CN110334746A - A kind of image detecting method and device - Google Patents
A kind of image detecting method and device Download PDFInfo
- Publication number
- CN110334746A CN110334746A CN201910508310.7A CN201910508310A CN110334746A CN 110334746 A CN110334746 A CN 110334746A CN 201910508310 A CN201910508310 A CN 201910508310A CN 110334746 A CN110334746 A CN 110334746A
- Authority
- CN
- China
- Prior art keywords
- image
- information
- detected
- characteristic information
- pond
- 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
- 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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image detecting method and devices, the method includes obtaining image to be detected, according to preset Multi resolution feature extraction model, extract the Analysis On Multi-scale Features information of described image to be detected, the Analysis On Multi-scale Features information is subjected to dimensionality reduction, to obtain target image characteristics information, according to the target image characteristics information, search historical image characteristic information, to obtain lookup result, if there is the historical image characteristic information with the target image characteristics information matches in the lookup result, then call the corresponding history detection record of the historical image characteristic information, testing result by history detection record as image to be detected exports.The method avoids repetition detection, while can extract more abundant feature, being capable of preferably picture engraving.
Description
Technical field
The present invention relates to field of image detection more particularly to a kind of image detecting methods and device.
Background technique
Feature extraction is a concept in computer vision and image procossing.It refers to extracting image using computer
Information, determines whether the point of each image belongs to a characteristics of image.Its result is referred to as feature description or feature vector.Often
Characteristics of image has color characteristic, textural characteristics, shape feature, spatial relation characteristics.Picture feature extraction is widely used in
Under various scenes, such as recognition of face or image audit etc..
In the prior art, the feature that picture materials fingerprint only uses last convolutional layer is extracted, top spy is only used
Sign, the feature extracted is insufficient, and the feature to low layer is simultaneously unmodeled.When needing the similarity for carrying out image to judge, mentioning
In the case that the feature taken is insufficient, it is easy the similarity degree of erroneous judgement image, so as to cause repeating to detect, increases the meter of system
Calculation amount.
Summary of the invention
In order to solve the problems, such as feature information extraction and repeat to detect, obtains more fully characteristic information and spy can be retained
The spatial information of sign, while the technical effect for repeating to detect is avoided, the present invention provides a kind of image detecting method and devices.
On the one hand, the present invention provides a kind of image detecting methods, which comprises
Obtain image to be detected；
According to preset Multi resolution feature extraction model, the Analysis On Multi-scale Features information of described image to be detected is extracted；
The Analysis On Multi-scale Features information is subjected to dimensionality reduction, to obtain target image characteristics information；
According to the target image characteristics information, historical image characteristic information is searched, to obtain lookup result；
If there is the historical image characteristic information with the target image characteristics information matches in the lookup result, adjust
With the corresponding history detection record of the historical image characteristic information；
Testing result by history detection record as image to be detected exports.
On the other hand a kind of image detection device is provided, described device includes: image to be detected input module, image spy
It levies extraction module, characteristics of image dimensionality reduction module, image duplicate checking module, history detection record calling module and detection request and generates mould
Block；
Image to be detected input module is for obtaining image to be detected；
Image characteristics extraction module is used to extract described image to be detected according to preset Multi resolution feature extraction model
Analysis On Multi-scale Features information；
Characteristics of image dimensionality reduction module is used to the Analysis On Multi-scale Features information carrying out dimensionality reduction, to obtain target image characteristics letter
Breath；
Image duplicate checking module is used to historical image characteristic information is searched, to obtain according to the target image characteristics information
Lookup result；
If history detection record calling module in the lookup result for existing and target image characteristics information matches
Historical image characteristic information then calls the corresponding history detection record of historical image characteristic information, and the history is detected and is remembered
It records and is exported as the testing result of image to be detected；
Detection request generation module is used to be not present and target image characteristics information matches according in the lookup result
Historical image characteristic information then generates the detection request of described image to be detected.
On the other hand a kind of computer readable storage medium is provided, for storing program, described program is performed reality
A kind of existing image detecting method.
On the other hand a kind of terminal device is provided, the terminal device includes a kind of above-mentioned image detection device.
