CN108038413A - Cheat probability analysis method, apparatus and storage medium - Google Patents
Cheat probability analysis method, apparatus and storage medium Download PDFInfo
- Publication number
- CN108038413A CN108038413A CN201711061172.XA CN201711061172A CN108038413A CN 108038413 A CN108038413 A CN 108038413A CN 201711061172 A CN201711061172 A CN 201711061172A CN 108038413 A CN108038413 A CN 108038413A
- Authority
- CN
- China
- Prior art keywords
- video
- sample
- analyzed
- fraud
- probability
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 71
- 238000003860 storage Methods 0.000 title description 17
- 238000012549 training Methods 0.000 claims abstract description 51
- 230000007935 neutral effect Effects 0.000 claims abstract description 27
- 230000001815 facial effect Effects 0.000 claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 17
- 238000000605 extraction Methods 0.000 claims description 43
- 230000006870 function Effects 0.000 claims description 25
- 230000015654 memory Effects 0.000 claims description 17
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000002360 preparation method Methods 0.000 claims description 6
- 210000004218 nerve net Anatomy 0.000 claims 2
- 239000000284 extract Substances 0.000 description 7
- 238000004891 communication Methods 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241000209202 Bromus secalinus Species 0.000 description 1
- 238000012550 audit Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06V40/174—Facial expression recognition
-
- 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
- G06V40/70—Multimodal biometrics, e.g. combining information from different biometric modalities
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Neurology (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of fraud probability analysis method, apparatus and computer-readable recording medium.This method comprises the following steps:Collect Sample video and distribute fraud mark;The characteristics of image and audio frequency characteristics of each Sample video are extracted, combination obtains the video features of each Sample video;Neutral net is built according to the dimension of the sequence length of Sample video and video features;With the video features and fraud mark training neutral net of each Sample video, optimize training parameter, obtain fraud probability analysis model;Gather the facial video of object scheduled duration to be analyzed;The characteristics of image and audio frequency characteristics of the video are extracted, combination obtains the video features of the video;By the video features input fraud probability analysis model, the probability of cheating of the object to be analyzed is exported and without probability of cheating, the output result for taking probable value big whether there is the analysis result of fraud as the object to be analyzed.Using the present invention, it objective can judge that personage whether there is fraud suspicion.
Description
Technical field
The present invention relates to technical field of information processing, more particularly to a kind of fraud probability analysis method, apparatus and storage
Medium.
Background technology
At present, personage's fraud analysis is generally realized by way of face is examined, the experience of extreme dependency analysis personnel and judgement,
And take a substantial amount of time and manpower, analysis result are often inaccurate objective.Also there is the instrument and equipment using specialty, pass through detection
The indexs such as breathing, pulse, skin electricity judge fraud suspect whether there is fraud, but such instrument and equipment is generally expensive and holds
Easy right in personam, which is formed, to be invaded.
The content of the invention
For these reasons, it is necessary to a kind of fraud probability analysis method, apparatus and storage medium are provided, pass through analysis
The facial video of personage, it is objective, judge that personage whether there is fraud suspicion exactly.
To achieve the above object, the present invention provides a kind of fraud probability analysis method, and this method includes:
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, one is distributed for each sample and takes advantage of
Swindleness mark;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains each sample
Video features;
Network struction step:According to the dimension of the sequence length of each sample and video features set the neutral net number of plies and
Neuron number per layer network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample number
According to, the neutral net is trained, exports the probability of cheating of each sample and without probability of cheating, it is each to train the renewal nerve
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud can
Can property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis
The face video of the model analysis object to be analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
Preferably, the sample characteristics extraction step includes:
Each sample is decoded and pre-processed, obtains the video frame and audio-frequency unit of each sample;
Feature extraction is carried out to the video frame of each sample, obtains the characteristics of image of each sample;
Feature extraction is carried out to the audio-frequency unit of each sample, obtains the audio frequency characteristics of each sample.
Preferably, the dimension of the video features for dimension and the corresponding audio frequency characteristics of described image feature dimension it
With.
Preferably, the Softmax loss functions formula is as follows:
Wherein, θ be the neutral net training parameter, XjRepresent j-th of sample, yjRepresent that the fraud of j-th of sample is general
Rate.
Preferably, the training parameter includes iterations.
Preferably, the model applying step further includes:
The object video to be analyzed is decoded and pre-processed, obtain the object video to be analyzed audio-frequency unit and
Video frame;
Feature extraction is carried out to the video frame of the object video to be analyzed, the image for obtaining the object video to be analyzed is special
Sign;
Feature extraction is carried out to the audio-frequency unit of the object video to be analyzed, the audio for obtaining the object video to be analyzed is special
Sign;
The characteristics of image and audio frequency characteristics of the object video to be analyzed are combined, obtain the object video to be analyzed
Video features;
The fraud probability analysis model that video features input training is obtained, the fraud for exporting the object to be analyzed are general
Rate and without probability of cheating.
