CN110427843A - A kind of face intelligent identification Method - Google Patents
A kind of face intelligent identification Method Download PDFInfo
- Publication number
- CN110427843A CN110427843A CN201910651887.3A CN201910651887A CN110427843A CN 110427843 A CN110427843 A CN 110427843A CN 201910651887 A CN201910651887 A CN 201910651887A CN 110427843 A CN110427843 A CN 110427843A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- atlas
- piecemeal
- personage
- 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
- G06V40/161—Detection; Localisation; Normalisation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of face intelligent identification Method, which comprises in the case where that can not identify personage to be identified, select original graph；It positions and tracks the personage to be identified；Facial image and original figure are obtained into original atlas, the facial image for acquiring the personage to be identified of at least one different angle is correction atlas；Missing identification sample set and complete face contour area comprising front face image sample are obtained by adjusting the human face posture of original atlas and the facial image for correcting atlas；Piecemeal is carried out to the face sample in missing identification sample set according to complete face contour area and forms block regional ensemble；Labeled as selected piecemeal；It is restored after obtaining amendment image using selected piecemeal, recognition of face is carried out to amendment image.The problem of solving conventional face's identification caused image section region after being adjusted human face posture to be distorted or can not identify, and then causing the face of personage to be identified that can not be correctly validated.
Description
Technical field
The present invention relates to facial images to identify field, more particularly, to a kind of face intelligent identification Method.
Background technique
Networking or some independent integration tools by means of camera and computer, face recognition technology have been modern
A very universal and common technology in people's life.People can be remotely or intelligently by means of face recognition technology
Complete the identification to face.But traditional recognition of face is highly dependent on the direction and camera shooting of a certain range of personage to be identified
The shooting angle of head.In the image comprising different human face postures of acquisition, recognition of face is often due to excessive angle
And its discrimination is caused exponentially to decline.
For described problem, most-often used mostly of the prior art be concentrate on to the two dimension and threedimensional model of facial image it
Between conversion, but the method is very high to the modeling demand of threedimensional model, and resource requirement needed for identifying face is too high, in reality
It is often difficult to reach the resource requirement of the method in scene.Based on described problem, the prior art proposes a kind of using polygonal
The optimal synthetic method of facial image is spent, but the method not only needs user to be identified the feature locations in image,
And it is different due to plurality of picture angle, the single region carried out is compared often inaccurate；The prior art
Another recognition methods, which also proposed, carries out recognition of face directly against the facial image after pose adjustment, but facial image carries out appearance
State adjustment after often due to shooting angle and cause facial image can not completely restore or part restore region distortion.Therefore in needle
To conventional face's identification, caused image section region is distorted or can not identify after being adjusted to human face posture, and then causes
On the problem of face of personage to be identified can not be correctly validated, the prior art does not have very good solution yet.
Summary of the invention
The present invention is directed to overcome above-mentioned prior art problem, a kind of face intelligent identification Method is provided, the method is used for
It solves conventional face's identification caused image section region after being adjusted human face posture to be distorted or can not identify, and then leads
The problem of causing the face of personage to be identified can not be correctly validated.
A kind of face intelligent identification Method, which comprises
It obtains images to be recognized and identifies the personage to be identified in image；
In the case where that can not identify personage to be identified, the facial image for selecting the personage to be identified is original graph；
It is positioned by the image acquisition source of original graph and tracks the personage to be identified；
Acquisition time and identical image acquisition mode based on original graph obtain immediate at least with the original graph time
The facial image and original figure of one additional personage to be identified at original atlas, according to the orientation of the positioning and tracking by
Assisting face collector to acquire the facial image of the personage to be identified of at least one different angle is correction atlas；
Human face posture by adjusting original atlas and the facial image for correcting atlas is obtained comprising front face image sample
This missing identification sample set and complete face contour area；
Piecemeal shape agglomerated regions collection is carried out to the face sample in missing identification sample set according to complete face contour area
It closes；
It is selected piecemeal by the sample piecemeal result queue for having highest resolution in each piece of regional ensemble；
It facial image is carried out in complete face contour area using the symmetry of selected piecemeal and face restores to be repaired
After positive image, recognition of face is carried out to amendment image.
