CN109961046A - The video flowing face identification method of building dynamic sample collection is recalled based on key frame - Google Patents
The video flowing face identification method of building dynamic sample collection is recalled based on key frame Download PDFInfo
- Publication number
- CN109961046A CN109961046A CN201910231632.1A CN201910231632A CN109961046A CN 109961046 A CN109961046 A CN 109961046A CN 201910231632 A CN201910231632 A CN 201910231632A CN 109961046 A CN109961046 A CN 109961046A
- Authority
- CN
- China
- Prior art keywords
- frame
- sample set
- kth
- identification
- face
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of video flowing face identification methods that building dynamic sample collection is recalled based on key frame, this method can effectively solve the problem that face recognition technology generated in practical applications: personal user, which independently provides large-scale sample data set, has biggish difficulty, sample set update slower, it cannot reflect that newest face visual information, the quality of data of sample set of people to be identified are irregular, in time to carry out accurately identifying the problems such as causing negative effect.The present invention is according to the particular content and actual recognition result of specific video flowing to be measured, it is adjacent using video flowing or close on similitude between frame image, current sample set is updated and is expanded, achieve the purpose that dynamic construction high quality samples collection, is a kind of face identification method that can be timely updated, construct high quality samples collection automatically.
Description
Technical field
The invention belongs to image identification technical fields, are related to a kind of face identification method, in particular to a kind of based on key
The video flowing face identification method of frame backtracking building dynamic sample collection.
Background technique
Recognition of face refers to comparing face visual information computer technology for identification using analysis.Generally
For, face identification system includes image acquisition, image procossing, Face datection, recognition of face.In face recognition process,
Sample set is reasonably constructed, forms the face visual information recognition classifier of personage to be identified by machine learning training.
Common method has geometrical measurers, sub-space analysis method, statistical nature method etc. in recognition of face at present.Although close
Field of face identification makes great progress over year, but implements facial visual information verifying and identification on a large scale or to traditional
Method brings stern challenge.If guaranteeing higher recognition accuracy, these methods have certain requirement to sample set,
Following main problem is produced in practical applications:
(1) personal user, which independently provides large-scale sample data set, biggish difficulty；
(2) sample set updates slower, cannot reflect the newest face visual information of people to be identified in time；
(3) quality of data of sample set is irregular, causes negative effect accurately identify.
Currently, there is an urgent need to one kind to timely update, construct the face identification method of high quality samples collection automatically.
Summary of the invention
In order to solve the above-mentioned technical problem, the invention proposes it is a kind of can be according to the particular content of video flowing to be measured in time more
Newly, the video flowing face identification method that building dynamic sample collection is recalled based on key frame of high quality samples collection is constructed.
The technical scheme adopted by the invention is that: a kind of video flowing face for recalling building dynamic sample collection based on key frame
Recognition methods, it is characterised in that: adjacent using video flowing according to the content of specific video flowing to be measured and actual recognition result
Or similitude between frame image is closed on, current sample set is updated and is expanded, dynamic construction high quality samples are reached
The purpose of collection；
Step includes:
Step 1: being set according to actual conditions by user and adjusted threshold value T, guarantee identification under the premise of allowing missing inspection
Confidence level carries out general recognition of face 90% or more；
In face recognition process, the result that threshold value is used to define identification is success or failure.That associated is people
The confidence level of face identification, the value range of confidence level are [0%, 100%], and expression identifies a possibility that correct when identifying successfully.
Such as confidence level is 80%, indicates that the face currently identified has 80% a possibility that for the face of personage to be measured.Confidence level by
The result statistics of identification obtains, refers to the ratio for identifying the total identification number of correct face number Zhan.This patent is adjusted by dynamic
Save threshold value T, confidence level be bound so that in general face recognition process under the premise of allowing missing inspection, identification can
Reliability is 90% or more.For example, the value range of threshold value is [1,10], and when threshold value is 3, confidence level 40%；When threshold value is 6
When, confidence level 80%, such as when threshold value is 8, confidence level reaches 92%.Therefore, it when recognition result is greater than 8, is considered as working as
Preceding recognition result has the identification of 92% a possibility that correct.It is set in the explanation of this patent, when recognition result R is less than threshold value T
When, it is believed that recognition failures；When recognition result R is greater than or equal to threshold value T, it is believed that identify successfully.This patent meets the value of threshold value T
It is set and is adjusted according to the actual situation by user.