The present invention also provides a kind of computing device, including memory and processor, the memory includes fraud may
Property analysis program.The computing device is directly or indirectly connected with camera device, when camera device is by the human dialog of shooting
Facial video be sent to computing device.The processor of the computing device performs the fraud probability analysis program in memory
When, realize following steps:
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, one is distributed for each sample and takes advantage of
Swindleness mark;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains each sample
Video features;
Network struction step:According to the dimension of the sequence length of each sample and video features set the neutral net number of plies and
Neuron number per layer network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample number
According to, the neutral net is trained, exports the probability of cheating of each sample and without probability of cheating, it is each to train the renewal nerve
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud can
Can property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis
The face video of the model analysis object to be analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
Preferably, the sample characteristics extraction step includes:
Each sample is decoded and pre-processed, obtains the video frame and audio-frequency unit of each sample;
Feature extraction is carried out to the video frame of each sample, obtains the characteristics of image of each sample;
Feature extraction is carried out to the audio-frequency unit of each sample, obtains the audio frequency characteristics of each sample.
Preferably, the dimension for stating video features is the dimension of described image feature and the dimension of corresponding audio frequency characteristics
The sum of.
Preferably, the Softmax loss functions formula is as follows:
Wherein, θ be the neutral net training parameter, XjRepresent j-th of sample, yjRepresent that j-th of sample is corresponding to take advantage of
Cheat the probability of mark.
Preferably, the training parameter includes iterations.
Preferably, the model applying step further includes:
The object video to be analyzed is decoded and pre-processed, obtain the object video to be analyzed audio-frequency unit and
Video frame;
Feature extraction is carried out to the video frame of the object video to be analyzed, the image for obtaining the object video to be analyzed is special
Sign;
Feature extraction is carried out to the audio-frequency unit of the object video to be analyzed, the audio for obtaining the object video to be analyzed is special
Sign;
The characteristics of image and audio frequency characteristics of the object video to be analyzed are combined, obtain the object video to be analyzed
Video features;
The fraud probability analysis model that video features input training is obtained, the fraud for exporting the object to be analyzed are general
Rate and without probability of cheating.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer-readable recording medium
Storage medium includes cheating probability analysis program, when the fraud probability analysis program is executed by processor, realizes such as
Arbitrary steps in the upper fraud probability analysis method.
Fraud probability analysis method, apparatus provided by the invention and storage medium, pass through the facial video of a large amount of personages
Training neutral net, the training parameter of neutral net is updated according to Softmax loss functions, with the training after updating for the last time
Parameter obtains fraud probability analysis model as final argument.Afterwards, the face of scheduled duration when gathering object dialogue to be analyzed
Portion's video, extracts the audio frequency characteristics and characteristics of image of the video, combination obtains the video features of the video, and the video features are defeated
Enter the fraud probability analysis model that training obtains, you can obtain the analysis result of the object fraud possibility to be analyzed.Utilize
The present invention, can objective, effectively judge that personage whether there is fraud suspicion, also reduce cost, save the time.
Brief description of the drawings
Fig. 1 is the applied environment figure of present invention fraud probability analysis the first preferred embodiment of method.
Fig. 2 is the applied environment figure of present invention fraud probability analysis the second preferred embodiment of method.
Fig. 3 is the Program modual graph that probability analysis program is cheated in Fig. 1, Fig. 2.
Fig. 4 is the flow chart of present invention fraud probability analysis method preferred embodiment.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Embodiment
The principle of the present invention and spirit are described below with reference to some specific embodiments.It is it should be appreciated that described herein
Specific embodiment only to explain the present invention, be not intended to limit the present invention.
With reference to shown in Fig. 1, for the applied environment figure of present invention fraud probability analysis the first preferred embodiment of method.At this
In embodiment, camera device 3 connects computing device 1 by network 2, and camera device 3 shoots facial video during human dialog, leads to
Cross network 2 and be sent to computing device 1, computing device 1 is analyzed described using fraud probability analysis program 10 provided by the invention
Video, exports the probability of cheating of personage and without probability of cheating, is referred to for people.
Computing device 1 can be that server, smart mobile phone, tablet computer, pocket computer, desktop PC etc. have
Storage and the terminal device of calculation function.
The computing device 1 includes memory 11, processor 12, network interface 13 and communication bus 14.