This method has used the facial image of other angles to the first identification figure of the facial image of the personage to be identified
Identify solve due to resulting in the distortion of part identification region after the adjustment of single angular pose or can not identify after being restored again
The problem of face of personage to be identified caused by and then can not be correctly validated, and the method can be selection Facenet mould
The identification of type provides the less image sources of background interference of more reliable and non-personage's face to be identified and then more efficient knowledge
Face in other image.
Wherein, the method has first carried out adjustment to human face posture and has carried out piecemeal to facial image adjusted again, avoids
Directly again carry out after angle meter and piecemeal uneven to the segmented areas of sample caused by piecemeal to non-frontal image
Location error is larger, and then the problem for causing subsequent specimen discerning error larger, and saves and carry out individually to segmented areas
Angle meter the step of.
Preferably, the image by original graph obtains the step of source positions and tracks the personage to be identified and specifically wraps
It includes:
The lens focus and camera lens when source obtains the facial image of the personage to be identified are obtained by the image of original graph
Direction determines the cardinal points of the personage to be identified；
The personage of the cardinal points is accurately positioned by the camera of at least one other observation angle and with
Track.
Knowledge is treated using other auxiliary cameras after source positions by obtaining to personage to be identified by the first image
Others' object continue tracking to guarantee the subsequent processing to identification sample, so that other non-personages to be identified or background are to identification
The degree of disturbance of sample declines to a great extent, to meet degree of purity requirement of the method to identification sample.
Preferably, the human face posture by adjusting original atlas and the facial image for correcting atlas is obtained comprising front
The missing of facial image sample identifies that the step of sample set specifically includes:
The face of the personage to be identified in the facial image of the original atlas and correction atlas is separated as identification sample set；
Determine the human face posture of identification sample set；
Basic front human face outline region is defined by human face posture and original atlas；
Symmetry based on face middle line mentions the completion of basic front human face outline region after complete face contour area
The facial contour of the personage to be identified in correction atlas is taken to be modified complete face contour area；
The posture calibrated original atlas by complete face contour area and correct the face sample in atlas is obtained to positive
Sample set is identified to missing.
Extracting and defining effective facial contour is one of the main means for prompting the identification validity of the method, is based on
Data set defined in Facenet model can effectively extract the key data feature and its parameter of identification facial contour, be based on institute
State the input picture analysis of feature and parameter as a result, can Euclidean distance to the feature vector of the facial contour of personage to be identified into
Row calculates and restores facial contour region and is positive towards the certain area in plane.
Preferably, described that original atlas is calibrated by complete face contour area and corrects the appearance of the face sample in atlas
State to front obtain missing identification sample set the step of specifically include:
The human face posture parameter of personage to be identified is calculated by complete face contour area, the human face posture parameter is positive
Dough figurine face contour area personage to be identified in the plane domain and the identification sample set in one group of determining Eulerian angles
Human face characteristic point it is described with Eulerian angles determine plane domain on projector distance mean value；
It is modified to obtain positive face using posture of the face attitude parameter to identification sample set as missing identification sample
This collection.
Since the method based on Facenet identification model is a kind of identification method based on image pattern, identification process is held
Uncertainty vulnerable to human face posture is influenced, but distance between feature is easy to cause if ignoring human face posture and being identified
Confusion and sample between the identification mistake that is generated since integrity degree is different, the method is emphatically to correct face appearance
Image after state is as identification sample and provides a kind of reliable pose adjustment mode to guarantee to identify the reliability and knowledge of sample
Other accuracy rate.
Wherein, the missing identification sample set is the complete face after to identifying that human face posture is adjusted in sample set
The front projection region after pose adjustment in contour area comprising original sample and the absent region outside front projection.
