CN108629781A - A kind of hair method for drafting - Google Patents
A kind of hair method for drafting Download PDFInfo
- Publication number
- CN108629781A CN108629781A CN201810374586.6A CN201810374586A CN108629781A CN 108629781 A CN108629781 A CN 108629781A CN 201810374586 A CN201810374586 A CN 201810374586A CN 108629781 A CN108629781 A CN 108629781A
- Authority
- CN
- China
- Prior art keywords
- hair
- path
- gradient information
- tomographic image
- point
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/20—Drawing from basic elements, e.g. lines or circles
-
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of hair method for drafting, including：Hair zones segmentation is carried out to original image, obtains hair zones probability graph；The hair zones probability graph is single channel gray-scale map；The trend of hair is extracted from the original image；According to the trend of the hair, hair path is extracted in the hair zones probability graph；According to the hair path drawing hair.Technical solution provided by the invention can automatically and quickly carry out fine hair and draw on mobile terminals, meet user demand.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of hair method for drafting.
Background technology
In recent years, mobile phone terminal photography class application emerges one after another, these applications have the function of U.S. face, makeups etc., greatly rich
Rich people’s lives enjoyment, are greatly favored by consumers.In above-mentioned this kind of application, have a kind of specifically for portrait hair treatment
APP, to reach further U.S. face, makeups effect.But the existing method for portrait hair treatment it is all fairly simple or
It is that original hair is directly replaced using new hair material or is the color for simply replacing original hair, it is clear that is such
Processing method is unsatisfactory in U.S. face, makeups effect, cannot meet the needs of users.
The method for solving problem above is that need to finely be drawn as requested to the hair of original portrait.The prior art
In, also have and draws the relatively good algorithm of hair effect, for example,《Single-View Hair Modeling for Portrait
Manipulation》M Chai, L Wang, Y Weng, Y Yu, B Guo,《Acm Transactions on Graphics》,
2012,31 (4)：1-8, but the algorithm that this article is provided is complicated, the modeling time is long, and processing time calculates in minutes, uncomfortable
Conjunction uses on mobile terminal (such as in cell phone application).
Invention content
The present invention is intended to provide a kind of hair method for drafting, can automatically and quickly carry out fine on mobile terminals
Hair is drawn, and user demand is met.
In order to achieve the above objectives, the technical solution adopted by the present invention is as follows：
A kind of hair method for drafting, including：Hair zones segmentation is carried out to original image, obtains hair zones probability graph；
The hair zones probability graph is single channel gray-scale map；The trend of hair is extracted from the original image；According to the head
The trend of hair extracts hair path in the hair zones probability graph；According to the hair path drawing hair.
Preferably, the method for the trend that hair is extracted from the original image is：The original image is turned
It is changed to gray-scale map；Gaussian pyramid is established to the gray-scale map；Calculate the gradient of every tomographic image of the gaussian pyramid；To institute
The gradient for stating every tomographic image is modified, and obtains the gradient information figure per tomographic image；From top to bottom along the gaussian pyramid
The gradient information figure per tomographic image is successively merged in direction, obtains the gradient information figure after fusion；Melted according to described
Gradient information figure after conjunction obtains hair directional diagram and hair direction setting property degree figure；According to the hair directional diagram and hair side
To setting property degree figure, local smoothing method is carried out to the hair directional diagram, obtains the trend of hair.
Preferably, the gradient of every tomographic image of the gaussian pyramid is calculated using Sobel operators.
Preferably, described that the gradient per tomographic image is modified, obtain the side of the gradient information figure per tomographic image
Method is：Piecemeal is carried out to the gradient per tomographic image；Singular value decomposition is carried out to each block per tomographic image, is obtained every
Principal direction the setting property degree of the principal direction of a block and each block；Each piece of principal direction and principal direction setting property degree are formed into a two dimension
Vector is filled into corresponding piece of the bivector, obtains the gradient information figure per tomographic image.
Preferably, it is described along gaussian pyramid direction from top to bottom to the gradient information figure per tomographic image into
Row successively merges, and the method for obtaining the gradient information figure after fusion is：The gradient information figure per tomographic image adopt
Sample obtains the up-sampling gradient information figure per tomographic image；It is adopted on last layer image per the corresponding gradient information figure of tomographic image
Sample gradient information figure resolution ratio is identical；Along the direction of the gaussian pyramid from top to bottom, according to the up-sampling of tomographic image ladder
Hum pattern and gradient information figure are spent, the gradient information figure of adjacent next tomographic image is updated, until updating the gaussian pyramid
The gradient information figure of bottom layer image obtains the gradient information figure after fusion.