Camera device 3 is installed on particular place, such as office space, monitoring area, face during for shooting human dialog
Portion's video, then will shoot obtained transmission of video to memory 11 by network 2.Network interface 13 can include having for standard
Line interface, wave point (such as WI-FI interfaces).Communication bus 14 is used for realization the connection communication between these components.
Memory 11 includes the readable storage medium storing program for executing of at least one type.The readable storage medium storing program for executing of at least one type
Can be such as flash memory, hard disk, multimedia card, the non-volatile memory medium of card-type memory.In certain embodiments, it is described can
Read the internal storage unit that storage medium can be the computing device 1, such as the hard disk of the computing device 1.In other realities
Apply in example, the readable storage medium storing program for executing can also be the external memory storage 11 of the computing device 1, such as the computing device 1
The plug-in type hard disk of upper outfit, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital,
SD) block, flash card (Flash Card) etc..
In the present embodiment, the memory 11 stores the program code of the fraud probability analysis program 10, shooting
The dialogue video that device 3 is shot, and processor 12 perform the number that the program code of fraud probability analysis program 10 is applied to
Data exported according to this and finally etc..
Processor 12 can be in certain embodiments a central processing unit (Central Processing Unit,
CPU), microprocessor or other data processing chips.
Fig. 1 illustrate only the computing device 1 with component 11-14, it should be understood that being not required for implementing all show
The component gone out, what can be substituted implements more or less components.
Alternatively, which can also include user interface, and user interface can include input unit such as keyboard
(Keyboard), the equipment with speech identifying function such as speech input device such as microphone (microphone), voice are defeated
Go out device such as sound equipment, earphone etc., alternatively user interface can also include standard wireline interface and wireless interface.
Alternatively, which can also include display.Display can be that LED is shown in certain embodiments
Device, liquid crystal display, touch-control liquid crystal display and OLED (OrganicLight-Emitting Diode, organic light emission two
Pole pipe) touch device etc..Display is used for the information and visual user interface for showing that computing device 1 is handled.
Alternatively, which further includes touch sensor.What the touch sensor was provided is touched for user
The region for touching operation is known as touch area.In addition, touch sensor described here can be resistive touch sensor, capacitance
Formula touch sensor etc..Moreover, the touch sensor not only includes the touch sensor of contact, proximity may also comprise
Touch sensor etc..In addition, the touch sensor can be single sensor, or such as multiple biographies of array arrangement
Sensor.User, such as psychologist, can start fraud probability analysis program 10 by touching.
The computing device 1 can also include radio frequency (Radio Frequency, RF) circuit, sensor and voicefrequency circuit etc.
Deng details are not described herein.
With reference to shown in Fig. 2, for the applied environment figure of present invention fraud probability analysis the second preferred embodiment of method.Treat point
Analysis object realizes that personage cheats the analytic process of possibility by terminal 3, and the camera device 30 of terminal 3 shoots object to be analyzed
Facial video, and the computing device 1 is sent to by network 2, the processor 12 of computing device 1 performs what memory 11 stored
The program code of probability analysis program 10 is cheated, the audio-frequency unit and video frame of video are analyzed, it is to be analyzed to export this
The probability of cheating of object and without probability of cheating, refers to for object to be analyzed or audit crew etc..
The component of computing device 1 in Fig. 2, such as memory 11, processor 12, network interface 13 and the communication shown in figure
Bus 14, and the component not shown in figure, refer to the introduction on Fig. 1.
The terminal 3 can be smart mobile phone, tablet computer, pocket computer, desktop PC etc. have storage and
The terminal device of calculation function.
Fraud probability analysis program 10 in Fig. 1, Fig. 2, when being performed by processor 12, realizes following steps:
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, one is distributed for each sample and takes advantage of
Swindleness mark;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains each sample
Video features;
Network struction step:According to the dimension of the sequence length of each sample and video features set the neutral net number of plies and
Neuron number per layer network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample number
According to, the neutral net is trained, exports the probability of cheating of each sample and without probability of cheating, it is each to train the renewal nerve
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud can
Can property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis
The face video of the model analysis object to be analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
On being discussed in detail for above-mentioned steps, program moulds of following Fig. 3 on cheating probability analysis program 10 refer to
The explanation of the flow chart of block diagram and Fig. 4 on cheating probability analysis method preferred embodiment.
It is the Program modual graph that probability analysis program 10 is cheated in Fig. 1, Fig. 2 with reference to shown in Fig. 3.In the present embodiment,
Fraud probability analysis program 10 is divided into multiple modules, and the plurality of module is stored in memory 11, and by processor
12 perform, to complete the present invention.Module alleged by the present invention is that the series of computation machine program for referring to completion specific function refers to
Make section.
