CN106919917A - Face comparison method - Google Patents
Face comparison method Download PDFInfo
- Publication number
- CN106919917A CN106919917A CN201710101858.0A CN201710101858A CN106919917A CN 106919917 A CN106919917 A CN 106919917A CN 201710101858 A CN201710101858 A CN 201710101858A CN 106919917 A CN106919917 A CN 106919917A
- Authority
- CN
- China
- Prior art keywords
- face
- facial image
- track
- matching
- frame
- 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.)
- Pending
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of face comparison method, including：Feature extraction is carried out to the facial image in each frame of video flowing, the characteristic vector of the facial image is obtained；All characteristic vectors in same frame are matched with the face track stored in present frame, the facial image after matching belonging to each characteristic vector is to that there should be a Track ID；Every facial image in each frame is indexed, multiple Match ID of the correspondence facial image are obtained；Track ID and Match ID according to facial image in all frames carries out facial image matching and compares, and obtains the matching comparison result of the facial image.In the present invention, the one-sidedness that individual face picture database compares decision-making is overcome, improve final face alignment accuracy rate.
Description
Technical field
The present invention relates to technical field of face recognition, more particularly to a kind of face comparison method.
Background technology
Existing face comparison method is typically to extract feature to a face picture, then goes to carry out one in face characteristic storehouse
One matching, finds or several face pictures similar to its；Or obtain one using lasting method for detecting human face
The plurality of pictures of face is opened, feature is extracted, then goes to face characteristic storehouse repeatedly to be compared and inquired about, then design certain plan
Face alignment result is slightly given, but such case is usually under personnel's mated condition, with camera in geo-stationary and just
The state in face, such as face reading card device, intelligent entrance guard.
However, for above two face comparison method, being respectively provided with certain defect and existing.For the first side
Formula, its feature description by using individual face picture to face can in motion process with contingency, particularly face
There can be larger attitudes vibration, and face has the multi-angle information of three dimensions, therefore only voucher face figure in itself
The feature that piece is extracted is it is difficult to ensure that draw accurate comparison information；
For the second way, its lasting Face datection consumes larger to computing resource, if there is non-mated condition
Under Face datection, situations such as such as personnel move, then need plus Face tracking algorithm, required computing resource is more；More
Difficult situation is, if in the crowd scene that multiple faces coexist, in addition it is also necessary to using the method for multiple target tracking, Cai Nengbao
The card tracking effective to every face；Need to extract multiple face pictures in the case of along with this plurality of human faces picture
Feature, and then compare, the consumption to computing resource will be bigger, therefore in actual use, this kind of framework is difficult to obtain
Preferable real-time.
The content of the invention
In order to solve above mentioned problem of the prior art, i.e., in order to realize more accurately face alignment, the present invention is provided
A kind of face comparison method, including：
Feature extraction is carried out to the facial image in each frame of video flowing, the characteristic vector of the facial image is obtained；
All characteristic vectors in same frame are matched with the face track stored in present frame, each feature after matching
Facial image belonging to vector is to that should have a Track ID；
Every facial image in each frame is indexed, multiple Match ID of the correspondence facial image are obtained；
Track ID and Match ID according to facial image in all frames carries out facial image matching and compares, and obtains described
The matching comparison result of facial image.
Preferably, feature extraction is carried out to the facial image using deep neural network.
Preferably, before the facial image in each frame of video flowing carries out feature extraction, also include：It is each to video flowing
The face occurred in frame carries out face location detection, the facial image of each face location of output correspondence respectively.
Preferably, do what is stored in all characteristic vectors and present frame in same frame according to depth characteristic distance matrix metric
The matching of face track.
Preferably, it is described that all characteristic vectors in same frame are matched with the face track stored in present frame, have
Body includes：
If the characteristic vector matches face track, the facial image belonging to the characteristic vector obtains what is matched
The corresponding Track ID in face track；
If the characteristic vector does not match face track, the Face image synthesis belonging to the characteristic vector are new
Track ID。
Preferably, it is described that every facial image is indexed, multiple Match ID of the correspondence facial image are obtained,
Specifically include：
According to the characteristic vector of facial image, rope is carried out in multiple facial feature databases using quick indexing tree strategy
Draw, obtain the Match ID of facial image in each facial feature database matched with the characteristic vector of the facial image.