Step 2: finding in current identification segment and identify successful first frame FirstFrame, will wherein identify successful people
Face partial image FaceCut is added using the original sample collection SampleSet uploaded by user through this method by t (t >=0)
The sample set SampleSet of secondary update；
Step 3: being trained using updated sample set, obtain new classifier；
Step 4: utilizing the frame of recognition failures before the backtracking identification of updated classifier；
Step 5: will backtracking identification process in identify successfully, the boundary frame of recognition failures be respectively labeled as key frame
KeyFrame, critical frame CriticalFrame, and successful face parts of images SuccessFaceCut will be identified in key frame
Sample set is added；
Step 6: step 3~step 5 is repeated, until critical frame CriticalFrame a certain in trace-back process is continuous
Until recognition failures；
Step 7: the position before returning to backtracking, circulation execute step 1~step 7, identify remaining frame image.
In step 1, set in the reasonable scope higher threshold value T be in order to guarantee the accuracy in face recognition process,
Subsequent step is carried out under the premise of identification is accurate.Higher threshold value T is set in the process of face recognition, as recognition result R
When less than threshold value T, it is believed that recognition failures；When recognition result R is greater than or equal to threshold value T, it is believed that identify successfully.It can protect in this way
The accuracy of identification is demonstrate,proved, so that subsequent step carries out under the premise of identification is accurate.But threshold value T is excessively high to will cause certain journey
Missing inspection on degree, although there are the facial images of people to be identified in i.e. primitive frame image SourceFrame, due to sample number
Amount less, reasons, the recognition result R such as sample quality difference threshold value T may be not achieved, so that identification be caused to omit.Therefore subsequent step
To carry out backtracking identification process on the basis of dynamic construction sample set, dynamic construction is closest to personage's current state to be identified
Sample set is recognized, and is reduced and is omitted, reduces omission factor.
In step 2, during step 1 executes, the first frame FirstFrame for identifying successfully segment is found, and will know
After not successful face parts of images carries out frame choosing, cutting, image preprocessing, obtains face part and cut image FaceCut, and
FaceCut is added in existing sample set SampleSet, the dynamic for completing a sample set updates.Right in sequence
When frame image is identified, being bound to, there are one all to identify successful continuous fragment, finds and currently identifies successful segment
First frame FirstFrame is to reduce unnecessary workload for the frame of recognition failures before the forward trace since the frame.
It is adjacent according to video flowing or close on the similitude of frame image, identify identified in the first frame FirstFrame of successfully segment it is successful
Face parts of images FaceCut has part representative, i.e. FaceCut can reflect and represent people to be identified in FirstFrame institute
Face information in the local video segment at place, is added sample set for FaceCut, has reached timely according to video flowing particular content
Update the purpose of sample set.
In step 3, it is trained using the sample set SampleSet updated in step 2, obtains new classifier
Classifier.After sample set update, new sample set is more representative, using corresponding machine learning method to update
Sample set afterwards is trained, it is therefore an objective to obtain the new classifier of more practical significance.Because new sample set is known for current
Other video flowing is more representative, then carries out subsequent recognition of face with new classifier, and the accurate of identification both can be improved
Rate can also reduce the omission factor of identification.
In step 4, according to consecutive frame, the similitude of image between frame is closed on, utilizes the classifier updated in step 3
Classifier is recognized the frame image of recognition failures before according to the sequence recalled frame by frame.Because in step 1
In used higher threshold value T, so previous frame recognition failures the reason of may be threshold value T setting it is excessively high, sacrifice missing inspection
Rate guarantees accuracy rate, although there are the face informations of personage to be identified that is, in frame image, the knot obtained is identified with classifier
Threshold value T is not achieved in fruit R, causes to omit.For above-mentioned reason, with updated more representational classifier Classifier
According to the sequence recalled frame by frame, the frame image of recognition failures before is recognized, reduces omission factor.It can both protect in this way
Demonstrate,prove higher accuracy rate, it is also ensured that lower omission factor.
