CN109344710A - A kind of localization method of image characteristic point, device, storage medium and processor - Google Patents
A kind of localization method of image characteristic point, device, storage medium and processor Download PDFInfo
- Publication number
- CN109344710A CN109344710A CN201811002527.2A CN201811002527A CN109344710A CN 109344710 A CN109344710 A CN 109344710A CN 201811002527 A CN201811002527 A CN 201811002527A CN 109344710 A CN109344710 A CN 109344710A
- Authority
- CN
- China
- Prior art keywords
- characteristic point
- pixel
- adsorption capacity
- point
- neighborhood
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
This application discloses a kind of localization method of image characteristic point, device, storage medium and processors.The application corrects the position of the characteristic point of current position inaccurate using the suction-operated in image between pixel and pixel.Firstly, determining all pixels point in its neighborhood of characteristic point；Thereafter, adsorption capacity of each pixel for characteristic point in acquisition neighborhood；And then all pixels point is acquired in neighborhood for the synthesis adsorption capacity of characteristic point using adsorption capacity of each pixel to characteristic point in neighborhood.Finally, being corrected according to the size and Orientation of comprehensive adsorption capacity to the position of the characteristic point of position inaccurate, the position after obtaining characteristic point correction.The application can carry out position correction to the characteristic point of position inaccurate, and compared with the prior art, the accuracy of positioning feature point is significantly improved.
Description
Technical field
This application involves technical field of image processing more particularly to a kind of localization methods of image characteristic point, device, storage
Medium and processor.
Background technique
In technical field of image processing, characteristic point, which refers to, has the point for representing meaning, such as image angle point, edge in image
Point etc..Characteristic point is also known as key point or point of interest.Characteristic point is identified and positioned for image recognition, images match, image shape
Change and 3D modeling etc. play an important role.Such as the recognition of face in image recognition.The accurate positioning of characteristic point is figure
As the key of identification post-processing.The positioning of characteristic point is more accurate in image, and image recognition rate is higher.
Although there is the detection technique of some pairs of image key points at present, these methods are merely capable of positioning feature point
To near real characteristic point, the accuracy of positioning is lower.
Summary of the invention
In order to solve the above technical problem existing in the prior art, the application provides a kind of positioning side of image characteristic point
Method, device, storage medium and processor can optimize the positioning of image characteristic point, keep its positioning more accurate.
This application provides following technical solutions:
The application is in a first aspect, provide a kind of localization method of image characteristic point, comprising:
Determine all pixels point in feature vertex neighborhood；Each pixel has the characteristic point and inhales in the neighborhood
Attached effect；
Each pixel is obtained for the adsorption capacity of the characteristic point；
The all pixels point is obtained for the feature using adsorption capacity of each pixel to the characteristic point
The synthesis adsorption capacity of point；
The position of the characteristic point is corrected according to the size and Orientation of the comprehensive adsorption capacity, obtains the feature
Position after point correction.
It is described to obtain each pixel for the adsorption capacity of the characteristic point as a kind of possible implementation,
It specifically includes:
According to the gradient vector of the pixel, and, the position of the position of the pixel and the characteristic point obtains
Adsorption capacity of the pixel for the characteristic point.
As a kind of possible implementation, the gradient vector according to the pixel, and, the pixel
Position and the position of the characteristic point obtain the pixel for the adsorption capacity of the characteristic point, specifically:
According to the position of the position of the pixel and the characteristic point, obtain between the pixel and the characteristic point
Distance；
The direction that the characteristic point is directed toward the pixel is obtained, with the angle between the pixel gradient direction；
It is obtained by the distance between the gradient vector of the pixel, the pixel and described characteristic point and the angle
Adsorption capacity of the pixel for the characteristic point.
As a kind of possible implementation, each pixel is obtained for the adsorption capacity of the characteristic point described
Later, the method also includes:
The adsorption capacity of the characteristic point is normalized in each pixel；
The all pixels point is obtained for the feature using adsorption capacity of each pixel to the characteristic point
The synthesis adsorption capacity of point, specifically:
Using each pixel to the adsorption capacity after characteristic point normalization obtain all pixels point for
The synthesis adsorption capacity of the characteristic point.
It is described to be obtained using adsorption capacity of each pixel to the characteristic point as a kind of possible implementation
The all pixels point obtains the synthesis adsorption capacity of the characteristic point especially by following formula:
Wherein, poIndicate characteristic point, piIt indicates in the poNeighborhood ΩoIn any one pixel,Table
Show the piTo the poAdsorption capacity, ωiFor adsorption capacityWeight, obtained especially by following formula:
Wherein,Indicate the piWith the poDistance；σdIndicate the neighborhood ΩoMiddle all pixels point and institute
State poDistance standard deviation.
As a kind of possible implementation, the position to the characteristic point is corrected, and obtains the characteristic point
After position after correction, the method also includes:
Change the feature neighborhood of a point, continues to be corrected the position of the characteristic point for new neighborhood, until
The size of the synthesis adsorption capacity of new neighbor assignment is less than or equal to preset threshold.
The application second aspect, provides a kind of positioning device of image characteristic point, and described device includes:
Determination unit, for determining all pixels point in feature vertex neighborhood；Each pixel is for institute in the neighborhood
Characteristic point is stated with suction-operated；
Adsorption capacity acquiring unit, for obtaining each pixel for the adsorption capacity of the characteristic point；
Comprehensive adsorption capacity acquiring unit, described in being obtained using adsorption capacity of each pixel to the characteristic point
Synthesis adsorption capacity of all pixels point for the characteristic point；
Positioning unit, for carrying out school to the position of the characteristic point according to the size and Orientation of the comprehensive adsorption capacity
Just, the position after obtaining the characteristic point correction.
As a kind of possible implementation, the adsorption capacity acquiring unit includes the first acquisition subelement；
Described first obtains subelement, for the gradient vector according to the pixel, and, the position of the pixel
The pixel is obtained for the adsorption capacity of the characteristic point with the position of the characteristic point.