Preferably, the step of face sample in described pair of missing identification sample set carries out piecemeal formation block regional ensemble has
Body includes:
Segmented areas is formed based on complete face contour area, according to segmented areas in original atlas and correction atlas
The face sample that missing identification sample set includes carries out piecemeal；
Piecemeal result is marked based on segmented areas；
The set in each piece of region is formed according to label result.
Piecemeal result is carried out classifying directly to save establishing the step of pixel coordinate is to classify by block region, and by
In the method be handle pose adjustment after facial image, the resolution ratio or clarity of the facial image all not be
Being uniformly distributed in one plane, and the block region deviding mode based on facial contour region can more effectively separate picture area
Domain is to reach independent purpose between its region.
Preferably, described to divide the sample piecemeal result queue for having highest resolution in each piece of regional ensemble to be selected
The step of block, specifically includes:
It verifies each piecemeal result simultaneously including original atlas sample and correction atlas sample and gathers intermediate-resolution most
Big optimal piecemeal result belongs in the block regional ensemble of correction atlas, in the optimal piecemeal result and same regional ensemble
The piecemeal result for belonging to original atlas sample compare, if having minimum similarity needed for regarding as similar block, wherein
After the minimum similarity is converted to grayscale image for the histogram equalization that a pair of of image carries out the image based on lowest resolution,
Minimum Pasteur's distance of down-sampling treated the pixel grayscale distribution of image pyramid is carried out to the grayscale image；
If the optimal piecemeal result is compared with the piecemeal result for belonging to original atlas sample in same regional ensemble
Have minimum similarity needed for regarding as similar block, then marks the optimal piecemeal result for the choosing in the block region belonging to it
Determine piecemeal, and it is selected for marking the piecemeal result queue of the highest resolution in the block regional ensemble of remaining unmarked selected piecemeal
Piecemeal.
Preferably, the symmetry using selected piecemeal and face carries out facial image in complete face contour area
After recovery obtains amendment image, the step of amendment image progress recognition of face, is specifically included:
Piecemeal is selected according to the affiliated block regional ensemble of the selected piecemeal in complete face contour area
Home position；
Facial image is restored using selected piecemeal according to the home position；
Facial image after recovery is converted to by feature vector using Facenet depth convolutional network；
Between the feature vector stored in the feature vector of recovery facial image after calculating normalized and identification library
After Euclidean distance, corresponding triple is obtained by Euclidean distance between described eigenvector；
Face recognition result is formed by the triple.
The picture handled through the method can be effectively used for Facenet model and be identified, using recovery picture to mention
The precision and range of fit of high Facenet identification.
Compared with prior art, the invention has the benefit that
1. solving conventional face's identification caused image section region distortion or nothing after being adjusted to human face posture
Method identification, and then the problem of cause the face of personage to be identified that can not be correctly validated；
2. the identification method based on Facenet makes the precision of identification higher；
3. being carried out using Multi-angle human face image preferentially complementary so that the facial image of the frontal pose after restoring is easier to
It is identified in by the identification method based on Facenet identification model；
4. the personage to be identified of the location tracking of a camera more than ensure that the independent pure of identification sample；
5. the set of the piecemeal based on facial contour region, which solves, is necessarily dependent upon the partitioned mode of pixel coordinate in turn
Lead to the problem that the method is irregular to sample piecemeal.
Detailed description of the invention
Fig. 1 is the flow chart of this method.
Fig. 2 is the pose adjustment and piecemeal schematic diagram of this method.