The fraud probability analysis program 10 can be divided into:Acquisition module 110, extraction module 120, training module
130 and prediction module 140.
Acquisition module 110, the facial video of scheduled duration during for obtaining human dialog.The video can be passed through
It is that the camera device 30 of the camera device 3 of Fig. 1 or Fig. 2 obtain or chosen from the network information or video database
It is clearly present the facial video of fraud and without the facial video of fraud.To be taken advantage of for the distribution of the Sample video of neural metwork training
Mark is cheated, fraud mark represents personage whether there is fraud suspicion in the Sample video, such as 1 indicates fraud suspicion, and 0, which represents nothing, takes advantage of
Swindleness suspicion.
Extraction module 120, for extracting the audio frequency characteristics and characteristics of image of the video, and audio frequency characteristics and image are special
Sign combination, obtains the video features of each video.The video obtained to acquisition module 110 is decoded and pre-processed, and is obtained every
The audio-frequency unit and video frame of a video, carry out feature extraction to the audio-frequency unit and video frame, obtain each video respectively
Audio frequency characteristics and characteristics of image, the audio frequency characteristics and characteristics of image are combined, obtain the video features of each video.
, can be by regarding by processing such as normalization, removal noises when extraction module 120 extracts the characteristics of image of the video
The HOG features of frequency frame, LBP features etc. are used as characteristics of image, can also directly utilize the spy of convolutional neural networks extraction video frame
Sign vector.
When extraction module 120 extracts the audio frequency characteristics of the video, the amplitude of the audio-frequency unit of the video can be made
For audio frequency characteristics.For example, it is assumed that the scheduled duration of the video is 3 minutes, audio sample rate 8000HZ, then 3 minutes videos
Audio-frequency unit extract 8000*60*3 amplitude as audio frequency characteristics.
When extraction module 120 combines described image feature and audio frequency characteristics, the dimension of the video features combined is every
The sum of the characteristics of image dimension of two field picture and corresponding audio frequency characteristics dimension.According to above-mentioned example, it is assumed that the face of human dialog regards
The audio sample rate of frequency is 8000HZ, and video sampling rate is 20HZ, then often reading a two field picture needs 50ms, 50ms to correspond to 400
A audio amplitude value, if characteristics of image dimension of the video per two field picture is k1, the dimension k2=of corresponding audio frequency characteristics
400, the dimension k=k1+k2 of the video features combined.
Training module 130, for the neutral net by repetitive exercise optimization structure, obtains trained fraud possibility
Analysis model.The video frame and audio frame of the facial video of human dialog are sequentially arranged, therefore the present invention is using circulation
Shot and long term memory network (Long Short-Term Memory, LSTM) in neutral net.
Build LSTM when, first according to acquisition module 110 obtain human dialog when scheduled duration face video sequence
The dimension of video features that length and the extraction combination of extraction module 120 obtain defines network shape, sets the number of plies of LSTM and every
The neuron number of layer LSTM.With above-mentioned example, it is assumed that the scheduled duration of the video is 3 minutes, and video sampling rate is 20HZ,
The dimension for combining obtained video features is k, then the sequence length of each video is denoted as 3*60*20, and the shape of the LSTM can use
The code in tflearn deep learnings storehouse is expressed as form:
Net=tflearn.input_data (shape=[None, 3*60*20, k])
Then build two hidden layers, every layer of 128 neural unit, with the code in tflearn deep learnings storehouse represent as
Under:
Net=tflearn.lstm (net, 128)
Net=tflearn.lstm (net, 128)
Next it is as follows that Softmax loss function formula are defined:
After the completion of LSTM and Softmax loss functions structure, training parameter is set.Assuming that iterations is 100, gradient is excellent
Change algorithm is adam, verification collection is 0.1, then LSTM model trainings represent as follows with the code in tflearn deep learnings storehouse:
Net=tflearn.regression (net, optimizer=' adam ', loss=' categorical_
Crossentropy ',
Name=' output1 ')
Model=tflearn.DNN (net, tersorboard_verbose=2)
Model.fit (X, Y, n_epoch=100, validation_set=0.1, snapshot_step=100)
The video features that training module 130 is marked using the fraud of each sample and combination obtains are trained LSTM,
Training updates the training parameter of the LSTM every time, minimizes the Softmax loss functions, with the instruction after updating for the last time
Practice parameter as final argument, obtain fraud probability analysis model.
Analysis module 140, for analyzing the fraud possibility of personage.Acquisition module 110 obtains the pre- timing of object to be analyzed
Long facial video, extraction module 120 extract the characteristics of image and audio frequency characteristics of the video, and the characteristics of image and audio is special
Sign is combined into the video features of the video, and analysis module 140 can by video features input 130 trained fraud of training module
Energy property analysis model, exports the probability of cheating of object to be analyzed and without probability of cheating.