As a kind of possible implementation, the first acquisition subelement includes:
Distance obtains subelement, for obtaining the picture according to the position of the pixel and the position of the characteristic point
The distance between vegetarian refreshments and the characteristic point；
Angle obtains subelement, the direction for being directed toward the pixel for obtaining the characteristic point, with the pixel ladder
Spend the angle between direction；
Adsorption capacity obtains subelement, for by the pixel gradient vector, the pixel and the characteristic point it
Between distance and the angle obtain the pixel for the adsorption capacity of the characteristic point.
As a kind of possible implementation, described device further include: normalization unit；
The normalization unit, for the adsorption capacity of the characteristic point to be normalized in each pixel；
The comprehensive adsorption capacity acquiring unit, is specifically used for:
Using each pixel to the adsorption capacity after characteristic point normalization obtain all pixels point for
The synthesis adsorption capacity of the characteristic point.
As a kind of possible implementation, the comprehensive adsorption capacity acquiring unit obtains institute especially by following formula
All pixels point is stated for the synthesis adsorption capacity of the characteristic point:
Wherein, poIndicate characteristic point, piIt indicates in the poNeighborhood ΩoIn any one pixel,Table
Show the piTo the poAdsorption capacity, ωiFor adsorption capacityWeight, obtained especially by following formula:
Wherein,Indicate the piWith the poDistance；σdIndicate the neighborhood ΩoMiddle all pixels point and institute
State poDistance standard deviation.
As a kind of possible implementation, described device further include:
Circulation correction unit continues for new neighborhood to the characteristic point for changing the feature neighborhood of a point
Position is corrected, until the size of the synthesis adsorption capacity of new neighbor assignment is less than or equal to preset threshold.
The application third aspect provides a kind of computer readable storage medium, is stored with computer program on the medium,
The localization method of the image characteristic point provided such as the above-mentioned first aspect of the application is provided when described program is executed by processor.
The application fourth aspect, provides a kind of processor, and the processor is held when running for running program, described program
The localization method for the image characteristic point that the row such as above-mentioned first aspect of the application provides.
Compared with prior art, the application has at least the following advantages:
Subject to determine the characteristic point of bit image, localization method provided by the present application using pixel in image and pixel it
Between suction-operated, correct the position of the characteristic point of current position inaccurate.Firstly, determining all pictures in its neighborhood of characteristic point
Vegetarian refreshments；Thereafter, adsorption capacity of each pixel for characteristic point in acquisition neighborhood；And then using each pixel to neighborhood
The adsorption capacity of interior characteristic point acquires all pixels point for the synthesis adsorption capacity of characteristic point.Finally, according to the big of comprehensive adsorption capacity
Small and direction is corrected the position of the characteristic point of position inaccurate, the position after obtaining characteristic point correction.
Due to synthesis suction-operated of all pixels point to it in comprehensive adsorption capacity characteristic feature vertex neighborhood, comprehensive adsorption capacity
For a vector, including size and Orientation, the position of size characterization current signature point and the gap of accurate location, and its direction table
Levy the accurate location of characteristic point and the relative direction of current location.Correction refers to that characteristic point is based on current location inhales according to comprehensive
The corresponding distance of its mobile size of the direction of attached power, the position after movement are the positions after characteristic point correction.It can be seen that should
Method can carry out position correction to the characteristic point of position inaccurate, and compared with the prior art, the accuracy of positioning feature point obtains
To significantly improving.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The some embodiments recorded in application, for those of ordinary skill in the art, without creative efforts,
It can also be obtained according to these attached drawings other attached drawings.
Fig. 1 is a kind of localization method flow chart of image characteristic point provided by the embodiments of the present application；
Fig. 2 is each characteristic point suffered comprehensive adsorption capacity in its neighborhood in eyes image provided by the embodiments of the present application
Distribution map；
Fig. 3 is the localization method flow chart of another image characteristic point provided by the embodiments of the present application；
Fig. 4 is pixel in feature vertex neighborhood provided by the embodiments of the present application to the adsorption capacity schematic diagram of characteristic point；
Fig. 5 is a kind of positioning device structure schematic diagram of image characteristic point provided by the embodiments of the present application；
Fig. 6 is a kind of positioning device hardware structure diagram of image characteristic point provided by the embodiments of the present application.
Specific embodiment
In order to make those skilled in the art more fully understand application scheme, below in conjunction in the embodiment of the present application
Attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is only this
Apply for a part of the embodiment, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art exist
Every other embodiment obtained under the premise of creative work is not made, shall fall in the protection scope of this application.
Image recognition has a wide range of applications in people's lives.In image recognition application, the feature of image is lighted
To crucial effect.The accuracy of positioning feature point is capable of the identifiability of influence diagram picture and processing quality, positioning feature point are got over
Accurately, image recognition and treatment effect are more ideal.For example, characteristic point, which is accurately positioned, to be helped to make the same object in two pictures
More accurate more natural image registration is realized, so that image mosaic and image synthetic effect are more ideal；And it is special in image
The positioning of sign point helps clearly to distinguish the different objects in image；Help to realize more true image enhancement, image shape
Become the effect with three-dimensional modeling.
For example, being applied to image " beauty " scene in image recognition below with characteristic point in image, in order to image
Intuitively understand the significance that characteristic point is accurately positioned.
Operation has the application program (Application, APP) of face " beauty " function in mobile terminal, such as to people
As makeup.Beauty is carried out to the portrait photographs that mobile terminal is locally stored using the APP, adds beauty automatically for personage in photo
Eyelashes.But due to characteristic point (constituting the pixel of the eyelid of personage in image) identification inaccuracy, the eyelid that APP is identified
Personage needs to add the eyelid positions of making-up eyelashes there are deviation in position and photo, and leading to APP is the beauty that picture adds automatically
Holding eyelashes root cannot be overlapped with the eyelid position of personage in picture, and image " beauty " is ineffective.And if Feature point recognition
Accurately, then the root of making-up eyelashes accurately can be added to the eyelid position of personage, thus realize cosmetic result, otherwise suitable
It is anti-.