Preferably, Track ID and the Match ID according to the facial image carries out facial image matching comparison,
Specifically include：
Face images and the corresponding all Match ID of the facial image with identical Track ID are obtained,
The Match ID with maximum matching times are elected as matching comparison result.
Preferably, the Match ID with maximum matching times are elected by maximum votes decision-making mechanism and is used as matching
Comparison result
Compared with prior art, the present invention at least has advantages below：
Designed by the face alignment mode in the present invention, overcome individual face picture database and compare the unilateral of decision-making
Property, realize more accurately face alignment.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of the face comparison method that the present invention is provided.
Specific embodiment
The preferred embodiment of the present invention described with reference to the accompanying drawings.It will be apparent to a skilled person that this
A little implementation methods are used only for explaining know-why of the invention, it is not intended that limit the scope of the invention.
Multiple target tracking is the difficulties of field of machine vision, especially for this overall appearance of face than relatively similar
Target, under complicated crowd scene multiple human face targets are carried out with effectively tracking even more extremely difficult；It is demonstrated experimentally that conventional
Color, textural characteristics all it cannot be guaranteed that multiple human face targets are carried out effectively distinguish and track, and by deep neural network instruct
The depth characteristic that can express hidden feature for getting is more beneficial for carrying out face tracking, therefore this patent is proposed based on more
Plus the multiple target face tracking framework of the depth characteristic of reliable robust.
Based on this, in the present invention, there is provided a kind of face comparison method, as shown in figure 1, specifically including：
Step 101, face automatic detection.
Each frame in video flowing (including input or offline video file of on line real-time monitoring camera) is carried out certainly
Dynamic Face datection, obtains wherein all face locations, and the facial image of each face location of correspondence is exported afterwards；Wherein, it is acquired
Face location be a rectangle frame.
Step 102, the feature extraction of facial image.
In this step, feature extraction is carried out to facial image using deep neural network, obtains the facial image
Characteristic vector；After feature extraction, after facial image is projected into a characteristic vector, obtain the feature of the facial image to
Amount；Wherein, feature extraction is carried out for existing mode using the deep neural network, therefore specific treatment is not illustrated.
Step 103, the matching of face track following.
All characteristic vectors in same frame are matched with the face track stored in present frame, each feature after matching
Facial image belonging to vector is to that should have a Track ID.
If specifically, the characteristic vector matches face track, the facial image belonging to the characteristic vector is obtained
The corresponding Track ID in face track for matching；
If the characteristic vector does not match face track, the Face image synthesis belonging to the characteristic vector are new
Track ID。
For the facial image detected in every frame, matched with existing face track, matching is based on depth characteristic
Metric matrix；Illustrate：If present frame detects N number of face, there is M face track before present frame, then when
The multiple target face tracking problem of previous frame can be converted into N number of face and the M matching problem of face track, this matching here
Problem can be obtained by the best allotting algorithm to depth characteristic distance matrix metric, such as Hungarian Method, specific
Be can be described as follows with process：
In N number of face, the face of existing track is matched, continue to use the Track ID of existing track；
In N number of face, the face (usually emerging face) without matching track, newly-generated Track ID；
The face that every frame is detected is ensured with this, a Track ID is all corresponded to.
Step 104, facial image index.
Every facial image in each frame is indexed, multiple Match ID of the correspondence facial image are obtained；
Specifically, according to the characteristic vector of facial image, using quick indexing tree strategy in multiple facial feature databases
In be indexed, obtain facial image in each facial feature database matched with the characteristic vector of the facial image
Match ID。
Because in real time video processing, several faces to be detected to every frame will carry out the inquiry of database,
If million grades of facial feature databases of millions, inquiry velocity can turn into the bottleneck of system effectiveness.Therefore our logarithms
Disposable quick indexing tree is set up using the strategy of quick indexing tree according to the face depth characteristic in storehouse, it is this it is tactful relative to
Common sorting query, speed can lift tens times, further ensure the real-time of system；Every detected for every frame
Face, by quick indexing, can obtain the Match ID that several databases most like with its feature are compared.
Step 105, facial image matching is compared.