In step 5, during step 4 executes, if the frame image Framek of recognition failures is in trace-back process before
It identifies successfully, then continues trace-back process；If the frame image of recognition failures still recognition failures in trace-back process before,
The next frame Framek+1 of frame Framek and the frame are respectively seen as recognition failures, the successful boundary frame of identification, label
Framek is critical frame CriticalFrame, and flag F ramek+1 is key frame KeyFrame.And it will be identified as in Framek+1
The FaceCutk+1 that the face parts of images of function obtain after frame choosing, cutting, image preprocessing is added to existing sample set
In SampelSet, the dynamic for completing a sample set updates.Key frame KeyFrame is identified as the boundary frame of function, according to view
Frequency flows similitude that is adjacent or closing on frame image, and KeyFrame also has local representativeness.It therefore, will be in key frame KeyFrame
Identify that sample set SampleSet is added in successful face parts of images FaceCut, can reach according to video flowing particular content and
The purpose of Shi Gengxin sample set, so that updated sample set is more representative.Only being identified as in key frame KeyFrame
Sample set is added in the face parts of images FaceCut of function, rather than sample is added in all successful face parts of images of identification
Collection, is the scale and quality in order to control sample set.Since video flowing is adjacent or closes on the similitude of frame image, continuously it is identified as
The face parts of images of function is also similar, it is not necessary that sample set is added in all of which, this ensure that sample set updates
Efficiency.
It in step 6, is trained, is obtained new by the method for step 3 using the sample set SampleSet updated in step 5
Classifier Classifier.Since step 5 mark key frame KeyFrame position, by the method for step 4 continue by
According to the sequence recalled frame by frame, the frame image of recognition failures before is again identified that.It handles in trace-back process and is identified as by step 5
The frame image of function, recognition failures, and mark new critical frame CriticalFrame and new key frame KeyFrame.Repetition is held
Row above-mentioned steps, until a certain critical frame when being identified with updated classifier still know by recognition failures, at this time stopping backtracking
Other process.As described in step 4, critical frame and key frame are recognition failures, the successful boundary frame of identification, are identified successfully with being added
The classifier that trains of sample set of key frame face parts of images the boundary frame of recognition failures is identified again, can
It avoids because of missing inspection caused by threshold value T setting is higher.If boundary frame identifies successfully at this time, continues backtracking and identified
Journey；If boundary frame still recognition failures at this time, then it is assumed that the terminal for reaching backtracking identification is thought present threshold value T's
Under setting, really there is no the face informations of personage to be identified in the boundary frame of recognition failures.
In step 7, the position before recalling identification process is returned to, i.e., currently identifies the end position of successfully segment
EndFrame, then gradually identified remaining since EndFrame+1 frame with step after the same method since step 1
Frame image.
The present invention compared with the existing technology, according to specific video streaming content to be measured and actual recognition result, utilizes view
Frequency stream closes on the similitude between frame image, and current sample set is updated and is expanded, and reaches dynamic construction high quality sample
The purpose of this collection, efficiently solves above-mentioned technical problem.
Detailed description of the invention
Fig. 1: for the logic relation picture of the embodiment of the present invention；
Fig. 2: for the flow chart of the embodiment of the present invention.
Specific embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, with reference to the accompanying drawings and embodiments to this hair
It is bright to be described in further detail, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, not
For limiting the present invention.
Referring to Fig.1, knowing method for distinguishing using backtracking proposed by the present invention, the frame image in video flowing is identified, is chosen
It selects representative key frame and is put into sample set SampleSet, dynamic updates sample set, and updates it every time in sample set
Afterwards, the classifier Classifier new with new sample set data training, identifies subsequent frame image using new classifier.
See Fig. 2, a kind of video flowing recognition of face for recalling building dynamic sample collection based on key frame provided by the invention
Method, specific implementation process is:
Firstly, being set according to actual conditions by user and being adjusted threshold value T, guarantee identification can under the premise of allowing missing inspection
Reliability carries out general recognition of face 90% or more.
It is secondary more by t (t >=0) by this method using the original sample collection SampleSet uploaded by user as shown in 1.