For the problem of positioning feature point inaccuracy, inventor has found that in image pixel and pixel it
Between there are suction-operateds.In the application, inventor will use " adsorption capacity " this concept, specifically describe absorption between points
Effect.As vector, adsorption capacity has both size and direction.It should be noted that adsorption capacity be different from physics solid substance it
Between mechanical function, but to existing correlation between the data information that different pixels point respectively contains in same piece image
It is described.Adsorption capacity, which refers to, gives some interactively between its surrounding pixel point in image.
In the application, localization method, device, storage medium and the processor of a kind of image characteristic point are inventor provided,
Using the suction-operated in image between pixel and pixel, position correction is carried out to the characteristic point of current position inaccurate.
The localization method of image characteristic point, device, storage medium and processor are described respectively below with reference to embodiment and attached drawing.
First embodiment
Referring to Fig. 1, which is a kind of localization method flow chart of image characteristic point provided by the embodiments of the present application.
As shown in Figure 1, the localization method of image characteristic point provided by the embodiments of the present application, comprising:
Step 101: determining all pixels point in feature vertex neighborhood；Each pixel is for the feature in the neighborhood
Point has suction-operated.
It should be noted that a width, which has been obtained ahead of time, located the image of characteristic point using the prior art in the present embodiment.
Since positional accuracy is limited, leading to the characteristic point position positioned in the image and accurate location, there are deviations.The present embodiment
It is the image being obtained ahead of time based on this, characteristic point therein is relocated.
Find mutually there is suction-operated in image between pixel and pixel according to the inventor's study.This implementation
Example is based on the suction-operated location feature point between pixel and pixel.Pixel in feature vertex neighborhood can be used as this reality
It applies example and corrects the pixel played a significant role to characteristic point position, and the pixel except feature vertex neighborhood is to this characteristic point
Position correction does not have important function.
It should be noted that the magnitude range of neighborhood specifically can be according to the positional accuracy and locating speed to characteristic point
Demand is arranged accordingly, and the range of neighborhood is bigger, and the pixel number in neighborhood is more, then positioning feature point accuracy
It is higher.But correspondingly since pixel number is more, positioning required calculation amount can be larger, and calculation amount is larger, causes to position
Speed is slower.Therefore, neighborhood small as far as possible can be selected, to improve positioning under the premise of meeting positional accuracy requirement
Speed.It is not specifically limited, can be chosen accordingly according to practical application scene and demand in the present embodiment.
For example, for characteristic point, 9 × 9 pixel models that neighborhood can be set as centered on characteristic point are accurately positioned
It encloses；If there is higher demand to positioning accuracy, the contiguous range of setting can be will be enlarged by, for example, be set as with characteristic point be
15 × 15 pixel point ranges of the heart；And locating speed is if desired improved, the contiguous range of setting can be reduced, such as be set as
7 × 7 pixel point ranges centered on characteristic point.
Step 102: adsorption capacity of each pixel for characteristic point in acquisition neighborhood.
The present embodiment is using the suction-operated in image between pixel and pixel, to the feature of current position inaccurate
Point carries out position correction.Each pixel has suction-operated, i.e., each pixel pair for this feature point in feature vertex neighborhood
Adsorption capacity is all had in the characteristic point in neighborhood.Pixel for characteristic point adsorption capacity generally with the gradient vector phase of pixel
It closes, gradient vector is bigger, then the pixel is bigger for the adsorption capacity of characteristic point.
Step 103: obtaining all pixels point for characteristic point using adsorption capacity of each pixel to the characteristic point
Comprehensive adsorption capacity.
Comprehensive adsorption capacity characterizes in the feature vertex neighborhood of current position inaccurate, synthesis of all pixels point to this feature point
Suction-operated.The each pixel obtained using step 102 is obtained comprehensive adsorption capacity, had a variety of to the adsorption capacity of characteristic point
Implementation.
It as an example, can be according to pixel each in neighborhood at a distance from characteristic point, for apart from different pixels
Weight is respectively configured in the corresponding adsorption capacity of point, and finally the adsorption capacity for being configured with weight adds up, obtains all pixels point pair
In the synthesis adsorption capacity of characteristic point.
Step 104: the position of characteristic point being corrected according to the size and Orientation of comprehensive adsorption capacity, obtains characteristic point school
Position after just.
The synthesis adsorption capacity that step 103 obtains has size and Orientation.The size of comprehensive adsorption capacity characterizes current signature point
Position and accurate location gap, comprehensive adsorption capacity is bigger, indicates that the position change of correction front and back characteristic point is bigger.It is comprehensive to inhale
The accurate location of the direction characteristic feature point of attached power and the relative direction of current location need to inhale the position of characteristic point to comprehensive
The correction for direction of attached power.
This step understands in combination with Fig. 2.Fig. 2 is each characteristic point in eyes image provided in this embodiment in its neighborhood
The distribution map of suffered comprehensive adsorption capacity.The each arrow being distributed in Fig. 2, i.e., the characteristic point where expression arrow is by its neighborhood
Comprehensive adsorption capacity.
Wherein, arrow length indicates that the size of comprehensive adsorption capacity, arrow direction indicate the direction of comprehensive adsorption capacity.Such as arrow
Head is longer, and the characteristic point where indicating is bigger by comprehensive adsorption capacity, and accordingly, arrow length is shorter, the characteristic point where indicating
It is smaller by comprehensive adsorption capacity.The direction of arrow indicates that current signature point position needs to be corrected towards the direction of arrow in Fig. 2.
As can be seen from FIG. 2, marginal point and the interior arrow length being distributed of angle point neighborhood are significantly longer in image, that is, indicate these
Synthesis adsorption capacity suffered by marginal point and angle point is bigger.The arrow length in region is shorter between eyebrow and upper eyelid in Fig. 2, this is
Without marginal point and angle point in feature vertex neighborhood as where arrow, in neighborhood each pixel to the suction-operated of characteristic point compared with
For equilibrium, cause comprehensive adsorption capacity not significant.
In addition, eyebrow center and pupil center's arrow length are relatively small in Fig. 2, this is because although spy where arrow
Have marginal point and angle point in sign vertex neighborhood, but each pixel is more balanced to the suction-operated of characteristic point in neighborhood, also leads
Cause comprehensive adsorption capacity not significant.