Preferably, the trend according to the hair extracts the side in hair path in the hair zones probability graph
Method includes：To each point in the hair zones probability graph, extraction path point is moved towards along the hair；By the path
The first hair path of point composition.
Further, further include：It carries out curve fitting to first hair path, obtains smooth hair path；Root
According to the smooth hair path drawing hair.
Preferably, described each point in the hair zones probability graph moves towards extraction path along the hair
Point method be：Since preset, along the positive direction extraction point of the trend of the hair, stops after reaching predetermined condition, obtain
Take positive direction path point；Since the preset, along the negative direction extraction point of the trend of the hair, reach the predetermined item
Stop after part, obtains negative direction path point；The positive direction path point and the negative direction path point are combined, institute is obtained
State the first hair path；The predetermined condition is：The point extracted is located at the outside of the hair zones probability graph；Alternatively, institute
The angle for stating the vector for walking upward two adjacent points of hair is more than 90 degree；Alternatively, the two adjacent points extracted is vertical
It is opposite to sit aiming symbol.
Preferably, hair zones segmentation is carried out to the original image using the method for deep learning.
Preferably, according to the hair path drawing hair by the way of picture point.
Hair method for drafting provided in an embodiment of the present invention, using gaussian pyramid on multiple dimensioned to the trend of hair into
Row extraction in this way can also efficiently extract the hair of hair texture weaker area trend, to promote hair drafting
Fineness；Meanwhile singular value decomposition (SVD) algorithm of not overlap partition is used to extract every layer of figure in hair moves towards extraction
As the principal direction of each block, significantly reduces gradient noise, improves arithmetic speed；In addition, the extraction in hair path can
Rough hair path is relatively quickly extracted, and is avoided that and is absorbed in hair circle endless loop；Hair path it is further
The power of fit procedure dynamic calculated curve achievees the purpose that the various hair forms of auto-adapted fitting, provided by the invention quasi-
Conjunction method is simple, it is efficient to calculate.As it can be seen that hair method for drafting provided by the invention, it can automatically and quickly on mobile terminals
It carries out fine hair to draw, meets user demand.
Description of the drawings
Fig. 1 is the method flow diagram of the embodiment of the present invention；
Fig. 2 is the schematic diagram to carry out curve fitting to path in the embodiment of the present invention.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with attached drawing, to the present invention into
Row is further described.
Step 101, hair zones segmentation is carried out to original image, obtains hair zones probability graph；The hair zones are general
Rate figure is single channel gray-scale map；
In the present embodiment, original image is a width RGB image Irgb, and set the width of the RGB image as width, it is a height of
height.Hair zones segmentation is carried out to the RGB image by the method for image segmentation, obtains hair zones probability graph
HairMask, HairMask are a width single channel gray-scale map.Image segmentation has been the technology of comparative maturity, and the present invention is using deep
The method for spending study carries out hair zones segmentation to above-mentioned original image.Partitioning algorithm based on deep learning can refer to hereafter：
Long,Jonathan,Evan Shelhamer,and Trevor Darrell.《Fully convolutional networks
for semantic segmentation》Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition.2015。
Step 102, the trend of hair is extracted from the original image；
This part mainly models the hair zones of extraction, extracts the specific trend of hair, the specific steps are：
A) by the original image IrgbBe converted to gray-scale map Igray；
B) to the gray-scale map IgrayEstablish gaussian pyramid；Gaussian pyramid by a series of different resolutions image group
At each layer of the image construction gaussian pyramid of these different resolutions.With PyrIiEach layer of gaussian pyramid is represented,
In the present embodiment, gaussian pyramid has 4 layers, then i=0, and 1,2,3, wherein PyrI0=Igray。
C) gradient of every tomographic image of the gaussian pyramid is calculated；In the present embodiment, using Sobel operators to Gauss
Pyramidal each layer of PyrIiIt calculates gradient and obtains GxyPyrIi, i=0,1,2,3.Wherein, GxyPyrIi(x, y) represents figure
As PyrIiGradient at point (x, y), the gradient are a bivector, GxyPyrIi(x, y) [0] represents at point (x, y)
The gradient of horizontal direction, GxyPyrIi(x, y) [1] represents the gradient of the vertical direction at point (x, y).