A kind of image detecting method and device provided by the invention, the method can extract the multiple dimensioned of image to be detected
Feature, and by Analysis On Multi-scale Features dimensionality reduction to obtain target image characteristics information.Search whether exist and target image characteristics information
Matched historical image characteristic information, then skip pictures detecting step if it exists, with the corresponding history of historical image characteristic information
Detection record is exported as the testing result of image to be detected.The method avoids repetition detection, while can extract more
The feature of horn of plenty, can preferably picture engraving, in addition, the method can also pass through ternary damage when carrying out data training
It loses function to optimize the model of image detection, obtains feature more with distinction.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of application scenarios schematic diagram of image detecting method provided in an embodiment of the present invention；
Fig. 2 is a kind of flow chart of image detecting method provided in an embodiment of the present invention；
Fig. 3 is the multiple dimensioned spy that described image to be detected is extracted in a kind of image detecting method provided in an embodiment of the present invention
The flow chart of the method for reference breath；
Fig. 4 is the schematic diagram of Multi resolution feature extraction model in a kind of image detecting method provided in an embodiment of the present invention；
Fig. 5 is to carry out pond to each foundation characteristic information in a kind of image detecting method provided in an embodiment of the present invention
The flow chart of method；
Fig. 6 is that the Analysis On Multi-scale Features information is carried out dimensionality reduction in a kind of image detecting method provided in an embodiment of the present invention
Method flow chart；
Fig. 7 is to be looked into a kind of image detecting method provided in an embodiment of the present invention according to the target image characteristics information
Look for the method flow diagram of historical image characteristic information；
Fig. 8 is to judge whether matched side according to Euclidean distance in a kind of image detecting method provided in an embodiment of the present invention
The flow chart of method；
Fig. 9 is a kind of schematic diagram of the training data model of image detecting method provided in an embodiment of the present invention；
Figure 10 is the stream of the calculation method of preset threshold described in a kind of image detecting method provided in an embodiment of the present invention
Cheng Tu；
Figure 11 is a kind of structural schematic diagram of image detection device provided in an embodiment of the present invention；
Figure 12 is that the structure of the image characteristics extraction module in a kind of image detection device provided in an embodiment of the present invention is shown
It is intended to；
Figure 13 is that the structure of the characteristics of image dimensionality reduction module in a kind of image detection device provided in an embodiment of the present invention is shown
It is intended to；
Figure 14 is a kind of equipment for realizing method provided by the embodiment of the present invention provided in an embodiment of the present invention
Hardware structural diagram.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.Obviously, described embodiment is only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, those of ordinary skill in the art are obtained without making creative work
Every other embodiment, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that, term " first ", " second " are used for description purposes only, and cannot
It is interpreted as indication or suggestion relative importance or implicitly indicates the quantity of indicated technical characteristic.Define as a result, " the
One ", the feature of " second " can explicitly or implicitly include one or more of the features.Moreover, term " first ",
" second " etc. is suitable for distinguishing similar object, without being used to describe a particular order or precedence order.It should be understood that in this way
The data used are interchangeable under appropriate circumstances, so that the embodiment of the present invention described herein can be in addition to scheming herein
Sequence other than those of showing or describe is implemented.
Relational language involved in the embodiment of the present invention is done first explained below:
VGG:VGG is Oxford University's visual geometric group (Visual Geometry Group, Department of
Engineering Science, University of Oxford) abbreviation.2014 are participated in for Oxford University's visual geometric group
Year image recognition network challenge match (ImageNet Large Scale Visual Recognition Challenge,
ILSVRC the network structure submitted when) is named as VGG with group name abbreviation, or is VGG network.
Referring to Figure 1, which show a kind of application scenarios schematic diagram of image detecting method provided in an embodiment of the present invention,
The application scenarios include user terminal 110 and server 120, and the user terminal includes but is not limited to smart phone, plate
Computer, laptop, desktop computer etc..The server obtains image to be detected information from user terminal, extract it is described to
The Analysis On Multi-scale Features of detection image information, and dimensionality reduction is carried out to the Analysis On Multi-scale Features of image to be detected information, to obtain mesh
Logo image characteristic information.Whether there is matched historical image characteristic to believe according in target image characteristics information searching server
Breath then calls directly testing result of the corresponding history detection record as image to be detected if it exists, server is avoided to treat
The repetition of detection image detects.
