电子科技大学 | Bus entrance and exit passenger head detection method based on empirical mode decomposition |
| CN109657534A (en) * | 2018-10-30 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | The method, apparatus and electronic equipment analyzed human body in image |
| CN109934835A (en) * | 2019-01-25 | 2019-06-25 | 广西科技大学 | Profile testing method based on the adjacent connection of deeply network |
| CN109934835B (en) * | 2019-01-25 | 2023-03-31 | 广西科技大学 | Contour detection method based on deep strengthening network adjacent connection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106446870B (en) | 2019-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103679158B (en) | Face authentication method and device | |
| CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
| Zhou et al. | Scene classification using a multi-resolution bag-of-features model | |
| CN105894047B (en) | A kind of face classification system based on three-dimensional data | |
| CN112734775B (en) | Image labeling, image semantic segmentation and model training methods and devices | |
| CN103810490B (en) | A kind of method and apparatus for the attribute for determining facial image | |
| CN108108751B (en) | Scene recognition method based on convolution multi-feature and deep random forest | |
| CN103593654B (en) | A kind of method and apparatus of Face detection | |
| CN104866829A (en) | Cross-age face verify method based on characteristic learning | |
| CN105956560A (en) | Vehicle model identification method based on pooling multi-scale depth convolution characteristics | |
| Zhang et al. | Semantic classification of high-resolution remote-sensing images based on mid-level features | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN103971091B (en) | Automatic plane number recognition method | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN110532970A (en) | Age-sex's property analysis method, system, equipment and the medium of face 2D image | |
| CN106778687A (en) | Method for viewing points detecting based on local evaluation and global optimization | |
| CN105469050B (en) | Video behavior recognition methods based on local space time's feature description and pyramid words tree | |
| CN106682641A (en) | Pedestrian identification method based on image with FHOG- LBPH feature | |
| CN103020614B (en) | Based on the human motion identification method that space-time interest points detects | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN109726619A (en) | A kind of convolutional neural networks face identification method and system based on parameter sharing | |
| CN108537181A (en) | A kind of gait recognition method based on the study of big spacing depth measure | |
| CN107122711A (en) | A kind of night vision video gait recognition method based on angle radial transformation and barycenter | |
| CN104050460B (en) | The pedestrian detection method of multiple features fusion | |
| CN109359549A (en) | A kind of pedestrian detection method based on mixed Gaussian and HOG_LBP |
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 |