Finally, position correction is carried out according to characteristic point of the size and Orientation of comprehensive adsorption capacity to position inaccurate, obtained
Position after characteristic point correction.As an example, the position of characteristic point is (x before correcting0, y0), the size of comprehensive adsorption capacity is D, comprehensive
It is (x ', y ') that close adsorption capacity direction, which be the characteristic point position along x-axis forward direction, then after correcting, wherein x '=x0+ D, y '=y0。
It should be noted that the application can be by successive ignition localization method provided in this embodiment, final realization pair
The accurate positionin of characteristic point.Position and characteristic point before the number of iterations and the correction of the contiguous range size and characteristic point of selection
The factors such as the distance of accurate location are related.For example, if characteristic point correction before position and its accurate location between distance it is smaller, very
It is the accurate positionin completed to characteristic point possibly also with the less number of localization method iteration provided in this embodiment.
For localization method provided by the present application using the suction-operated in image between pixel and pixel, it is current fixed to correct
The position of the characteristic point of position inaccuracy.Since all pixels point adsorbs the comprehensive of its in comprehensive adsorption capacity characteristic feature vertex neighborhood
Effect, comprehensive adsorption capacity be a vector, including size and Orientation, and size characterizes position and the accurate location of current signature point
Gap, and the relative direction of the accurate location of its direction characteristic feature point and current location.Correction refers to be based on working as by characteristic point
According to the corresponding distance of its mobile size of direction of comprehensive adsorption capacity, the position after movement is after characteristic point corrects for front position
Position.It can be seen that this method can carry out position correction, compared with the prior art, feature to the characteristic point of position inaccurate
The accuracy of point location is significantly improved.
Localization method based on the image characteristic point that previous embodiment provides, present invention also provides obtain feature vertex neighborhood
Interior any pixel is to its adsorption capacity, and obtains specific implementation of all pixels point to the synthesis adsorption capacity of characteristic point in neighborhood
Mode, and then provide the localization method of another image characteristic point.Second embodiment will be situated between to this method in conjunction with attached drawing
It continues.
Second embodiment
Referring to Fig. 3, which is the localization method flow chart of another image characteristic point provided by the embodiments of the present application.
As shown in figure 3, the localization method of image characteristic point provided by the embodiments of the present application, comprising:
In the present embodiment, step 301 is identical as step 101 in first embodiment, and the specific implementation of step 301 can refer to
Description in one embodiment, details are not described herein.
Step 302: according to the gradient vector of pixel, and, the position of pixel and the position of characteristic point obtain pixel
Adsorption capacity of the point for characteristic point.
The characteristic point of current position inaccurate, ladder of any pixel to its adsorption capacity, with the pixel in neighborhood
The position for spending vector and two o'clock is related.
Image is stored with discrete digital signal form, if image is regarded as two-dimensional discrete function, picture in image
The gradient vector of vegetarian refreshments is actually the result to two-dimensional discrete function derivation.As vector, the existing direction of gradient vector is again
There is size.The gradient direction of any pixel characterizes data information (such as gray value) that the pixel includes to feature in neighborhood
The absorption direction for the data information that point includes, therefore gradient direction of the pixel to the adsorption capacity direction of characteristic point, with pixel
Unanimously.
Seek the gradient vector of pixel calculation formula include it is a variety of, due in field of image processing, gradient vector category
It is therefore, no longer specific herein to introduce in the concept of comparative maturity.Not to the specific side for obtaining pixel gradient vector in this step
Formula is defined.
As a kind of specific implementation, step 302 be may include steps of.
Step 302a: according to the position of the position of pixel and characteristic point, the distance between pixel and characteristic point are obtained.
As an example, can directly utilize Euclidean distance formula in step 302a, the distance between two o'clock is obtained.Certainly,
This step can use other calculation formula also to obtain the distance between pixel and characteristic point.In the present embodiment, for obtaining
The specific calculation of distance, is not limited between capture vegetarian refreshments and characteristic point.
Step 302b: the folder between the direction that the characteristic point is directed toward the pixel, with pixel gradient direction is obtained
Angle.
Step 302c: by the distance between the gradient vector of the pixel, the pixel and characteristic point and the folder
Angle obtains pixel for the adsorption capacity of characteristic point.
Step 302a to step 302c for ease of understanding, reference can be made to Fig. 4 and formula (1).Fig. 4 provides for the embodiment of the present application
Feature vertex neighborhood in pixel to the adsorption capacity schematic diagram of characteristic point.
In Fig. 4, poIndicate the characteristic point of current position inaccurate, piIt indicates in poAny one pixel in neighborhood,Indicate piTo poAdsorption capacity, gpiIndicate piGradient vector, θ poIt is directed toward piDirection and piGradient direction
Angle.
Specifically, it can use following formula and calculate piTo poAdsorption capacity
From formula (1) as can be seen that the size of adsorption capacity is directly proportional to the gradient vector of pixel, gradient vector is bigger,
Then corresponding adsorption capacity is bigger.
Step 303: adsorption capacity of each pixel for characteristic point is normalized.
In image, can more or less there are some noise spots, the presence of these noise spots will to the accurate positioning of characteristic point
It impacts, reduces the accuracy of positioning.Such as the gradient vector for the noise spot having may be very big, will cause very big suction in this way
Attached power, if the movement of mistake will be caused along the adsorption capacity moving characteristic point.Therefore, in order to inhibit noise in image point pair
It is influenced caused by positioning feature point, characteristic point positioning method provided by the present application proposes, by pixel each in neighborhood to feature
The adsorption capacity of point is normalized.The adsorption capacity of characteristic point is avoided to weaken noise spot in neighborhood by normalization adsorption capacity
Adsorption capacity as noise spot is excessive to cause positioning feature point to generate deviation.
Present embodiments provide normalization adsorption capacity embodiment, the adsorption capacity that step 302 is acquired and normalization because
Son is multiplied, the adsorption capacity after being normalized.Each pixel is to the adsorption capacity after the normalization of this feature point in feature vertex neighborhoodIt can be indicated by formula (2).