With reference to shown in Fig. 4, for the flow chart of present invention fraud probability analysis method preferred embodiment.Utilize Fig. 1 or Fig. 2
Shown framework, starts computing device 1, processor 12 performs the fraud probability analysis program 10 stored in memory 11, real
Existing following steps:
Step S10, using acquisition module 110 collect human dialog when scheduled duration face video and taken advantage of for video distribution
Swindleness mark.The video can be obtained by the camera device 30 of the camera device 3 of Fig. 1 or Fig. 2 or from network
The facial video of fraud and normally regarding without fraud are clearly present during the human dialog chosen in information or video database
Frequently.
Step S20, the audio frequency characteristics and characteristics of image of each video are extracted using extraction module 120, and combine the audio
Feature and characteristics of image, obtain the video features of each video.Described image feature can be the HOG features of video frame, LBP spies
The feature vector of the low-level image features such as sign or the video frame directly extracted using convolutional neural networks.The audio frequency characteristics
It can be the set for the amplitude that every two field picture corresponds to audio.The dimension of the video features is the characteristics of image dimension of video frame
The sum of with corresponding audio frequency characteristics dimension.
Step S30, neutral net is built according to the dimension of the sequence length of the video of the scheduled duration, video features.
The sequence length and extraction module 120 of the facial video of the scheduled duration obtained according to acquisition module 110 extract, combine and obtain
Video features dimension set neutral net the number of plies and per layer network neuron number because the output knot of neutral net
Fruit is for the probability of cheating of personage and without probability of cheating, so the neuron number as the grader of network output layer is 2.
Step S40, according to the video features of each video and the fraud mark training neutral net, obtains trained
Cheat probability analysis model.The fraud mark and extraction module 120 of the Sample video obtained with acquisition module 110 extract, group
It is sample data to close obtained video features, and training is iterated to neutral net, and training every time updates the instruction of the neutral net
Practice parameter, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining trained fraud can
Can property analysis model.
Step S50, the facial video of object scheduled duration to be analyzed is gathered using acquisition module 110.The face video leads to
The camera device 30 of the camera device 3 or Fig. 2 of crossing Fig. 1 obtains.
Step S60, the characteristics of image and audio frequency characteristics of the object video to be analyzed are extracted using extraction module 120, will
The characteristics of image and audio frequency characteristics combination, obtain the video features of the object video to be analyzed.Feature extraction is specific with combination
Process, refer to being discussed in detail for extraction module 120 and step S20.
Step S70, by the video features input fraud probability analysis model, the fraud for obtaining object to be analyzed can
Can property analysis result.The fraud possibility that the video features input training for the object to be analyzed that extraction module 120 is obtained obtains
Analysis model, exports the probability of cheating value of the object to be analyzed and without probability of cheating value, takes the output result conduct that probable value is big
The object to be analyzed whether there is the analysis result of fraud.
In addition, the embodiment of the present invention also proposes a kind of computer-readable recording medium, the computer-readable recording medium
Can be hard disk, multimedia card, SD card, flash card, SMC, read-only storage (ROM), Erasable Programmable Read Only Memory EPROM
(EPROM), any one in portable compact disc read-only storage (CD-ROM), USB storage etc. or several timess
Meaning combination.The computer-readable recording medium includes Sample video and fraud probability analysis program 10, and the fraud can
Energy property analysis program 10 realizes following operation when being executed by processor:.
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, one is distributed for each sample and takes advantage of
Swindleness mark;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains each sample
Video features;
Network struction step:According to the dimension of the sequence length of each sample and video features set the neutral net number of plies and
Neuron number per layer network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample number
According to, the neutral net is trained, exports the probability of cheating of each sample and without probability of cheating, it is each to train the renewal nerve
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud can
Can property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis
The face video of the model analysis object to be analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
The embodiment of the computer-readable recording medium of the present invention and above-mentioned fraud probability analysis method and
The embodiment of computing device 1 is roughly the same, and details are not described herein.
It should be noted that herein, term " comprising ", "comprising" or its any other variant are intended to non-row
His property includes, so that process, device, article or method including a series of elements not only include those key elements, and
And other elements that are not explicitly listed are further included, or further include as this process, device, article or method institute inherently
Key element.In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that including this
Also there are other identical element in the process of key element, device, article or method.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.Embodiment party more than
The description of formula, it is required general that those skilled in the art can be understood that above-described embodiment method can add by software
The mode of hardware platform is realized, naturally it is also possible to which by hardware, but the former is more preferably embodiment in many cases.It is based on
Such understanding, the part that technical scheme substantially in other words contributes the prior art can be with software products
Form embody, which is stored in a storage medium (such as ROM/RAM, magnetic disc, light as described above
Disk) in, including some instructions use is so that a station terminal equipment (can be mobile phone, computer, server, or the network equipment
Deng) perform method described in each embodiment of the present invention.