Specific embodiment
Attached drawing of the present invention only for illustration, is not considered as limiting the invention.It is following in order to more preferably illustrate
Embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent the size of actual product；For art technology
For personnel, the omitting of some known structures and their instructions in the attached drawings are understandable.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
As depicted in figs. 1 and 2, a kind of face intelligent identification Method is present embodiments provided, which comprises
It obtains images to be recognized and identifies the personage to be identified in image；
In the case where that can not identify personage to be identified, the facial image for selecting the personage to be identified is original graph；
It is positioned by the image acquisition source of original graph and tracks the personage to be identified；
Acquisition time and identical image acquisition mode based on original graph obtain immediate at least with the original graph time
The facial image and original figure of one additional personage to be identified at original atlas, according to the orientation of the positioning and tracking by
Assisting face collector to acquire the facial image of the personage to be identified of at least one different angle is correction atlas；
Human face posture by adjusting original atlas and the facial image for correcting atlas is obtained comprising front face image sample
This missing identification sample set and complete face contour area；
Piecemeal shape agglomerated regions collection is carried out to the face sample in missing identification sample set according to complete face contour area
It closes；
It is selected piecemeal by the sample piecemeal result queue for having highest resolution in each piece of regional ensemble；
It facial image is carried out in complete face contour area using the symmetry of selected piecemeal and face restores to be repaired
After positive image, recognition of face is carried out to amendment image.
This method has used the facial image of other angles to the first identification figure of the facial image of the personage to be identified
Identify solve due to resulting in the distortion of part identification region after the adjustment of single angular pose or can not identify after being restored again
The problem of face of personage to be identified caused by and then can not be correctly validated, and the method can be selection Facenet mould
The identification of type provides the less image sources of background interference of more reliable and non-personage's face to be identified and then more efficient knowledge
Face in other image.
Wherein, the method has first carried out adjustment to human face posture and has carried out piecemeal to facial image adjusted again, avoids
Directly again carry out after angle meter and piecemeal uneven to the segmented areas of sample caused by piecemeal to non-frontal image
Location error is larger, and then the problem for causing subsequent specimen discerning error larger, and saves and carry out individually to segmented areas
Angle meter the step of.
In the specific implementation process, the image by original graph obtains source and positions and track the personage's to be identified
Step specifically includes:
The lens focus and camera lens when source obtains the facial image of the personage to be identified are obtained by the image of original graph
Direction determines the cardinal points of the personage to be identified；
The personage of the cardinal points is accurately positioned by the camera of at least one other observation angle and with
Track.
Knowledge is treated using other auxiliary cameras after source positions by obtaining to personage to be identified by the first image
Others' object continue tracking to guarantee the subsequent processing to identification sample, so that other non-personages to be identified or background are to identification
The degree of disturbance of sample declines to a great extent, to meet degree of purity requirement of the method to identification sample.
Specifically, the human face posture by adjusting original atlas and the facial image for correcting atlas is obtained comprising front
The missing of facial image sample identifies that the step of sample set specifically includes:
The face of the personage to be identified in the facial image of the original atlas and correction atlas is separated as identification sample set；
Determine the human face posture of identification sample set；
Basic front human face outline region is defined by human face posture and original atlas；
Symmetry based on face middle line mentions the completion of basic front human face outline region after complete face contour area
The facial contour of the personage to be identified in correction atlas is taken to be modified complete face contour area；
The posture calibrated original atlas by complete face contour area and correct the face sample in atlas is obtained to positive
Sample set is identified to missing.
Extracting and defining effective facial contour is one of the main means for prompting the identification validity of the method, is based on
Data set defined in Facenet model can effectively extract the key data feature and its parameter of identification facial contour, be based on institute
State the input picture analysis of feature and parameter as a result, can Euclidean distance to the feature vector of the facial contour of personage to be identified into
Row calculates and restores facial contour region and is positive towards the certain area in plane.
Specifically, described calibrate original atlas by complete face contour area and correct the appearance of the face sample in atlas
State to front obtain missing identification sample set the step of specifically include:
The human face posture parameter of personage to be identified is calculated by complete face contour area, the human face posture parameter is positive
Dough figurine face contour area personage to be identified in the plane domain and the identification sample set in one group of determining Eulerian angles
Human face characteristic point it is described with Eulerian angles determine plane domain on projector distance mean value；
It is modified to obtain positive face using posture of the face attitude parameter to identification sample set as missing identification sample
This collection.