D) gradient per tomographic image is modified, obtains the gradient information figure per tomographic image；
In the present embodiment, since the hair texture strength of different hair zones differs, and due to illumination, the screening of hair level
The factors such as gear cause the gradient noise of obtained hair zones larger, cannot be directly used to the trend for judging hair, therefore, this
Inventive embodiments are by the way of section technique principal direction to the gradient G xyPyrI of every tomographic imageiIt is modified, specific amendment side
Method is as follows：
(1) to the gradient G xyPyrI per tomographic imageiCarry out piecemeal；If the width of block is Bw, a height of B of blockh, each block
It is formulated as：
BGxyPyrIi(j, k)={ GxyPyrIi(x,y)Bw* j≤x ＜ min (Bw*(j+1),width),Bh* k≤y ＜ min
(Bh*(k+1),height)}
Wherein, width indicates that the width of original image, height indicate that the height of original image, j are the serial number of horizontal direction, k
For the serial number of vertical direction, andCeil is the function that rounds up.For example, ceil
(x) smallest positive integral not less than x is indicated.
(2) singular value (Singular Value Decomposition, SVD) is carried out to each block per tomographic image
It decomposes, obtains the principal direction of each block and principal direction the setting property degree of each block；
In the present embodiment, to each piece of BGxyPyrIi(j, k) carries out SVD and decomposes to obtain the characteristic value S of blocki(j, k), and
This feature is worth corresponding feature vector Vi(j, k), wherein Si(j, k) [0] indicates first characteristic value, Si(j, k) [1] indicates the
Two characteristic values, and Si(j,k)[0]≥Si(j, k) [1], Vi(j, k) [0] be block principal direction (principal direction is defined as larger
The corresponding feature vector of one characteristic value), the principal direction V of calculation blockiThe setting property degree of (j, k) [0]
(3) each piece of principal direction and principal direction setting property degree are formed into a bivector, is filled into the bivector
In corresponding piece, the gradient information figure per tomographic image is obtained.
In the present embodiment, by Vi(j,k)[0],Ri(j, k) composition of vector, which is filled into block, obtains modified gradient information figure,
Gradient information figure i.e. per tomographic image
, wherein floor indicates downward bracket function, for example, floor (x) indicates the maximum integer no more than x, with VPyrIi(x,y)
[0] gradient at coordinate (x, y) is indicated, with VPyrIi(x, y) [1] indicates gradient the setting property degree at coordinate (x, y).In order to
Computational efficiency is improved, the embodiment of the present invention takes a height of B of the width of piecemealw=Bh=16, for example, GxyPyrIiResolution ratio be
640x640, then can be by GxyPyrIiIt is divided into 40x40 block.
E) the gradient information figure per tomographic image is successively melted along the direction of the gaussian pyramid from top to bottom
It closes, obtains the gradient information figure after fusion；
Specifically, according to i=3,2,1 sequence carries out following operation successively：
(1) to the gradient information figure VPyrI per tomographic imageiIt is up-sampled, obtains the up-sampling ladder per tomographic image
Spend hum pattern VPyrUpIi；Up-sampling gradient information figure resolution ratio per the corresponding gradient information figure of tomographic image and last layer image
It is identical, i.e. VPyrUpIiResolution ratio and VPyrIi-1Resolution ratio it is identical；
(2) direction along the gaussian pyramid from top to bottom, according to the up-sampling gradient information figure and ladder of a tomographic image
Hum pattern is spent, the gradient information figure of adjacent next tomographic image is updated, the ladder of the bottom layer image until updating the gaussian pyramid
Hum pattern is spent, the gradient information figure after fusion is obtained.
In the present embodiment, it is updated, is merged using following formula：
(2a) corrects the gradient of the gaussian pyramid adjacent layer：
VPyrIi(x, y) [0]=VPyrIi(x,y)[0]*sign(VPyrIi(x,y)[0]·VPyrUpIi(x, y) [0]),
Wherein, dot product is indicated, sign is sign function, is defined as
(2b) updates gradient information：
Calculate the weighting coefficient between pyramidal layer：
Linear weighted function is carried out to two layers of information of pyramid using weighting coefficient：
VPyrIi-1(x, y) [0]=(1-alpha (x, y)) * VPyrUpIi(x,y)[0]+alpha(x,y)*VPyrIi(x,
y)[0]
Then result is normalized to obtain unit vector：
The mode that this pyramid successively merges can take into account the gradient information of different scale, effectively restore hair texture compared with
The gradient of weak-strong test.