It these are only the preferred embodiment of the present invention, be not intended to limit the scope of the invention, it is every to utilize this hair
The equivalent structure or equivalent flow shift that bright specification and accompanying drawing content are made, is directly or indirectly used in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. one kind fraud probability analysis method, it is characterised in that this method includes:
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, a fraud mark is distributed for each sample
Note;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains the video of each sample
Feature;
Network struction step:The neutral net number of plies and every layer are set according to the dimension of the sequence length of each sample and video features
The neuron number of network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample data,
The neutral net is trained, the probability of cheating of each sample is exported and updates the nerve net without probability of cheating, every time training
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud may
Property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis model
The face video of the object to be analyzed is analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
2. fraud probability analysis method as claimed in claim 1, it is characterised in that the sample characteristics extraction step bag
Include:
Each sample is decoded and pre-processed, obtains the video frame and audio-frequency unit of each sample;
Feature extraction is carried out to the video frame of each sample, obtains the characteristics of image of each sample;
Feature extraction is carried out to the audio-frequency unit of each sample, obtains the audio frequency characteristics of each sample.
3. fraud probability analysis method as claimed in claim 1, it is characterised in that the dimension of the video features is described
The sum of the dimension of characteristics of image and dimension of corresponding audio frequency characteristics.
4. fraud probability analysis method as claimed in claim 1, it is characterised in that the Softmax loss functions formula
It is as follows:
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>&theta;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<mi>n</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>h</mi>
<mi>&theta;</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>+</mo>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
</mrow>
<mo>)</mo>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>h</mi>
<mi>&theta;</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
Wherein, θ be the neutral net training parameter, XjRepresent j-th of sample, yjRepresent the probability of cheating of j-th of sample.
5. fraud probability analysis method as claimed in claim 1, it is characterised in that the training in the network training step
Parameter includes iterations.
6. fraud probability analysis method as claimed in claim 1, it is characterised in that the model applying step further includes:
The object video to be analyzed is decoded and pre-processed, obtains the audio-frequency unit and video of the object video to be analyzed
Frame;
Feature extraction is carried out to the video frame of the object video to be analyzed, obtains the characteristics of image of the object video to be analyzed;
Feature extraction is carried out to the audio-frequency unit of the object video to be analyzed, obtains the audio frequency characteristics of the object video to be analyzed;
The characteristics of image and audio frequency characteristics of the object video to be analyzed are combined, obtain the video of the object video to be analyzed
Feature;
By the obtained fraud probability analysis model of video features input training, export the object to be analyzed probability of cheating and
Without probability of cheating.
7. a kind of computing device, including memory and processor, it is characterised in that the memory includes cheating possibility point
Program is analysed, the fraud probability analysis program realizes following steps when being performed by the processor:
Sample preparation process:The facial video of personage's scheduled duration is collected as sample, a fraud mark is distributed for each sample
Note;
Sample characteristics extraction step:The characteristics of image and audio frequency characteristics of each sample are extracted, combination obtains the video of each sample
Feature;
Network struction step:The neutral net number of plies and every layer are set according to the dimension of the sequence length of each sample and video features
The neuron number of network;
Network training step:Softmax loss functions are defined, using the fraud mark and video features of each sample as sample data,
The neutral net is trained, the probability of cheating of each sample is exported and updates the nerve net without probability of cheating, every time training
The training parameter of network, so that the training parameter that the Softmax loss functions minimize, as final argument, obtaining fraud may
Property analysis model;And
Model applying step:The facial video of object scheduled duration to be analyzed is gathered, utilizes the fraud probability analysis model
The face video of the object to be analyzed is analyzed, obtains the analysis result of the object fraud possibility to be analyzed.
8. computing device as claimed in claim 7, it is characterised in that the sample characteristics extraction step includes:
Each sample is decoded and pre-processed, obtains the video frame and audio-frequency unit of each sample;
Feature extraction is carried out to the video frame of each sample, obtains the characteristics of image of each sample;
Feature extraction is carried out to the audio-frequency unit of each sample, obtains the audio frequency characteristics of each sample.