Track ID and Match ID according to facial image in all frames carries out facial image matching and compares, and obtains described
The matching comparison result of facial image；Specifically,
Face images and the corresponding all Match ID of the facial image with identical Track ID are obtained,
The Match ID with maximum matching times are elected as matching comparison result.
Operation based on step 103 and step 104, several can be corresponded to for a face, on sequential pursuit path (false
It is set to L) face picture with same Track ID, and every face picture, can correspond to again several (it is assumed that quantity is P)
Match ID, so relative to the system that only individual face picture is compared, this system all has for every decision-making of face
The similar Query Result of L*P database carries out decision support；From the decision-making mechanism of maximum votes, finally select with most
The Match ID of the face database of matching times as the final face database comparison result.
In the present invention, the feature for face tracking is multiplexed with the feature for face alignment, that is, make full use of depth
Feature effectively prevent influence of its larger computed losses to system real time again to the powerful sign ability of detailed information,
Improve the overall real-time performance of system；The multiple database of multiple face pictures based on reliable face tracking is quickly compared simultaneously
The maximum voting mechanism of result, for final face alignment decision-making provides decision support as much as possible, overcomes individual people
Face picture database compares the one-sidedness of decision-making, improves final face alignment accuracy rate.
So far, combined preferred embodiment shown in the drawings describes technical scheme, but, this area
Technical staff is it is easily understood that protection scope of the present invention is expressly not limited to these specific embodiments.Without departing from this
On the premise of the principle of invention, those skilled in the art can make equivalent change or replacement to correlation technique feature, these
Technical scheme after changing or replacing it is fallen within protection scope of the present invention.
Claims (8)
1. a kind of face comparison method, it is characterised in that including：
Feature extraction is carried out to the facial image in each frame of video flowing, the characteristic vector of the facial image is obtained；
All characteristic vectors in same frame are matched with the face track stored in present frame, each characteristic vector after matching
Affiliated facial image is to that should have a Track ID；
Every facial image in each frame is indexed, multiple Match ID of the correspondence facial image are obtained；
Track ID and Match ID according to facial image in all frames carries out facial image matching and compares, and obtains the face
The matching comparison result of image.
2. face comparison method according to claim 1, it is characterised in that using deep neural network to the face figure
As carrying out feature extraction.
3. face comparison method according to claim 1, it is characterised in that the facial image in each frame of video flowing enters
Before row feature extraction, also include：Face to occurring in each frame of video flowing carries out face location detection, output correspondence respectively
The facial image of each face location.
4. face comparison method according to claim 1, it is characterised in that done together according to depth characteristic distance matrix metric
All characteristic vectors are matched with the face track stored in present frame in one frame.
5. face comparison method according to claim 4, it is characterised in that it is described by all characteristic vectors in same frame with
The face track stored in present frame is matched, and is specifically included：
If the characteristic vector matches face track, the facial image belonging to the characteristic vector obtains the face for matching
The corresponding Track ID in track；
If the characteristic vector does not match face track, the Face image synthesis belonging to the characteristic vector are new
Track ID。
6. face comparison method according to claim 1, it is characterised in that described to be indexed to every facial image,
Multiple Match ID of the correspondence facial image are obtained, is specifically included：
According to the characteristic vector of facial image, it is indexed in multiple facial feature databases using quick indexing tree strategy,
Obtain the Match ID of facial image in each facial feature database matched with the characteristic vector of the facial image.
7. face comparison method according to claim 1, it is characterised in that the Track according to the facial image
ID and Match ID carry out facial image matching and compare, and specifically include：
Obtain face images and the corresponding all Match ID of the facial image with identical Track ID, election
Go out to have the Match ID of maximum matching times as matching comparison result.