(2c) updates setting property degree information：
VPyrIi-1(x, y) [1]=0.5*VPyrUpIi(x,y)[1]+0.5*VPyrIi(x,y)[1]
F) according to the gradient information figure VPyrI after the fusion0, obtain hair directional diagram VhairWith hair direction setting property degree
Scheme Rhair；
Vhair(x, y)=⊥ VPyrI0(x,y)[0],
Rhair(x, y)=VPyrI0(x,y)[1]
Wherein, ⊥ operators expression takes vertical vector, for example,Indicate vectorVertical vector.
G) according to the hair directional diagram and hair direction setting property degree figure, local smoothing method is carried out to the hair directional diagram,
Obtain the trend of hair.
Vector in the range of i.e. pair radius is r is weighted averagely, and the hair direction at coordinate (x, y) is updated to：
Wherein, r is smooth radius, and 5 are taken in the present embodiment；The value range of m, n are [- r, r].
Step 103, V is moved towards according to the hairhair, head is extracted in the hair zones probability graph HairMask
Send out path；Specifically, to each point in the hair zones probability graph, with pre- fixed step size step along the trend of the hair
Extract path point；The path point is formed into the first hair path.Specific steps are described below in detail：
A) since preset, the positive direction along the trend of the hair reaches predetermined with pre- fixed step size step extraction points
Stop after condition, obtains positive direction path point；Since the preset, the negative direction along the trend of the hair is with predetermined step
Long step extraction points, stop after reaching the predetermined condition, obtain negative direction path point；By the positive direction path point and described
Negative direction path point is combined, and obtains first hair path；
The each position (x, y) for scanning hair area probability figure HairMask starts path with predetermined probability p and extracts：If
Current point is Li=(xi,yi), particularly, L0=(x0,y0)=(x, y), along Vhair(xi,yi) direction, using step as step-length
It moves forward and reaches Li+1=Li+Vhair(xi,yi), and so on, stop after reaching predetermined condition, if point when stopping is Ln；From
L0Start, along Vhair(x0,y0) negative direction, using step as step-length move forward, i.e. Li-1=Li-Vhair(xi,yi), reach pre-
Stop after fixed condition, if point when stopping is Lm；The path point of both forward and reverse directions is combined to form into a complete path L,
That is Lm, Lm+1, Lm+2..., L0,L1..., Ln.Wherein, " predetermined condition " refers to meeting one of following state：The point L extractediPosition
In the outside of the hair zones probability graph HairMask, it is maintained in hair zones probability graph for limiting hair path；Or
Person, the vectorial V for walking upward two adjacent points of the hairhair(xi,yi) and Vhair(xi+1,yi+1) angle be more than 90
Degree, the condition assume that the trend of hair will not be mutated, and the place of mutation can be the place that the hair of different levels crosses mostly；Or
Person, the two adjacent point L extractediWith Li+1Ordinate symbol on the contrary, the condition can avoid track algorithm enter " circle
Circle " endless loop.Wherein, Probability p be a value that can be arranged, to control extraction path density, ranging from [0,1] of p.
It should be noted that Probability p is for preset, it is exactly to give a preset, may moves towards to extract road along hair
Diameter, it is also possible to not do so, and leap to next preset；Along hair move towards extract path when, not with
Probability p carries out.
B) in order to keep the hair of drafting smooth as much as possible, need to carry out curve fitting to the path of extraction, to path into
The process of row curve matching is：
1) certain paths is set as L, n point is shared, with Li, i=0,1,2 ..., n-1 indicate the point on path, if waiting being fitted
The highest power Q of curve.If initial Q=3, L is begun stepping through from i=0iIf from straight line L in ergodic process0Ln-1Side advances to
The other side, then Q increases by 1, as shown in Figure 2.
2) position of each point on waiting for matched curve is calculated：
It is apparent that Tn-1It approximate can regard the total length in path as.