New current sample set SampleSettTrained classifier Classifiert, general to a new identification segment progress
In identification process, from segment starting to the equal recognition failures of -1 frame of kth, kth frame identifies successfully that kth+p frame is known to kth+p-1 frame
Do not fail, then kth frame is that above-mentioned current identification segment identifies that successful first frame FirstFrame, kth+p frame are above-mentioned current
Identify the end position EndFrame of segment；
As shown in 2., according to above-mentioned step 2, finds and identify successful first frame FirstFrame namely this example in segment
In kth frame.At this point, -1 frame of kth and kth frame can be considered recognition failures, the successful boundary of identification.
As shown in 3., label -1 frame of kth is critical frame CriticalFrameq, label kth frame is key frame KeyFrameq。
As shown in 4., by key frame KeyFrameqNamely successful face parts of images FaceCut is identified in kth framekAdd
Enter sample set SampleSett, sample set is updated, the sample set SampleSet of the t+1 times update is obtainedt+1.According to upper
Step 3 is stated, updated sample set SampleSet is utilizedt+1Training generates new classifier Classifiert+1, use
Classifiert+1Identification process after continuation.Namely according to above-mentioned step 4, utilize Classifiert+1Before backtracking identification
The frame of recognition failures.
As shown in 5., in this example, from critical frame CriticalFrameqNamely -1 frame of kth starts to carry out backtracking identification
Process.It is adjacent according to video flowing or close on the similitude of frame image, add key frame KeyFrameqIn the successful people of identification
Sample set SampleSet after face information updatet+1It more can really reflect the face face letter of people to be identified in video flowing instantly
Breath, therefore under conditions of threshold value T is constant, recalling the recognition result of identification process, confidence level is higher than before.
As shown in 5., new classifier Classifier is utilizedt+1Identify critical frame CriticalFrameqNamely kth -1
Identified when frame successfully, then according to above-mentioned step 6, continue according to before the step of carry out backtracking identification, until again identifying that failure.
As shown in 6., in backtracking identification process, -1 frame of kth to kth-m+1 frame identifies that success, the identification of kth-m frame are lost
It loses.At this point, kth-m frame and kth-m+1 frame can be considered that recognition failures, the successful boundary frame of identification, label kth-m frame are critical frame
CriticalFrameq+1, label kth-m+1 frame is key frame KeyFrameq+1。
As shown in 7., by key frame KeyFrameq+1Namely successful face parts of images is identified in kth-m+1 frame
FaceCutk-m+1Sample set SampleSet is addedt+1, sample set is updated, the sample set of the t+2 times update is obtained
SampleSett+2, and updated sample set SampleSet is utilized againt+2Training generates new classifier
Classifiert+2, use Classifiert+2Identification process after continuation.According to above-mentioned step 6, Classifier is usedt+2Again
The secondary critical frame CriticalFrame to recognition failuresq+1Namely kth-m frame is identified.
As shown in 8., if kth-m frame still recognition failures, being considered as kth-m frame is that can identify in current identification segment
Successfully near preceding frame position, backtracking identification process is terminated.Then the position according to above-mentioned step 7, before returning to backtracking
EndFrame。
As shown in 9., in this example, that is, kth+p-1 frame is returned to, then since kth+p frame, starts a new identification piece
Section, then successively remaining frame image is identified according to above-mentioned steps.
The present embodiment includes backtracking identification process, dynamic update sample set process, separator iteration renewal process；Backtracking is known
Other process can identify the starting frame position FirstFrame of current identification segment, terminate frame position EndFrame, and with returning
Identification process of tracing back updates starting frame position FirstFrame and terminates frame position EndFrame；Backtracking identification process can find knowledge
Do not fail, identify successful boundary frame position, and is individually identified as critical frame CriticalFrame, key frame KeyFrame；It returns
The face part identified in the key frame KeyFrame found can be cut image and carry out image procossing by identification process of tracing back, and
By treated, face part cutting image FaceCut is automatically added in the sample set SanpleSet of corresponding personage, completes sample
The update of this collection；Recalling identification process can be after the completion of sample set updates, automatically training life on the basis of new samples collection
The classifier of Cheng Xin.
After the completion of a video sub-segments identify, the starting frame position FirstFrame and knot of the sub-piece can recorde
Beam frame position EndFrame, and continue to identify remaining video content since EndFrame, and then find video flowing to be identified
In the successful video sub-segments of all identifications starting frame positions and terminate frame position.