In formula (2), η is normalization factor.The calculation formula of normalization factor η are as follows:
In formula (3), max (| | gpi| |) indicate in characteristic point poThe corresponding gradient vector norm of all pixels point in neighborhood
Maximum value.
Step 304: using each pixel to the characteristic point normalization after adsorption capacity obtain all pixels point for
The synthesis adsorption capacity of characteristic point.
In order to avoid characteristic point falls into local optimum, the close pixel of characteristic point of adjusting the distance and distance feature point are far
The adsorption capacity of pixel uses restraint respectively.Such as, on the one hand, avoid the corresponding adsorption capacity of pixel that distance feature point is close
Influence it is excessive, can prevent from falling into local optimum and the influence of ignoring pixel farther out；On the other hand distance feature point is avoided
The corresponding adsorption capacity of the very big pixel of far but gradient (such as the extraordinary images such as noise spot vegetarian refreshments) influences excessive.Therefore, this reality
One class Gauss model of introducing in example is applied, such Gauss model is used to be poPixel in neighborhood divides the adsorption capacity of characteristic point
With weight, to constrain in neighborhood each pixel to poThe synthesis suction-operated of point.
The formula of class Gauss model is specific as follows:
In formula (4),It is characterized point poWith its neighborhood ΩoInterior any pixel point piDistance, σdIndicate neighborhood
ΩoMiddle all pixels point and poDistance standard deviation, ωiFor piTo characteristic point poAdsorption capacity weight.It can by formula (4)
Know, neighborhood ΩoIt is interior, with characteristic point poThe smaller pixel of distance, corresponding adsorption capacity weight is relatively bigger.
After introducing class Gauss model, the comprehensive adsorption capacity Ψ of the calculating of propositionoFormula are as follows:
In above formula,It can be the characteristic point p that step 303 obtainsoNeighborhood ΩoIt is interior, any pixel point piIt is corresponding
Adsorption capacity after normalization, or the characteristic point p that step 302 obtainsoNeighborhood ΩoIt is interior, any pixel point piTo characteristic point po
Adsorption capacity.
Step 305: the position of characteristic point being corrected according to the size and Orientation of comprehensive adsorption capacity, obtains characteristic point school
Position after just.
In the present embodiment, step 305 is identical as step 104 in first embodiment, and the specific implementation of step 305 can refer to
Description in one embodiment, details are not described herein.
The above are the localization methods of image characteristic point provided by the embodiments of the present application.In this method, feature vertex neighborhood is determined
After interior all pixels point, pixel is obtained according to the position of the position and characteristic point of the gradient vector of pixel and pixel
Adsorption capacity of the point for characteristic point；Later, adsorption capacity of each pixel for characteristic point is normalized, is recycled each
Adsorption capacity after pixel normalizes the characteristic point obtains all pixels point for the synthesis adsorption capacity of characteristic point；Finally,
The position of characteristic point is corrected according to the size and Orientation of comprehensive adsorption capacity, the position after obtaining characteristic point correction.
In this method, for reduction image in position inaccurate feature vertex neighborhood memory noise spot to positioning feature point
Caused by influence, adsorption capacity of each pixel to characteristic point in neighborhood is normalized.In addition, the institute in calculating field
It is each by class Gauss model relevant at a distance from pixel and characteristic point when having synthesis adsorption capacity of the pixel to characteristic point
The corresponding adsorption capacity of a pixel distributes weight, is not only able to avoid that its corresponding adsorption capacity influenced apart from close pixel
Greatly, meanwhile, even if it is big apart from remote but gradient also to can avoid the extraordinary images vegetarian refreshments such as noise spot, the accurate positionin of interference characteristic point.By
This, further increases the accuracy of this method location feature point.
It is further to note that under normal circumstances, using image characteristic points positioning method provided in this embodiment, for
The correction of characteristic point will not be completed disposably, be the process of loop iteration.That is, relative to the position before each correction,
The accurate location of characteristic point position more Approximation Characteristic point after secondary correction.Characteristics of image point location side provided in this embodiment
Method may further comprise: after step 305
Step 306: changing the feature neighborhood of a point, continue to carry out school to the position of the characteristic point for new neighborhood
Just, until the size of the synthesis adsorption capacity of new neighbor assignment is less than or equal to preset threshold.
The specific implementation process of step 306 is illustrated below with one.
For example, there are the characteristic point A of a position inaccurate in image, position in the picture is (x1, y1), is passed through
Implementation steps 301 obtain all pixels point in characteristic point A neighborhood for the synthesis adsorption capacity of characteristic point A to step 305, and root
Finally the position of characteristic point A is corrected according to the synthesis adsorption capacity, the position (x2, y2) after obtaining characteristic point A correction.Its
Afterwards, the corresponding new neighborhood of characteristic point of the position (x2, y2) can be determined by the pixel of the position (x2, y2) as characteristic point,
Above-mentioned steps 301 are repeated to step 304, seek in new neighborhood all pixels point to the comprehensive of (x2, y2) position feature point
Close adsorption capacity.
Thereafter, the size of the synthesis adsorption capacity is compared with preset threshold.The preset threshold is that the present embodiment is to mention
High characteristic point position calibration accuracy and the synthesis adsorption capacity size limit value being arranged: when comprehensive adsorption capacity size is greater than preset threshold
When, showing characteristic point current location and accurate location, there is also relatively large deviations, and the accuracy of current location is not high enough, needs into one
Walk correction feature point position；When comprehensive adsorption capacity size is less than or equal to preset threshold, show characteristic point current location and standard
It is very small that there are deviations for true position, and the accuracy of current location is sufficiently high, is not necessarily to further correction feature point position.
That is, if the synthesis adsorption capacity size of the neighbor assignment of the pixel of the position (x2, y2) is less than or equal to pre-
, can be by (x2, y2) as the characteristic point after correction if threshold value, primary correction completes the precise calibration of characteristic point position.