It is obtained after normalization：
3) it is fitted with polynomial curve, enables the polynomial curve equation beajFor two dimension to
Amount indicates that coefficient to be asked, j=0,1,2 ..., Q define energy function：
It minimizes E and acquires parameter aj。
C) smooth hair path is obtained
If the points in new hair path are m=α Tn-1, α ∈ [1,10], wherein α is bigger, then puts more, finally draws
Curve is more smooth, but point is more, and the calculation amount needed is also bigger, considers, and takes 1.5 herein, and smooth hair path is
Step 104, according to smooth hair path drawing hair.
In this step, according to smooth hair path Lnew, hair is drawn in the way of picture point:To LnewIn it is each
Point is in original image IrgbIn corresponding position draw point can form a smooth hair.All hairs are drawn in this approach.
Hair method for drafting provided in an embodiment of the present invention, using gaussian pyramid on multiple dimensioned to the trend of hair into
Row extraction in this way can also efficiently extract the hair of hair texture weaker area trend, to promote hair drafting
Fineness；Meanwhile singular value decomposition (SVD) algorithm of not overlap partition is used to extract every layer of figure in hair moves towards extraction
As the principal direction of each block, significantly reduces gradient noise, improves arithmetic speed；In addition, the extraction in hair path can
Rough hair path is relatively quickly extracted, and is avoided that and is absorbed in hair circle endless loop；Hair path it is further
The power of fit procedure dynamic calculated curve achievees the purpose that the various hair forms of auto-adapted fitting, provided by the invention quasi-
Conjunction method is simple, it is efficient to calculate.In conclusion hair method for drafting provided by the invention, it can be automatically and quickly mobile whole
Fine hair is carried out on end to draw, and meets user demand.Experiment shows technical solution provided by the invention, can be on mobile phone
The automatic drafting of hair is completed with the time less than 3 seconds.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.
Claims (10)
1. a kind of hair method for drafting, which is characterized in that including：
Hair zones segmentation is carried out to original image, obtains hair zones probability graph；The hair zones probability graph is single channel
Gray-scale map；
The trend of hair is extracted from the original image；
According to the trend of the hair, hair path is extracted in the hair zones probability graph；
According to the hair path drawing hair.
2. hair method for drafting according to claim 1, which is characterized in that described extracted from the original image is lifted one's head
The method of the trend of hair is：
The original image is converted into gray-scale map；
Gaussian pyramid is established to the gray-scale map；
Calculate the gradient of every tomographic image of the gaussian pyramid；
The gradient per tomographic image is modified, the gradient information figure per tomographic image is obtained；
The gradient information figure per tomographic image is successively merged along the direction of the gaussian pyramid from top to bottom, is obtained
Gradient information figure after fusion；
According to the gradient information figure after the fusion, hair directional diagram and hair direction setting property degree figure are obtained；
According to the hair directional diagram and hair direction setting property degree figure, local smoothing method is carried out to the hair directional diagram, obtains head
The trend of hair.
3. hair method for drafting according to claim 2, which is characterized in that calculate the Gauss gold using Sobel operators
The gradient of every tomographic image of word tower.
4. hair method for drafting according to claim 2, which is characterized in that described to be carried out to the gradient per tomographic image
It corrects, the method for obtaining the gradient information figure per tomographic image is：
Piecemeal is carried out to the gradient per tomographic image；
Singular value decomposition is carried out to each block per tomographic image, the principal direction of the principal direction and each block that obtain each block is set
Property degree；
Each piece of principal direction and principal direction setting property degree are formed into a bivector, are filled into corresponding piece of the bivector
It is interior, obtain the gradient information figure per tomographic image.
5. hair method for drafting according to claim 4, which is characterized in that it is described along the gaussian pyramid from top to bottom
Direction the gradient information figure per tomographic image is successively merged, the method for obtaining the gradient information figure after merging is：
The gradient information figure per tomographic image is up-sampled, the up-sampling gradient information figure per tomographic image is obtained；Every layer
The corresponding gradient information figure of image is identical as the up-sampling gradient information figure resolution ratio of last layer image；
Along the direction of the gaussian pyramid from top to bottom, according to the up-sampling gradient information figure and gradient information of a tomographic image
Figure updates the gradient information figure of adjacent next tomographic image, the gradient information of the bottom layer image until updating the gaussian pyramid
Figure obtains the gradient information figure after fusion.
6. hair method for drafting according to claim 5, which is characterized in that the trend according to the hair, in institute
Stating the method that hair path is extracted in hair zones probability graph includes：
To each point in the hair zones probability graph, extraction path point is moved towards along the hair；
The path point is formed into the first hair path.