8. face comparison method according to claim 7, it is characterised in that elected by maximum votes decision-making mechanism
Match ID with maximum matching times are used as matching comparison result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710101858.0A CN106919917A (en) | 2017-02-24 | 2017-02-24 | Face comparison method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710101858.0A CN106919917A (en) | 2017-02-24 | 2017-02-24 | Face comparison method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106919917A true CN106919917A (en) | 2017-07-04 |
Family
ID=59453743
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710101858.0A Pending CN106919917A (en) | 2017-02-24 | 2017-02-24 | Face comparison method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106919917A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657216A (en) * | 2017-09-11 | 2018-02-02 | 安徽慧视金瞳科技有限公司 | 1 to the 1 face feature vector comparison method based on interference characteristic vector data collection |
| CN108229322A (en) * | 2017-11-30 | 2018-06-29 | 北京市商汤科技开发有限公司 | Face identification method, device, electronic equipment and storage medium based on video |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102165464A (en) * | 2008-07-14 | 2011-08-24 | 谷歌公司 | Method and system for automated annotation of persons in video content |
| CN105488478A (en) * | 2015-12-02 | 2016-04-13 | 深圳市商汤科技有限公司 | Face recognition system and method |
| CN106022220A (en) * | 2016-05-09 | 2016-10-12 | 西安北升信息科技有限公司 | Method for performing multi-face tracking on participating athletes in sports video |
-
2017
- 2017-02-24 CN CN201710101858.0A patent/CN106919917A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102165464A (en) * | 2008-07-14 | 2011-08-24 | 谷歌公司 | Method and system for automated annotation of persons in video content |
| CN105488478A (en) * | 2015-12-02 | 2016-04-13 | 深圳市商汤科技有限公司 | Face recognition system and method |
| CN106022220A (en) * | 2016-05-09 | 2016-10-12 | 西安北升信息科技有限公司 | Method for performing multi-face tracking on participating athletes in sports video |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657216A (en) * | 2017-09-11 | 2018-02-02 | 安徽慧视金瞳科技有限公司 | 1 to the 1 face feature vector comparison method based on interference characteristic vector data collection |
| CN108229322A (en) * | 2017-11-30 | 2018-06-29 | 北京市商汤科技开发有限公司 | Face identification method, device, electronic equipment and storage medium based on video |
| CN108229322B (en) * | 2017-11-30 | 2021-02-12 | 北京市商汤科技开发有限公司 | Video-based face recognition method and device, electronic equipment and storage medium |
| US11068697B2 (en) | 2017-11-30 | 2021-07-20 | Beijing Sensetime Technology Development Co., Ltd | Methods and apparatus for video-based facial recognition, electronic devices, and storage media |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11188783B2 (en) | Reverse neural network for object re-identification | |
| KR102554724B1 (en) | Method for identifying an object in an image and mobile device for practicing the method | |
| Günther et al. | Unconstrained face detection and open-set face recognition challenge | |
| CN100568262C (en) | Human face recognition detection device based on the multi-video camera information fusion | |
| US20170032182A1 (en) | System for adaptive real-time facial recognition using fixed video and still cameras | |
| CN102902986A (en) | Automatic gender identification system and method | |
| CN106997629A (en) | Access control method, apparatus and system | |
| CN104091171B (en) | Vehicle-mounted far infrared pedestrian detecting system and method based on local feature | |
| CN102521565A (en) | Garment identification method and system for low-resolution video | |
| CN101587485B (en) | Face information automatic login method based on face recognition technology | |
| CN105138954A (en) | Image automatic screening, query and identification system | |
| CN103279768A (en) | Method for identifying faces in videos based on incremental learning of face partitioning visual representations | |
| CN101980242A (en) | Human face discrimination method and system and public safety system | |
| CN111539351B (en) | Multi-task cascading face frame selection comparison method | |
| Park et al. | Detection of construction workers in video frames for automatic initialization of vision trackers | |
| CN108846269A (en) | One kind is towards manifold identity identifying method and identification authentication system | |
| CN110728252A (en) | Face detection method applied to regional personnel motion trail monitoring | |
| CN112926522B (en) | Behavior recognition method based on skeleton gesture and space-time diagram convolution network | |
| CN101950448A (en) | Detection method and system for masquerade and peep behaviors before ATM (Automatic Teller Machine) | |
| CN114092877A (en) | Garbage can unattended system design method based on machine vision | |
| CN107862298A (en) | It is a kind of based on the biopsy method blinked under infrared eye | |
| CN106919917A (en) | Face comparison method | |
| Nosheen et al. | Efficient Vehicle Detection and Tracking using Blob Detection and Kernelized Filter | |
| CN114581990A (en) | Intelligent running test method and device | |
| CN103971100A (en) | Video-based camouflage and peeping behavior detection method for automated teller machine |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170704 |
|
| RJ01 | Rejection of invention patent application after publication |