Since the method based on Facenet identification model is a kind of identification method based on image pattern, identification process is held
Uncertainty vulnerable to human face posture is influenced, but distance between feature is easy to cause if ignoring human face posture and being identified
Confusion and sample between the identification mistake that is generated since integrity degree is different, the method is emphatically to correct face appearance
Image after state is as identification sample and provides a kind of reliable pose adjustment mode to guarantee to identify the reliability and knowledge of sample
Other accuracy rate.
Wherein, the missing identification sample set is the complete face after to identifying that human face posture is adjusted in sample set
The front projection region after pose adjustment in contour area comprising original sample and the absent region outside front projection, wherein Fig. 2
In Lack be absent region.
Specifically, the face sample in described pair of missing identification sample set carries out the step of piecemeal forms block regional ensemble and has
Body includes:
Segmented areas is formed based on complete face contour area, according to segmented areas in original atlas and correction atlas
The face sample that missing identification sample set includes carries out piecemeal；
Piecemeal result is marked based on segmented areas；
The set in each piece of region is formed according to label result.
Piecemeal result is carried out classifying directly to save establishing the step of pixel coordinate is to classify by block region, and by
In the method be handle pose adjustment after facial image, the resolution ratio or clarity of the facial image all not be
Being uniformly distributed in one plane, and the block region deviding mode based on facial contour region can more effectively separate picture area
Domain is to reach independent purpose between its region.
Wherein, if n block block region is divided into facial contour region, then n block regional ensemble { S is defined1,S2,S3,…,
Sn, the sample of label is sorted out according to the region that block region is divided, such as set S1M sample for being included then has setIt willIt is defined as the average or maximum resolution of m-th of sample in block regional ensemble 1
Or clarity；IfCorresponding piece of region includes absent region, then defines
Specifically, described divide the sample piecemeal result queue for having highest resolution in each piece of regional ensemble to be selected
The step of block, specifically includes:
It verifies each piecemeal result simultaneously including original atlas sample and correction atlas sample and gathers intermediate-resolution most
Big optimal piecemeal result belongs in the block regional ensemble of correction atlas, in the optimal piecemeal result and same regional ensemble
The piecemeal result for belonging to original atlas sample compare, if having minimum similarity needed for regarding as similar block, wherein
After the minimum similarity is converted to grayscale image for the histogram equalization that a pair of of image carries out the image based on lowest resolution,
Minimum Pasteur's distance of down-sampling treated the pixel grayscale distribution of image pyramid is carried out to the grayscale image；
If the optimal piecemeal result is compared with the piecemeal result for belonging to original atlas sample in same regional ensemble
Have minimum similarity needed for regarding as similar block, then marks the optimal piecemeal result for the choosing in the block region belonging to it
Determine piecemeal, and it is selected for marking the piecemeal result queue of the highest resolution in the block regional ensemble of remaining unmarked selected piecemeal
Piecemeal.
Wherein, the selected piecemeal p ' of n-th piece of regional ensemblenSelected mode be
Specifically, the symmetry using selected piecemeal and face carries out facial image in complete face contour area
After recovery obtains amendment image, the step of amendment image progress recognition of face, is specifically included:
Piecemeal is selected according to the affiliated block regional ensemble of the selected piecemeal in complete face contour area
Home position；
Facial image is restored using selected piecemeal according to the home position；
Facial image after recovery is converted to by feature vector using Facenet depth convolutional network；
Between the feature vector stored in the feature vector of recovery facial image after calculating normalized and identification library
After Euclidean distance, corresponding triple is obtained by Euclidean distance between described eigenvector；
Face recognition result is formed by the triple.
The picture handled through the method can be effectively used for Facenet model and be identified, using recovery picture to mention
The precision and range of fit of high Facenet identification.
Wherein, mean value is calculated for the resolution ratio of restored image, uses the laplacian pyramid based on face database
And gaussian pyramid is adjusted the resolution ratio of restored image, is uniformly distributed the pixel of restored image to meet recognition of face
Identification requirement.