7. hair method for drafting according to claim 6, which is characterized in that further include：
It carries out curve fitting to first hair path, obtains smooth hair path；
According to the smooth hair path drawing hair.
8. hair method for drafting according to claim 6, which is characterized in that described in the hair zones probability graph
Each point, the method for moving towards to extract path point along the hair are：
Since preset, along the positive direction extraction point of the trend of the hair, stop after reaching predetermined condition, obtains positive direction
Path point；
Since the preset, along the negative direction extraction point of the trend of the hair, stops after reaching the predetermined condition, obtain
Negate direction path point；
The positive direction path point and the negative direction path point are combined, first hair path is obtained；
The predetermined condition is：The point extracted is located at the outside of the hair zones probability graph；Alternatively, the trend of the hair
The angle of the vector of upper two adjacent points is more than 90 degree；Alternatively, the symbol phase of the ordinate of the two adjacent points extracted
Instead.
9. hair method for drafting according to claim 1, which is characterized in that using the method for deep learning to described original
Image carries out hair zones segmentation.
10. hair method for drafting according to claim 1, which is characterized in that according to the hair by the way of picture point
Path drawing hair.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810374586.6A CN108629781B (en) | 2018-04-24 | 2018-04-24 | Hair drawing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810374586.6A CN108629781B (en) | 2018-04-24 | 2018-04-24 | Hair drawing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108629781A true CN108629781A (en) | 2018-10-09 |
| CN108629781B CN108629781B (en) | 2022-04-22 |
Family
ID=63694599
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810374586.6A Active CN108629781B (en) | 2018-04-24 | 2018-04-24 | Hair drawing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108629781B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112598610A (en) * | 2020-12-11 | 2021-04-02 | 杭州海康机器人技术有限公司 | Depth image obtaining method and device, electronic equipment and storage medium |
| CN114565507A (en) * | 2022-01-17 | 2022-05-31 | 北京新氧科技有限公司 | Hair processing method and device, electronic equipment and storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800129A (en) * | 2012-06-20 | 2012-11-28 | 浙江大学 | Hair modeling and portrait editing method based on single image |
| CN103035030A (en) * | 2012-12-10 | 2013-04-10 | 西北大学 | Hair model modeling method |
| CN103093488A (en) * | 2013-02-02 | 2013-05-08 | 浙江大学 | Virtual haircut interpolation and tweening animation producing method |
| CN103606186A (en) * | 2013-02-02 | 2014-02-26 | 浙江大学 | Virtual hair style modeling method of images and videos |
| CN103927526A (en) * | 2014-04-30 | 2014-07-16 | 长安大学 | Vehicle detecting method based on Gauss difference multi-scale edge fusion |
| US20140198108A1 (en) * | 2013-01-16 | 2014-07-17 | Disney Enterprises, Inc. | Multi-linear dynamic hair or clothing model with efficient collision handling |
| US20140267225A1 (en) * | 2013-03-13 | 2014-09-18 | Microsoft Corporation | Hair surface reconstruction from wide-baseline camera arrays |
| CN105844706A (en) * | 2016-04-19 | 2016-08-10 | 浙江大学 | Full-automatic three-dimensional hair modeling method based on single image |
| CN106611160A (en) * | 2016-12-15 | 2017-05-03 | 中山大学 | CNN (Convolutional Neural Network) based image hair identification method and device |
| CN107451555A (en) * | 2017-07-27 | 2017-12-08 | 安徽慧视金瞳科技有限公司 | A kind of hair based on gradient direction divides to determination methods |
| CN107886516A (en) * | 2017-11-30 | 2018-04-06 | 厦门美图之家科技有限公司 | The method and computing device that hair moves towards in a kind of calculating portrait |
-
2018
- 2018-04-24 CN CN201810374586.6A patent/CN108629781B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800129A (en) * | 2012-06-20 | 2012-11-28 | 浙江大学 | Hair modeling and portrait editing method based on single image |
| CN103035030A (en) * | 2012-12-10 | 2013-04-10 | 西北大学 | Hair model modeling method |
| US20140198108A1 (en) * | 2013-01-16 | 2014-07-17 | Disney Enterprises, Inc. | Multi-linear dynamic hair or clothing model with efficient collision handling |