Fig. 2 is referred to, which show a kind of image detecting methods, can be applied to server side, which comprises
S210. image to be detected is obtained；
S220. according to preset Multi resolution feature extraction model, the Analysis On Multi-scale Features information of described image to be detected is extracted；
Further, Fig. 3 is referred to, it is described according to preset Multi resolution feature extraction model, extract the mapping to be checked
The Analysis On Multi-scale Features information of picture includes:
S310. described image to be detected is input in preset convolutional neural networks；
S320. according to preset convolutional neural networks, the foundation characteristic of described image to be detected in each convolution group is extracted
Information；
S330. pond is carried out to each foundation characteristic information, to obtain the pond of described image to be detected in each convolution group
Change characteristic information；
S340. splice the pond characteristic information of described image to be detected in each convolution group, to obtain target image
Characteristic information.
Specifically, it after image to be detected being adjusted size, is input in convolutional neural networks and carries out feature extraction, the volume
Product neural network can be 16 layers convolutional neural networks VGG network or depth residual error network etc. more rich in the net of expressivity
Network, in a specific example, when selecting VGG network as feature extraction network, the dimension of picture of input is 224*224's
Size, that is, need first to be adjusted to image to be detected the size of 224*224, then is input in VGG network.
The Analysis On Multi-scale Features are the feature of each convolution group in convolutional network, and the feature emphasis of different convolution groups is not
Together, for example, the convolutional layer of low layer extract be more low layer feature, such as the color of picture, edge etc., high-rise convolutional layer
What is extracted is more abstract feature, such as shape etc..By way of Multi resolution feature extraction, obtains different convolutional layers and mentioned
The feature taken can obtain more fully characteristic information, to preferably portray the feature of image to be detected.
In a specific example, Fig. 4 is referred to, when carrying out feature extraction by VGG network, 16 layers of VGG network
There are five convolution groups for tool, extract the foundation characteristic information in each convolution group, obtain the first convolution group foundation characteristic, second respectively
Convolution group foundation characteristic, third convolution group foundation characteristic, Volume Four product group foundation characteristic and the 5th convolution group foundation characteristic.
For different convolution group features, the dimension of feature is different.Such as first convolution group foundation characteristic have 64
Channel, the dimension in each channel are 224*224, and the second convolution group foundation characteristic has 128 channels, and the dimension in each channel is
112*112.Pond is carried out to the foundation characteristic information in each convolution group, obtains the pond characteristic information of each convolution group, institute
Fig. 5, the tool of the max pooling can be referred to by maximum value pond (max pooling) Lai Jinhang by stating pondization operation
Gymnastics is made as follows:
S510. the foundation characteristic information is divided into multiple pond regions；
S520. the provincial characteristics data in each pond region are obtained；
S530. the maximum value of the provincial characteristics data is selected, to obtain pond characteristic；
S540. splice the pond characteristic in each pond region, to obtain pond characteristic information.
Alternatively, the pondization operation can also activate pond (Regional by maximum value in convolution region
Maximum Activation of Convolutions Pooling, rmac pooling) Lai Jinhang, the rmac
Pooling, by carrying out sliding window by way of becoming window, by different size of window, can be obtained each on characteristic image
The provincial characteristics in a region.Max pooling is carried out to characteristic image, obtains global feature.By provincial characteristics and entirety
Feature is directly added or concatenates, and obtains the foundation characteristic information of Chi Huahou.
In a specific example, pondization operates the form using max pooling.By max pooling
Later, the first convolution group pond feature, the second convolution group pond feature, third convolution group pond feature, Volume Four are obtained respectively
Product group pond feature and the 5th convolution group pond feature.The dimension of first convolution group pond feature is 1*64*1*1, described
The dimension of second convolution group pond feature is 1*128*1*1, and the dimension of third convolution group pond feature is 1*256*1*1,
The dimension of the Volume Four product group pond feature is 1*512*1*1, and the dimension of the 5th convolution group pond feature is 1*512*
1*1。
Finally by the pond merging features of each convolution group, the Analysis On Multi-scale Features information of picture to be detected is obtained, it is described more
Scale feature information is the material fingerprint of image to be detected, is the feature vector that can represent the image, this feature vector
Dimension be 1472.