The present invention can effectively solve the problem that face recognition technology generated in practical applications: personal user independently provides big rule
It is slower that the sample data set of mould has biggish difficulty, sample set to update, and cannot reflect the newest face vision of people to be identified in time
Information, the quality of data of sample set are irregular, to carry out accurately identifying the problems such as causing negative effect.The present invention is according to tool
The particular content and actual recognition result of the video flowing to be measured of body are adjacent using video flowing or close on similar between frame image
Property, current sample set is updated and is expanded, achievees the purpose that dynamic construction high quality samples collection, is that one kind can be timely
It updates, the face identification method of automatic building high quality samples collection.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (4)
1. it is a kind of based on key frame recall building dynamic sample collection video flowing face identification method, which is characterized in that including with
Lower step:
Step 1: general recognition of face is carried out to video flowing；
Step 2: utilizing current sample set SampleSettTrained classifier Classifiert, to a new identification segment
It carries out in general identification process, from segment starting to the equal recognition failures of -1 frame of kth, kth frame is identified as to kth+p-1 frame
Function, kth+p frame recognition failures, then kth frame is that current identification segment identifies that successful first frame FirstFrame, kth+p frame are
The end position EndFrame of current identification segment；Wherein, current sample set SampleSettTo utilize the original uploaded by user
Beginning sample set SampleSet passes through the sample set of t update, t >=0 by this method；
Step 3: finding in current identification segment and identify successful first frame FirstFrame, at this point, -1 frame of kth and kth frame view
For recognition failures, the successful boundary of identification；
Step 4: label -1 frame of kth is critical frame CriticalFrameq, label kth frame is key frame KeyFrameq；
Step 5: by key frame KeyFrameqThe middle successful face parts of images FaceCut of identificationkSample set is added
SampleSett, sample set is updated, the sample set SampleSet of the t+1 times update is obtainedt+1；
Step 6: utilizing updated sample set SampleSett+1Training generates new classifier Classifiert+1, utilize
Classifiert+1The frame of recognition failures before backtracking identification；
From critical frame CriticalFrameqStart to carry out backtracking identification, utilizes new classifier Classifiert+1It identifies critical
Frame CriticalFrameqIf identifying successfully, continue backtracking identification, until again identifying that failure；
In backtracking identification process, if -1 frame of kth identifies success, kth-m frame recognition failures to kth-m+1 frame；Then kth-m frame
It is considered as recognition failures, the successful boundary frame of identification with kth-m+1 frame, label kth-m frame is critical frame CriticalFrameq+1,
Label kth-m+1 frame is key frame KeyFrameq+1；
Step 7: by key frame KeyFrameq+1The middle successful face parts of images FaceCut of identificationk-m+1Sample set is added
SampleSett+1, sample set is updated, the sample set SampleSet of the t+2 times update is obtainedt+2, and again using more
Sample set SampleSet after newt+2Training generates new classifier Classifiert+2；Use Classifiert+2Again to knowledge
Not Shi Bai critical frame CriticalFrameq+1It is identified；
If critical frame CriticalFrameq+1Still recognition failures are considered as critical frame CriticalFrameq+1Currently to know
It can be identified in other segment successfully near preceding frame position, terminate backtracking identification process；
Step 8: the position EndFrame before returning to backtracking；And since EndFrame, circulation executes step 1~step 8, continues
Remaining video image is identified.
2. the video flowing face identification method according to claim 1 that building dynamic sample collection is recalled based on key frame,
It is characterized in that: in step 1, given threshold T, when recognition result R is less than threshold value T, it is believed that recognition failures；When recognition result R is big
When threshold value T, it is believed that identify successfully；Wherein, the value of threshold value T is set and is adjusted according to the actual situation by user.
3. the video flowing face identification method according to claim 1 that building dynamic sample collection is recalled based on key frame,
It is characterized in that: in step 5, after identifying that successful face parts of images carries out frame choosing, cutting, image preprocessing, obtaining face
Part cuts image FaceCutk, and by FaceCutkIt is added to existing sample set SampleSettIn, complete a sample set
Dynamic update.