Obviously, the above embodiment of the present invention is only intended to clearly illustrate technical solution of the present invention example, and
It is not the restriction to a specific embodiment of the invention.It is all made within the spirit and principle of claims of the present invention
Any modifications, equivalent replacements, and improvements etc., should all be included in the scope of protection of the claims of the present invention.
Claims (7)
1. a kind of face intelligent identification Method, it is characterised in that:
It obtains images to be recognized and identifies the personage to be identified in image；
In the case where that can not identify personage to be identified, the facial image for selecting the personage to be identified is original graph；
It is positioned by the image acquisition source of original graph and tracks the personage to be identified；
Acquisition time and identical image acquisition mode based on original graph obtain and the original graph time immediate at least one
The facial image and original figure of additional personage to be identified is at original atlas, according to the orientation of the positioning and tracking by assisting
The facial image that face collector acquires the personage to be identified of at least one different angle is correction atlas；
Human face posture by adjusting original atlas and the facial image for correcting atlas is obtained comprising front face image sample
Missing identification sample set and complete face contour area；
Piecemeal is carried out to the face sample in missing identification sample set according to complete face contour area and forms block regional ensemble；
It is selected piecemeal by the sample piecemeal result queue for having highest resolution in each piece of regional ensemble；
Facial image is carried out in complete face contour area using the symmetry of selected piecemeal and face to restore to obtain correction map
As after, recognition of face is carried out to amendment image.
2. a kind of face intelligent identification Method according to claim 1, which is characterized in that the image by original graph
Acquisition source positions and specifically includes the step of tracking the personage to be identified:
Lens focus and camera lens direction when source obtains the facial image of the personage to be identified are obtained by the image of original graph,
Determine the cardinal points of the personage to be identified；
The personage of the cardinal points is accurately positioned and is tracked by the camera of at least one other observation angle.
3. a kind of face intelligent identification Method according to claim 1, which is characterized in that described by adjusting original atlas
And correction atlas facial image human face posture obtain include front face image sample missing identify sample set the step of
It specifically includes:
The face of the personage to be identified in the facial image of the original atlas and correction atlas is separated as identification sample set；
Determine the human face posture of identification sample set；
Basic front human face outline region is defined by human face posture and original atlas；
Symmetry based on face middle line rectifys the completion of basic front human face outline region to extract after complete face contour area
The facial contour of personage to be identified in positive atlas is modified complete face contour area；
It is lacked by posture to the front that complete face contour area calibrates original atlas and corrects the face sample in atlas
Lose identification sample set.
4. a kind of face intelligent identification Method according to claim 3, which is characterized in that described to pass through complete facial contour
Posture to the front of the original atlas of regional calibration and the face sample in correction atlas obtains the step of missing identification sample set tool
Body includes:
Calculate the human face posture parameter of personage to be identified by complete face contour area, the human face posture parameter is positive dough figurine
The people of face contour area personage to be identified in the plane domain and the identification sample set in one group of determining Eulerian angles
Projector distance mean value of the face characteristic point on the plane domain determined with Eulerian angles；
It is modified to obtain positive face using posture of the face attitude parameter to identification sample set as missing identification sample set.
5. a kind of face intelligent identification Method according to claim 1, which is characterized in that described pair of missing identifies sample set
In face sample carry out piecemeal formed block regional ensemble the step of specifically include:
Segmented areas is formed based on complete face contour area, according to segmented areas to the missing in original atlas and correction atlas
The face sample that identification sample set includes carries out piecemeal；
Piecemeal result is marked based on segmented areas；
The set in each piece of region is formed according to label result.