And if the synthesis adsorption capacity size of the neighbor assignment of the pixel of the position (x2, y2) is greater than preset threshold, needs according to the synthesis
Adsorption capacity corrects the position (x2, y2), the characteristic point position (x3, y3) after obtaining second-order correction, (x3, y3) after being corrected
It is compared the synthesis adsorption capacity size of (x3, y3) position feature point by the neighborhood of pixel points of position with preset threshold,
Until stopping characteristic point position correction when determining that comprehensive adsorption capacity size is less than or equal to preset threshold.Multiple school is recycled as a result,
Just completing the precise calibration of characteristic point position.
In moving process, neighborhood can change characteristic point, and after neighborhood variation, the pixel in neighborhood can change, therefore, as
Vegetarian refreshments can also change the synthesis adsorption capacity of characteristic point in its affiliated neighborhood.When waiting comprehensive adsorption capacity smaller always or being 0, just tie
Position correction of the beam to characteristic point, is equivalent to and has corrected that in place.As an example, preset threshold can according to need setting, when
So in order to which accuracy is very high, 0 can be set to, that is to say, that when the size of comprehensive function power is 0, just stop to feature
The correction of point position.
Localization method based on the image characteristic point that previous embodiment provides, present invention also provides a kind of image characteristic points
Positioning device.3rd embodiment will be introduced the device in conjunction with attached drawing.
3rd embodiment
Referring to Fig. 5, which is a kind of positioning device structure schematic diagram of image characteristic point provided by the embodiments of the present application.
As shown in figure 5, the positioning device of image characteristic point provided by the embodiments of the present application, comprising: determination unit 51, absorption
Power acquiring unit 52, comprehensive adsorption capacity acquiring unit 53 and positioning unit 54.
Wherein it is determined that unit 51, for determining all pixels point in feature vertex neighborhood；Each pixel in the neighborhood
There is suction-operated for the characteristic point；
Adsorption capacity acquiring unit 52, for obtaining each pixel for the adsorption capacity of the characteristic point；
Comprehensive adsorption capacity acquiring unit 53, for obtaining institute using adsorption capacity of each pixel to the characteristic point
All pixels point is stated for the synthesis adsorption capacity of the characteristic point；
Positioning unit 54, for carrying out school to the position of the characteristic point according to the size and Orientation of the comprehensive adsorption capacity
Just, the position after obtaining the characteristic point correction.
The above are the positioning device of image characteristic point provided by the embodiments of the present application, which utilizes current position inaccurate
The neighbouring pixel of characteristic point to its adsorption capacity, obtain adjacent pixels point to the synthesis adsorption capacity of characteristic point, it is final to utilize
Comprehensive adsorption capacity is corrected the position of characteristic point.Due in comprehensive adsorption capacity characteristic feature vertex neighborhood all pixels point to it
Synthesis suction-operated, comprehensive adsorption capacity is a vector, including size and Orientation, size characterize the position of current signature point with
The gap of accurate location, and the relative direction of the accurate location of its direction characteristic feature point and current location.Correction refers to will be special
For sign point based on current location according to the corresponding distance of its mobile size of direction of comprehensive adsorption capacity, the position after movement is feature
Position after point correction.It can be seen that the device can carry out position correction to the characteristic point of position inaccurate, compared to existing
The accuracy of technology, positioning feature point is significantly improved.
In some possible implementations of the application, above-mentioned adsorption capacity acquiring unit 52 may include: the first acquisition
Unit；
Described first obtains subelement, for the gradient vector according to the pixel, and, the position of the pixel
The pixel is obtained for the adsorption capacity of the characteristic point with the position of the characteristic point.
In some possible implementations of the application, the first acquisition subelement includes: that distance obtains subelement, folder
Angle obtains subelement and adsorption capacity obtains subelement.
Wherein, distance obtains subelement, for obtaining institute according to the position of the pixel and the position of the characteristic point
State the distance between pixel and the characteristic point；
Angle obtains subelement, the direction for being directed toward the pixel for obtaining the characteristic point, with the pixel ladder
Spend the angle between direction；
Adsorption capacity obtains subelement, obtains institute for gradient vector, the distance and the angle by the pixel
Pixel is stated for the adsorption capacity of the characteristic point.
In some possible implementations of the application, the positioning device of image characteristic point provided in this embodiment can be with
It include: normalization unit；
The normalization unit, for the adsorption capacity of the characteristic point to be normalized in each pixel；
The comprehensive adsorption capacity acquiring unit 53, is specifically used for:
Using each pixel to the adsorption capacity after characteristic point normalization obtain all pixels point for
The synthesis adsorption capacity of the characteristic point.
The comprehensive adsorption capacity acquiring unit 53 can specifically obtain all pixels point for institute by following formula
State the synthesis adsorption capacity of characteristic point:
Wherein, poIndicate characteristic point, piIt indicates in the poNeighborhood ΩoIn any one pixel,Table
Show the piTo the poAdsorption capacity, ω i be adsorption capacityWeight, obtained especially by following formula:
Wherein,Indicate the piWith the poDistance；σdIndicate the neighborhood ΩoMiddle all pixels point and institute
State poDistance standard deviation.
In some possible implementations of the application, the positioning device of image characteristic point provided in this embodiment can be with
Include:
Circulation correction unit continues for new neighborhood to the characteristic point for changing the feature neighborhood of a point
Position is corrected, until the size of the synthesis adsorption capacity of new neighbor assignment is less than or equal to preset threshold.
In the present embodiment, preset threshold is that the present embodiment is the comprehensive suction for improving characteristic point position calibration accuracy and being arranged
Attached power size limit value: when comprehensive adsorption capacity size is greater than preset threshold, show that characteristic point current location is also deposited with accurate location
It is not high enough in the accuracy of relatively large deviation, current location, need further correction feature point position；When comprehensive adsorption capacity size is low
In or when being equal to preset threshold, show characteristic point current location and accurate location there is also deviations very small, the standard of current location
Exactness is sufficiently high, is not necessarily to further correction feature point position.Unit is corrected by the circulation of positioning device, it can be to feature
Point carries out continuous several times correction, the final precise calibration realized to characteristic point position.