9. computing device as claimed in claim 7, it is characterised in that the model applying step further includes:
The object video to be analyzed is decoded and pre-processed, obtains the audio-frequency unit and video of the object video to be analyzed
Frame;
Feature extraction is carried out to the video frame of the object video to be analyzed, obtains the characteristics of image of the object video to be analyzed;
Feature extraction is carried out to the audio-frequency unit of the object video to be analyzed, obtains the audio frequency characteristics of the object video to be analyzed;
The characteristics of image and audio frequency characteristics of the object video to be analyzed are combined, obtain the video of the object video to be analyzed
Feature;
By the obtained fraud probability analysis model of video features input training, export the object to be analyzed probability of cheating and
Without probability of cheating.
10. a kind of computer-readable recording medium, it is characterised in that the computer-readable recording medium includes fraud may
Property analysis program, the fraud probability analysis program realized such as any one of claim 1 to 6 institute when being executed by processor
The step of fraud probability analysis method stated.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711061172.XA CN108038413A (en) | 2017-11-02 | 2017-11-02 | Cheat probability analysis method, apparatus and storage medium |
PCT/CN2018/076122 WO2019085331A1 (en) | 2017-11-02 | 2018-02-10 | Fraud possibility analysis method, device, and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711061172.XA CN108038413A (en) | 2017-11-02 | 2017-11-02 | Cheat probability analysis method, apparatus and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108038413A true CN108038413A (en) | 2018-05-15 |
Family
ID=62092695
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711061172.XA Pending CN108038413A (en) | 2017-11-02 | 2017-11-02 | Cheat probability analysis method, apparatus and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN108038413A (en) |
WO (1) | WO2019085331A1 (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108776932A (en) * | 2018-05-22 | 2018-11-09 | 深圳壹账通智能科技有限公司 | Determination method, storage medium and the server of customer investment type |
CN109284371A (en) * | 2018-09-03 | 2019-01-29 | 平安证券股份有限公司 | Anti- fraud method, electronic device and computer readable storage medium |
WO2020087974A1 (en) * | 2018-10-30 | 2020-05-07 | 北京字节跳动网络技术有限公司 | Model generation method and device |
CN111382623A (en) * | 2018-12-28 | 2020-07-07 | 广州市百果园信息技术有限公司 | Live broadcast auditing method, device, server and storage medium |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11244050B2 (en) * | 2018-12-03 | 2022-02-08 | Mayachitra, Inc. | Malware classification and detection using audio descriptors |
CN110705585A (en) * | 2019-08-22 | 2020-01-17 | 深圳壹账通智能科技有限公司 | Network fraud identification method and device, computer device and storage medium |
CN111540375B (en) * | 2020-04-29 | 2023-04-28 | 全球能源互联网研究院有限公司 | Training method of audio separation model, and separation method and device of audio signals |
CN113630495B (en) * | 2020-05-07 | 2022-08-02 | 中国电信股份有限公司 | Training method and device for fraud-related order prediction model and order prediction method and device |
CN112926623B (en) * | 2021-01-22 | 2024-01-26 | 北京有竹居网络技术有限公司 | Method, device, medium and electronic equipment for identifying synthesized video |
CN114549026B (en) * | 2022-04-26 | 2022-07-19 | 浙江鹏信信息科技股份有限公司 | Method and system for identifying unknown fraud based on algorithm component library analysis |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104545950A (en) * | 2014-12-23 | 2015-04-29 | 上海博康智能信息技术有限公司 | Non-contact type lie detection method and lie detection system thereof |
CN105160318A (en) * | 2015-08-31 | 2015-12-16 | 北京旷视科技有限公司 | Facial expression based lie detection method and system |
CN106909896A (en) * | 2017-02-17 | 2017-06-30 | 竹间智能科技(上海)有限公司 | Man-machine interactive system and method for work based on character personality and interpersonal relationships identification |
CN106901758A (en) * | 2017-02-23 | 2017-06-30 | 南京工程学院 | A kind of speech confidence level evaluating method based on convolutional neural networks |
CN107103266A (en) * | 2016-02-23 | 2017-08-29 | 中国科学院声学研究所 | The training of two-dimension human face fraud detection grader and face fraud detection method |
CN107133481A (en) * | 2017-05-22 | 2017-09-05 | 西北工业大学 | The estimation of multi-modal depression and sorting technique based on DCNN DNN and PV SVM |
CN107256392A (en) * | 2017-06-05 | 2017-10-17 | 南京邮电大学 | A kind of comprehensive Emotion identification method of joint image, voice |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6826300B2 (en) * | 2001-05-31 | 2004-11-30 | George Mason University | Feature based classification |
CN105956572A (en) * | 2016-05-15 | 2016-09-21 | 北京工业大学 | In vivo face detection method based on convolutional neural network |