6. a kind of face intelligent identification Method according to claim 1, which is characterized in that described by each piece of regional ensemble
In have the step of sample piecemeal result queue of highest resolution is selected piecemeal and specifically include:
Verifying each includes the piecemeal result of original atlas sample and correction atlas sample simultaneously and set intermediate-resolution is maximum
Optimal piecemeal result belongs in the block regional ensemble of correction atlas, the category in the optimal piecemeal result and same regional ensemble
It is compared in the piecemeal result of original atlas sample, if having minimum similarity needed for regarding as similar block, wherein described
Minimum similarity is after the histogram equalization that a pair of of image carries out the image based on lowest resolution is converted to grayscale image, to institute
State minimum Pasteur's distance that grayscale image carries out down-sampling treated the pixel grayscale distribution of image pyramid；
If the optimal piecemeal result has compared with the piecemeal result for belonging to original atlas sample in same regional ensemble
Minimum similarity needed for similar block is regarded as, then marks the optimal piecemeal result for selected point of the block region belonging to it
Block, and marking the piecemeal result queue of the highest resolution in the block regional ensemble of remaining unmarked selected piecemeal is selected point
Block.
7. a kind of face intelligent identification Method according to claim 1, which is characterized in that described to use selected piecemeal and people
The symmetry of face carries out facial image in complete face contour area and restores after obtaining amendment image, carries out people to amendment image
The step of face identifies specifically includes:
Recovery of the piecemeal in complete face contour area is selected according to the affiliated block regional ensemble of the selected piecemeal
Position；
Facial image is restored using selected piecemeal according to the home position；
Facial image after recovery is converted to by feature vector using Facenet depth convolutional network；
Euclidean between the feature vector stored in the feature vector of recovery facial image after calculating normalized and identification library
After distance, corresponding triple is obtained by Euclidean distance between described eigenvector；
Face recognition result is formed by the triple.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910651887.3A CN110427843B (en) | 2019-07-18 | 2019-07-18 | Intelligent face recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910651887.3A CN110427843B (en) | 2019-07-18 | 2019-07-18 | Intelligent face recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110427843A true CN110427843A (en) | 2019-11-08 |
| CN110427843B CN110427843B (en) | 2021-07-13 |
Family
ID=68411149
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910651887.3A Active CN110427843B (en) | 2019-07-18 | 2019-07-18 | Intelligent face recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110427843B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000163592A (en) * | 1998-11-26 | 2000-06-16 | Victor Co Of Japan Ltd | Recognizing method for face image |
| CN103605965A (en) * | 2013-11-25 | 2014-02-26 | 苏州大学 | Multi-pose face recognition method and device |
| KR101491832B1 (en) * | 2014-05-23 | 2015-02-12 | 동국대학교 산학협력단 | Apparatus and method for selecting image |
| CN108564053A (en) * | 2018-04-24 | 2018-09-21 | 南京邮电大学 | Multi-cam dynamic human face recognition system based on FaceNet and method |
| CN109284722A (en) * | 2018-09-29 | 2019-01-29 | 佳都新太科技股份有限公司 | Image processing method, device, face recognition device and storage medium |
| CN109360170A (en) * | 2018-10-24 | 2019-02-19 | 北京工商大学 | Face restorative procedure based on advanced features |
| CN109377518A (en) * | 2018-09-29 | 2019-02-22 | 佳都新太科技股份有限公司 | Target tracking method, device, target tracking equipment and storage medium |
| CN109598210A (en) * | 2018-11-16 | 2019-04-09 | 三星电子（中国）研发中心 | A kind of image processing method and device |
| CN109902657A (en) * | 2019-03-12 | 2019-06-18 | 哈尔滨理工大学 | A kind of face identification method indicated based on piecemeal collaboration |
| CN109918998A (en) * | 2019-01-22 | 2019-06-21 | 东南大学 | A kind of big Method of pose-varied face |