| CN103093488A (en) * | 2013-02-02 | 2013-05-08 | 浙江大学 | Virtual haircut interpolation and tweening animation producing method |
| CN103606186A (en) * | 2013-02-02 | 2014-02-26 | 浙江大学 | Virtual hair style modeling method of images and videos |
| US20140267225A1 (en) * | 2013-03-13 | 2014-09-18 | Microsoft Corporation | Hair surface reconstruction from wide-baseline camera arrays |
| CN103927526A (en) * | 2014-04-30 | 2014-07-16 | 长安大学 | Vehicle detecting method based on Gauss difference multi-scale edge fusion |
| CN105844706A (en) * | 2016-04-19 | 2016-08-10 | 浙江大学 | Full-automatic three-dimensional hair modeling method based on single image |
| CN106611160A (en) * | 2016-12-15 | 2017-05-03 | 中山大学 | CNN (Convolutional Neural Network) based image hair identification method and device |
| CN107451555A (en) * | 2017-07-27 | 2017-12-08 | 安徽慧视金瞳科技有限公司 | A kind of hair based on gradient direction divides to determination methods |
| CN107886516A (en) * | 2017-11-30 | 2018-04-06 | 厦门美图之家科技有限公司 | The method and computing device that hair moves towards in a kind of calculating portrait |
Non-Patent Citations (4)
| Title |
|---|
| HONG CHEN 等: "A Generative Sketch Models for Human Hair Analysis and Synthesis", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| MENGLEI CHAI 等: "AutoHair: Fully Automatic Hair Modeling from A Single Image", 《ACM TRANSACTIONS ON GRAPHICS》 * |
| XIAOGUANG FENG 等: "Multiscale Principal Components Analysis for Image Local Orientation Estimation", 《CONFERENCE RECORD OF THE THIRTY-SIXTH ASILOMAR CONFERENCE ON SIGNALS，SYSTEMS AND COMPUTER》 * |
| 廖宇: "基于多尺度主成分分析的图像局部方向估计算法", 《计算机应用》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112598610A (en) * | 2020-12-11 | 2021-04-02 | 杭州海康机器人技术有限公司 | Depth image obtaining method and device, electronic equipment and storage medium |
| CN114565507A (en) * | 2022-01-17 | 2022-05-31 | 北京新氧科技有限公司 | Hair processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108629781B (en) | 2022-04-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102449664B (en) | Gradual-change animation generating method and apparatus | |
| WO2021185225A1 (en) | Image super-resolution reconstruction method employing adaptive adjustment | |
| KR102380222B1 (en) | Emotion recognition in video conferencing | |
| CN108596024B (en) | Portrait generation method based on face structure information | |
| CN109919830B (en) | Method for restoring image with reference eye based on aesthetic evaluation | |
| JP6788264B2 (en) | Facial expression recognition method, facial expression recognition device, computer program and advertisement management system | |
| US8300900B2 (en) | Face recognition by fusing similarity probability | |
| CN107194371B (en) | User concentration degree identification method and system based on hierarchical convolutional neural network | |
| CN111862274A (en) | Training method for generating confrontation network, and image style migration method and device | |
| CN108932536A (en) | Human face posture method for reconstructing based on deep neural network | |
| CN107767380A (en) | A kind of compound visual field skin lens image dividing method of high-resolution based on global empty convolution | |
| CN111931908B (en) | Face image automatic generation method based on face contour | |
| CN107220990A (en) | A kind of hair dividing method based on deep learning | |
| CN112633288B (en) | Face sketch generation method based on painting brush touch guidance | |
| CN105095857B (en) | Human face data Enhancement Method based on key point perturbation technique | |
| CN102609964A (en) | Portrait paper-cut generation method | |
| CN108629781A (en) | A kind of hair method for drafting | |
| CN106447720A (en) | Method for constructing golden-ratio face | |
| CN109684973A (en) | The facial image fill system of convolutional neural networks based on symmetrical consistency | |
| CN108596992B (en) | Rapid real-time lip gloss makeup method | |
| CN110298898A (en) | Change the method and its algorithm structure of automobile image body color | |
| CN111612802B (en) | Re-optimization training method based on existing image semantic segmentation model and application | |
| CN107305622A (en) | A kind of human face five-sense-organ recognition methods, apparatus and system | |
| CN106846399A (en) | A kind of method and device of the vision center of gravity for obtaining image | |
| CN109993135A (en) | A kind of gesture identification method based on augmented reality, system and device |
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 |