The embodiment of the present application also provides a kind of storage mediums, are stored thereon with program, when which is executed by processor
Realize part or Overall Steps in the localization method of the image characteristic point of the application first embodiment and second embodiment protection.It should
Storage medium can be USB flash disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory
The various media that can store program code such as (Random Access Memory, RAM), magnetic or disk.
The embodiment of the present application provides a kind of processor, and the processor is for running program, wherein described program operation
Part or Overall Steps in the localization method of Shi Zhihang first embodiment and the image characteristic point of second embodiment protection.
Based on storage medium and processor that previous embodiment provides, present invention also provides a kind of determining for image characteristic point
Position equipment.
Referring to Fig. 6, which is the positioning device hardware structure diagram of image characteristic point provided in this embodiment.
As shown in fig. 6, the positioning device of image characteristic point includes: memory 601, processor 602,603 and of communication bus
Communication interface 604.
Wherein, the program that can be run on a processor is stored on memory 601, program realizes the application first when executing
Part or Overall Steps in the localization method for the image characteristic point that embodiment and second embodiment provide.Memory 601 can wrap
High-speed random access memory is included, can also include nonvolatile memory, for example, at least disk memory, a flash memories
Part or other volatile solid-state parts.
In the positioning device, processor 602 and memory 601 pass through transmission signaling, logical order etc..This is fixed
Position equipment can carry out communication interaction by communication interface 604 and other equipment.
The above method is executed by program, position correction can be carried out to the characteristic point of position inaccurate, compared to existing
The accuracy of technology, positioning feature point is significantly improved.
It should be appreciated that in this application, " at least one (item) " refers to one or more, and " multiple " refer to two or two
More than a."and/or" indicates may exist three kinds of relationships, for example, " A and/or B " for describing the incidence relation of affiliated partner
It can indicate: only exist A, only exist B and exist simultaneously tri- kinds of situations of A and B, wherein A, B can be odd number or plural number.Word
Symbol "/" typicallys represent the relationship that forward-backward correlation object is a kind of "or"." at least one of following (a) " or its similar expression, refers to
Any combination in these, any combination including individual event (a) or complex item (a).At least one of for example, in a, b or c
(a) can indicate: a, b, c, " a and b ", " a and c ", " b and c ", or " a and b and c ", and wherein a, b, c can be individually, can also
To be multiple.
The above is only the preferred embodiment of the application, not makes any form of restriction to the application.Though
Right the application has been disclosed in a preferred embodiment above, however is not limited to the application.It is any to be familiar with those skilled in the art
Member, in the case where not departing from technical scheme ambit, all using the methods and technical content of the disclosure above to the application
Technical solution makes many possible changes and modifications or equivalent example modified to equivalent change.Therefore, it is all without departing from
The content of technical scheme, any simple modification made to the above embodiment of the technical spirit of foundation the application are equal
Variation and modification, still fall within technical scheme protection in the range of.
Claims (10)
1. a kind of localization method of image characteristic point characterized by comprising
Determine all pixels point in feature vertex neighborhood；Each pixel there is absorption to make the characteristic point in the neighborhood
With；
Each pixel is obtained for the adsorption capacity of the characteristic point；
The all pixels point is obtained for the characteristic point to the adsorption capacity of the characteristic point using each pixel
Comprehensive adsorption capacity；
The position of the characteristic point is corrected according to the size and Orientation of the comprehensive adsorption capacity, obtains the characteristic point school
Position after just.
2. the localization method of image characteristic point according to claim 1, which is characterized in that described to obtain each pixel
Point specifically includes the adsorption capacity of the characteristic point:
According to the gradient vector of the pixel, and, described in the position acquisition of the position of the pixel and the characteristic point
Adsorption capacity of the pixel for the characteristic point.
3. the localization method of image characteristic point according to claim 2, which is characterized in that described according to the pixel
Gradient vector, and, the position of the position of the pixel and the characteristic point obtains the pixel for the characteristic point
Adsorption capacity, specifically:
According to the position of the position of the pixel and the characteristic point, obtain between the pixel and the characteristic point away from
From；
The direction that the characteristic point is directed toward the pixel is obtained, with the angle between the pixel gradient direction；
As described in the distance between the gradient vector of the pixel, the pixel and described characteristic point and angle acquisition
Adsorption capacity of the pixel for the characteristic point.
4. the localization method of image characteristic point according to claim 1-3, which is characterized in that obtained often described
After a pixel is for the adsorption capacity of the characteristic point, the method also includes:
The adsorption capacity of the characteristic point is normalized in each pixel；
The all pixels point is obtained for the characteristic point to the adsorption capacity of the characteristic point using each pixel
Comprehensive adsorption capacity, specifically:
The all pixels point is obtained for described to the adsorption capacity after characteristic point normalization using each pixel
The synthesis adsorption capacity of characteristic point.
5. the localization method of image characteristic point according to claim 1-3, which is characterized in that described using each
The pixel obtains all pixels point for the synthesis adsorption capacity of the characteristic point, tool to the adsorption capacity of the characteristic point
Body is obtained by following formula:
Wherein, poIndicate characteristic point, piIt indicates in the poNeighborhood ΩoIn any one pixel,Described in expression
piTo the poAdsorption capacity, ωiFor adsorption capacityWeight, obtained especially by following formula:
Wherein,Indicate the piWith the poDistance；σdIndicate the neighborhood ΩoMiddle all pixels point and the po
Distance standard deviation.
6. the localization method of image characteristic point according to claim 1-3, which is characterized in that described to the spy
The position of sign point is corrected, after the position after obtaining the characteristic point correction, the method also includes:
Change the feature neighborhood of a point, continues to be corrected the position of the characteristic point for new neighborhood, until new
The size of the synthesis adsorption capacity of neighbor assignment is less than or equal to preset threshold.
7. a kind of positioning device of image characteristic point characterized by comprising
Determination unit, for determining all pixels point in feature vertex neighborhood；Each pixel is for the spy in the neighborhood
Sign point has suction-operated；
Adsorption capacity acquiring unit, for obtaining each pixel for the adsorption capacity of the characteristic point；
Comprehensive adsorption capacity acquiring unit, it is described all for being obtained using adsorption capacity of each pixel to the characteristic point
Synthesis adsorption capacity of the pixel for the characteristic point；
Positioning unit is obtained for being corrected according to the size and Orientation of the comprehensive adsorption capacity to the position of the characteristic point
Position after obtaining the characteristic point correction.