-
2019
- 2019-07-18 CN CN201910651887.3A patent/CN110427843B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000163592A (en) * | 1998-11-26 | 2000-06-16 | Victor Co Of Japan Ltd | Recognizing method for face image |
| CN103605965A (en) * | 2013-11-25 | 2014-02-26 | 苏州大学 | Multi-pose face recognition method and device |
| KR101491832B1 (en) * | 2014-05-23 | 2015-02-12 | 동국대학교 산학협력단 | Apparatus and method for selecting image |
| CN108564053A (en) * | 2018-04-24 | 2018-09-21 | 南京邮电大学 | Multi-cam dynamic human face recognition system based on FaceNet and method |
| CN109284722A (en) * | 2018-09-29 | 2019-01-29 | 佳都新太科技股份有限公司 | Image processing method, device, face recognition device and storage medium |
| CN109377518A (en) * | 2018-09-29 | 2019-02-22 | 佳都新太科技股份有限公司 | Target tracking method, device, target tracking equipment and storage medium |
| CN109360170A (en) * | 2018-10-24 | 2019-02-19 | 北京工商大学 | Face restorative procedure based on advanced features |
| CN109598210A (en) * | 2018-11-16 | 2019-04-09 | 三星电子（中国）研发中心 | A kind of image processing method and device |
| CN109918998A (en) * | 2019-01-22 | 2019-06-21 | 东南大学 | A kind of big Method of pose-varied face |
| CN109902657A (en) * | 2019-03-12 | 2019-06-18 | 哈尔滨理工大学 | A kind of face identification method indicated based on piecemeal collaboration |
Non-Patent Citations (5)
| Title |
|---|
| JIAWAN ZHANG等: ""Symmetry-Aware Face Completion with Generative Adversarial Networks"", 《ACCV 2018: COMPUTER VISION》 * |
| LEONARDO A. CAMENT等: ""Face recognition under pose variation with local Gabor features enhanced by Active Shape and Statistical Models"", 《PATTERN RECOGNITION》 * |
| 王亚南: ""多姿态人脸图像识别系统"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 翟东海等: ""破损区域分块划分的图像修复"", 《中国图象图形学报》 * |
| 谢鹏程: ""多姿态人脸识别的研究与实现"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110427843B (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106055091B (en) | A kind of hand gestures estimation method based on depth information and correcting mode | |
| JP6448223B2 (en) | Image recognition system, image recognition apparatus, image recognition method, and computer program | |
| CN107813310B (en) | Multi-gesture robot control method based on binocular vision | |
| CN103577815B (en) | A kind of face alignment method and system | |
| CN108717531B (en) | Human body posture estimation method based on Faster R-CNN | |
| CN102087703B (en) | The method determining the facial pose in front | |
| US10189162B2 (en) | Model generation apparatus, information processing apparatus, model generation method, and information processing method | |
| CN105740780B (en) | Method and device for detecting living human face | |
| CN108985137A (en) | A kind of licence plate recognition method, apparatus and system | |
| CN111784775B (en) | Identification-assisted visual inertia augmented reality registration method | |
| CN101377814A (en) | Face image processing apparatus, face image processing method, and computer program | |
| CN101968846A (en) | Face tracking method | |
| JP7147753B2 (en) | Information processing device, information processing method, and program | |
| CN102982524B (en) | Splicing method for corn ear order images | |
| CN111209874B (en) | Method for analyzing and identifying wearing attribute of human head | |
| WO2015165227A1 (en) | Human face recognition method | |
| CN113255608B (en) | Multi-camera face recognition positioning method based on CNN classification | |
| CN112017212B (en) | Training and tracking method and system of face key point tracking model | |
| JP6922348B2 (en) | Information processing equipment, methods, and programs | |
| Savinykh et al. | Darkslam: Gan-assisted visual slam for reliable operation in low-light conditions | |
| WO2022018811A1 (en) | Three-dimensional posture of subject estimation device, three-dimensional posture estimation method, and program | |
| CN111898552B (en) | Method and device for distinguishing person attention target object and computer equipment | |
| CN110427843A (en) | A kind of face intelligent identification Method | |
| CN108694348B (en) | Tracking registration method and device based on natural features | |
| CN116563391A (en) | Automatic laser structure calibration method based on machine vision |
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 |