4. the video flowing face for recalling building dynamic sample collection based on key frame according to claim 1 to 3 is known
Other method, it is characterised in that: updated sample set is trained using machine learning method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910231632.1A CN109961046B (en) | 2019-03-26 | 2019-03-26 | Video stream face identification method for building dynamic sample set based on keyframe backtracking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910231632.1A CN109961046B (en) | 2019-03-26 | 2019-03-26 | Video stream face identification method for building dynamic sample set based on keyframe backtracking |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109961046A true CN109961046A (en) | 2019-07-02 |
| CN109961046B CN109961046B (en) | 2022-03-15 |
Family
ID=67024862
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910231632.1A Active CN109961046B (en) | 2019-03-26 | 2019-03-26 | Video stream face identification method for building dynamic sample set based on keyframe backtracking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109961046B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449560A (en) * | 2020-03-26 | 2021-09-28 | 广州金越软件技术有限公司 | Technology for comparing human faces based on dynamic portrait library |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537389A (en) * | 2014-12-29 | 2015-04-22 | 生迪光电科技股份有限公司 | Human face recognition method and terminal equipment |
| US20180204111A1 (en) * | 2013-02-28 | 2018-07-19 | Z Advanced Computing, Inc. | System and Method for Extremely Efficient Image and Pattern Recognition and Artificial Intelligence Platform |
-
2019
- 2019-03-26 CN CN201910231632.1A patent/CN109961046B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180204111A1 (en) * | 2013-02-28 | 2018-07-19 | Z Advanced Computing, Inc. | System and Method for Extremely Efficient Image and Pattern Recognition and Artificial Intelligence Platform |
| CN104537389A (en) * | 2014-12-29 | 2015-04-22 | 生迪光电科技股份有限公司 | Human face recognition method and terminal equipment |
Non-Patent Citations (2)
| Title |
|---|
| JINLONG LI: "Video shot segmentation and key frame extraction based on SIFT feature", 《2012 INTERNATIONAL CONFERENCE ON IMAGE ANALYSIS AND SIGNAL PROCESSING》 * |
| 沈晴: "基于视频的人机交互中动作在线发现与时域分割", 《计算机学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113449560A (en) * | 2020-03-26 | 2021-09-28 | 广州金越软件技术有限公司 | Technology for comparing human faces based on dynamic portrait library |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109961046B (en) | 2022-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222701A (en) | A kind of bridge defect automatic identifying method | |
| CN109086785A (en) | A kind of training method and device of image calibration model | |
| US20220309772A1 (en) | Human activity recognition fusion method and system for ecological conservation redline | |
| CN103377647A (en) | Automatic music notation recording method and system based on audio and video information | |
| CN110176025B (en) | Invigilator tracking method based on posture | |
| CN103984943A (en) | Scene text identification method based on Bayesian probability frame | |
| CN105138953A (en) | Method for identifying actions in video based on continuous multi-instance learning | |
| CN105336342A (en) | Method and system for evaluating speech recognition results | |
| CN109708638A (en) | A kind of ship track point extracting method | |
| CN109961046A (en) | The video flowing face identification method of building dynamic sample collection is recalled based on key frame | |
| CN103824461A (en) | Vehicle driving situation data recognition and modification method | |
| CN110084129A (en) | A kind of river drifting substances real-time detection method based on machine vision | |
| CN109614896A (en) | A method of the video content semantic understanding based on recursive convolution neural network | |
| CN109684511A (en) | A kind of video clipping method, video aggregation method, apparatus and system | |
| CN107886125B (en) | MODIS satellite remote sensing image labeling method based on local spectrum decomposition scoring | |
| CN104347071B (en) | Method and system for generating reference answers of spoken language test | |
| CN114627411A (en) | Crop growth period identification method based on parallel detection under computer vision | |
| CN110659572A (en) | Video motion detection method based on bidirectional feature pyramid | |
| CN114494941A (en) | Comparison learning-based weak supervision time sequence action positioning method | |
| CN110189327A (en) | Eye ground blood vessel segmentation method based on structuring random forest encoder | |
| CN109948614B (en) | Wrist bone interest area cutting method based on machine learning | |
| CN103106633B (en) | A kind of video foreground object screenshot method based on gauss hybrid models and system | |
| CN106877955B (en) | Fm broadcast signal based on hidden Markov model gives the correct time characteristic recognition method | |
| CN110543675A (en) | Power transmission line fault identification method | |
| CN110046666A (en) | Mass picture mask method |
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 |