CN107007257B (en) * | 2017-03-17 | 2018-06-01 | 深圳大学 | The automatic measure grading method and apparatus of the unnatural degree of face |
-
2017
- 2017-11-02 CN CN201711061172.XA patent/CN108038413A/en active Pending
-
2018
- 2018-02-10 WO PCT/CN2018/076122 patent/WO2019085331A1/en active Application Filing
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104545950A (en) * | 2014-12-23 | 2015-04-29 | 上海博康智能信息技术有限公司 | Non-contact type lie detection method and lie detection system thereof |
CN105160318A (en) * | 2015-08-31 | 2015-12-16 | 北京旷视科技有限公司 | Facial expression based lie detection method and system |
CN107103266A (en) * | 2016-02-23 | 2017-08-29 | 中国科学院声学研究所 | The training of two-dimension human face fraud detection grader and face fraud detection method |
CN106909896A (en) * | 2017-02-17 | 2017-06-30 | 竹间智能科技(上海)有限公司 | Man-machine interactive system and method for work based on character personality and interpersonal relationships identification |
CN106901758A (en) * | 2017-02-23 | 2017-06-30 | 南京工程学院 | A kind of speech confidence level evaluating method based on convolutional neural networks |
CN107133481A (en) * | 2017-05-22 | 2017-09-05 | 西北工业大学 | The estimation of multi-modal depression and sorting technique based on DCNN DNN and PV SVM |
CN107256392A (en) * | 2017-06-05 | 2017-10-17 | 南京邮电大学 | A kind of comprehensive Emotion identification method of joint image, voice |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108776932A (en) * | 2018-05-22 | 2018-11-09 | 深圳壹账通智能科技有限公司 | Determination method, storage medium and the server of customer investment type |
CN109284371A (en) * | 2018-09-03 | 2019-01-29 | 平安证券股份有限公司 | Anti- fraud method, electronic device and computer readable storage medium |
CN109284371B (en) * | 2018-09-03 | 2023-04-18 | 平安证券股份有限公司 | Anti-fraud method, electronic device, and computer-readable storage medium |
WO2020087974A1 (en) * | 2018-10-30 | 2020-05-07 | 北京字节跳动网络技术有限公司 | Model generation method and device |
CN111382623A (en) * | 2018-12-28 | 2020-07-07 | 广州市百果园信息技术有限公司 | Live broadcast auditing method, device, server and storage medium |
Also Published As
Publication number | Publication date |
---|---|
WO2019085331A1 (en) | 2019-05-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108038413A (en) | Cheat probability analysis method, apparatus and storage medium | |
CN108021864A (en) | Character personality analysis method, device and storage medium | |
CN108053838B (en) | In conjunction with fraud recognition methods, device and the storage medium of audio analysis and video analysis | |
CN108038414A (en) | Character personality analysis method, device and storage medium based on Recognition with Recurrent Neural Network | |
CN109785928A (en) | Diagnosis and treatment proposal recommending method, device and storage medium | |
CN107491432A (en) | Low quality article recognition methods and device, equipment and medium based on artificial intelligence | |
CN107194158A (en) | A kind of disease aided diagnosis method based on image recognition | |
CN111260448A (en) | Artificial intelligence-based medicine recommendation method and related equipment | |
CN110619568A (en) | Risk assessment report generation method, device, equipment and storage medium | |
CN111813399B (en) | Machine learning-based auditing rule processing method and device and computer equipment | |
CN110163230A (en) | A kind of image labeling method and device | |
CN110246512A (en) | Sound separation method, device and computer readable storage medium | |
CN107168952A (en) | Information generating method and device based on artificial intelligence | |
CN109271493A (en) | A kind of language text processing method, device and storage medium | |
CN111785366A (en) | Method and device for determining patient treatment scheme and computer equipment | |
CN109670055A (en) | A kind of multi-medium data checking method, device, equipment and storage medium | |
CN109783749A (en) | A kind of Material for design intelligent recommendation method, apparatus and terminal device | |
CN110895568B (en) | Method and system for processing court trial records | |
CN110113634A (en) | A kind of information interaction method, device, equipment and storage medium | |
CN109431525A (en) | A kind of psychological sand table device and implementation method based on Internet of Things and virtual reality | |
CN111815169A (en) | Business approval parameter configuration method and device | |
CN113343106A (en) | Intelligent student recommendation method and system | |
CN107851113A (en) | Be configured as based on derived from performance sensor unit user perform attribute and realize the framework of automatic classification and/or search to media data, apparatus and method | |
CN110399933A (en) | Data mark modification method, device, computer-readable medium and electronic equipment | |
CN110503409A (en) | The method and relevant apparatus of information processing |
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 |