8. the positioning device of image characteristic point according to claim 7, which is characterized in that the adsorption capacity acquiring unit packet
Include the first acquisition subelement；
Described first obtains subelement, for the gradient vector according to the pixel, and, the position of the pixel and institute
The position for stating characteristic point obtains the pixel for the adsorption capacity of the characteristic point.
9. a kind of computer readable storage medium, which is characterized in that be stored with computer program, described program quilt on the medium
Processor realizes the localization method of image characteristic point as claimed in any one of claims 1 to 6 when executing.
10. a kind of processor, which is characterized in that the processor executes such as right for running program when described program is run
It is required that the localization method of the described in any item image characteristic points of 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811002527.2A CN109344710B (en) | 2018-08-30 | 2018-08-30 | Image feature point positioning method and device, storage medium and processor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811002527.2A CN109344710B (en) | 2018-08-30 | 2018-08-30 | Image feature point positioning method and device, storage medium and processor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109344710A true CN109344710A (en) | 2019-02-15 |
| CN109344710B CN109344710B (en) | 2020-12-18 |
Family
ID=65296993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811002527.2A Active CN109344710B (en) | 2018-08-30 | 2018-08-30 | Image feature point positioning method and device, storage medium and processor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109344710B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113077513A (en) * | 2021-06-03 | 2021-07-06 | 深圳市优必选科技股份有限公司 | Visual positioning method and device and computer equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106339693A (en) * | 2016-09-12 | 2017-01-18 | 华中科技大学 | Positioning method of face characteristic point under natural condition |
| US9659234B1 (en) * | 2015-12-01 | 2017-05-23 | Intel Corporation | Adaptive selection of scale invariant image feature keypoints |
| CN106897662A (en) * | 2017-01-06 | 2017-06-27 | 北京交通大学 | The localization method of the face key feature points based on multi-task learning |
| CN106991388A (en) * | 2017-03-27 | 2017-07-28 | 中国科学院自动化研究所 | Crucial independent positioning method |
| CN107392215A (en) * | 2017-08-02 | 2017-11-24 | 焦点科技股份有限公司 | A kind of multigraph detection method based on SIFT algorithms |
| CN107742312A (en) * | 2017-10-09 | 2018-02-27 | 沈阳东软医疗系统有限公司 | A kind of method and apparatus that key point is positioned in medical image |
-
2018
- 2018-08-30 CN CN201811002527.2A patent/CN109344710B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9659234B1 (en) * | 2015-12-01 | 2017-05-23 | Intel Corporation | Adaptive selection of scale invariant image feature keypoints |
| CN106339693A (en) * | 2016-09-12 | 2017-01-18 | 华中科技大学 | Positioning method of face characteristic point under natural condition |
| CN106897662A (en) * | 2017-01-06 | 2017-06-27 | 北京交通大学 | The localization method of the face key feature points based on multi-task learning |
| CN106991388A (en) * | 2017-03-27 | 2017-07-28 | 中国科学院自动化研究所 | Crucial independent positioning method |
| CN107392215A (en) * | 2017-08-02 | 2017-11-24 | 焦点科技股份有限公司 | A kind of multigraph detection method based on SIFT algorithms |
| CN107742312A (en) * | 2017-10-09 | 2018-02-27 | 沈阳东软医疗系统有限公司 | A kind of method and apparatus that key point is positioned in medical image |
Non-Patent Citations (2)
| Title |
|---|
| NANNAN DING等: ""Image Registration Based on Log-Polar Transform and SIFT Features"", 《2010 INTERNATIONAL CONFERENCE ON COMPUTATIONAL AND INFORMATION SCIENCES》 * |
| 游江等: ""基于邻域贡献权值细化的圆心亚像素定位算法"", 《光学技术》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113077513A (en) * | 2021-06-03 | 2021-07-06 | 深圳市优必选科技股份有限公司 | Visual positioning method and device and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109344710B (en) | 2020-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9679192B2 (en) | 3-dimensional portrait reconstruction from a single photo | |
| CN110175558B (en) | Face key point detection method and device, computing equipment and storage medium | |
| US9008460B2 (en) | Automatic adjustment of images using a homography | |
| US8818131B2 (en) | Methods and apparatus for facial feature replacement | |
| WO2021135627A1 (en) | Method for constructing three-dimensional model of target object, and related apparatus | |
| WO2016029768A1 (en) | 3d human face reconstruction method and apparatus | |
| WO2021129146A1 (en) | Processing method and device for protecting privacy information in picture | |
| JP6528764B2 (en) | Face matching device, method, and recording medium | |
| US10467793B2 (en) | Computer implemented method and device | |
| Shim et al. | A subspace model-based approach to face relighting under unknown lighting and poses | |
| CN113689503B (en) | Target object posture detection method, device, equipment and storage medium | |
| WO2020037963A1 (en) | Facial image identifying method, device and storage medium | |
| US9727776B2 (en) | Object orientation estimation | |
| US10169891B2 (en) | Producing three-dimensional representation based on images of a person | |
| CN104156998A (en) | Implementation method and system based on fusion of virtual image contents and real scene | |
| CN107895173B (en) | Method, device and equipment for labeling image target and readable storage medium | |
| CN105096249B (en) | Image processing method and device | |
| CN111815768B (en) | Three-dimensional face reconstruction method and device | |
| KR20230162107A (en) | Facial synthesis for head rotations in augmented reality content | |
| WO2019216879A1 (en) | Perspective distortion correction on faces | |
| CN109934165A (en) | A kind of joint point detecting method, device, storage medium and electronic equipment | |
| US9860441B1 (en) | Tile-based digital image correspondence | |
| CN110956106A (en) | Processing method, device, storage medium and equipment for live broadcast | |
| KR20230014607A (en) | Method and apparatus for generating mega size augmented reality image information | |
| CN110188630A (en) | A kind of face identification method and camera |
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 |