S230. the Analysis On Multi-scale Features information is subjected to dimensionality reduction, to obtain target image characteristics information；
Further, Fig. 6 is referred to, it is described that the Analysis On Multi-scale Features information is subjected to dimensionality reduction, to obtain target image spy
Reference ceases
S610. the mean vector of the Analysis On Multi-scale Features information is calculated；
S620. according to the mean vector, the correlation matrix of the Analysis On Multi-scale Features information is calculated；
S630. according to the correlation matrix, eigenvectors matrix is calculated；
S640 obtains target image characteristics information according to described eigenvector matrix and preset dimensionality reduction dimension.
Specifically, after obtaining target image characteristics information, dimensionality reduction is carried out to the target image characteristics information.The dimensionality reduction
Dimensionality reduction can be carried out using the method for principal component analysis (Principal Compinent Analysis, PCA).
PCA by calculate Analysis On Multi-scale Features information covariance matrix, obtain covariance matrix characteristic value and feature to
Amount selects characteristic value maximum, i.e. the matrix of the composition of feature vector corresponding to the maximum k feature of variance, so that it may by more rulers
Being transformed into new space for degree characteristic information, realizes the dimensionality reduction of data characteristics.
In the covariance matrix for calculating Analysis On Multi-scale Features information, when obtaining the characteristic value and feature vector of covariance matrix,
It can be calculated using the method for Eigenvalues Decomposition covariance matrix and singular value decomposition covariance matrix.
In a specific example, it can be realized by Eigenvalues Decomposition covariance matrix or singular value decomposition method
PCA dimensionality reduction.Analysis On Multi-scale Features information is inputted, and needs is set, the Analysis On Multi-scale Features information is dropped into k dimension.It will be described multiple dimensioned
Characteristic information removes average value, i.e. each feature subtracts respective average value.Covariance matrix is calculated, with Eigenvalues Decomposition method
Or singular value decomposition method seeks the eigen vector of covariance matrix.Characteristic value is sorted from large to small, selection is wherein
Maximum k.Then using its corresponding k feature vector as row vector composition characteristic vector matrix P.By more rulers
Degree characteristic information is transformed into the new space that k feature vector constructs to get to the Analysis On Multi-scale Features information after dimensionality reduction.
The method by PCA, can be by the Analysis On Multi-scale Features information of 1472 dimensions by Analysis On Multi-scale Features information dimensionality reduction
Dimensionality reduction reduces the feature of redundancy to 64 dimensions, avoids increasing calculation amount, causes the waste of time and resource, while passing through PCA dimensionality reduction
Have also been found that some potential characteristic variables.
S240. according to the target image characteristics information, historical image characteristic information is searched, to obtain lookup result；
Believe if S250. existing in the lookup result with the historical image characteristic of the target image characteristics information matches
Breath then calls the corresponding history detection record of the historical image characteristic information；
S260. the testing result by history detection record as image to be detected exports.
Further, Fig. 7 is referred to, it is described that historical image characteristic information is searched according to the target image characteristics information,
Include: to obtain lookup result
S710. historical image characteristic information is obtained；
S720. the Euclidean distance between the target image characteristics information and the historical image characteristic information is calculated.
Further, Fig. 8 is referred to, it is described to calculate the target image characteristics information and the historical image characteristic information
Between Euclidean distance include:
If S810. the Euclidean distance is greater than preset threshold, the detection request of described image to be detected is generated, with detection
Described image to be detected；
If S820. the Euclidean distance is less than preset threshold, the corresponding history of the historical image characteristic information is called
Detection record；
S830. the testing result by history detection record as image to be detected exports.
Specifically, when carrying out the comparison of target image characteristics information and historical image characteristic information, Euclidean can be used
Whether distance similar to judge target image characteristics information and historical image characteristic information, when the target image characteristics information and
When Euclidean distance between historical image characteristic information is less than preset value, then the target image characteristics information and the history figure
As characteristic information matching, the corresponding history detection record of historical image characteristic information is called, the history is detected and records conduct
The testing result of target image characteristics information exports.When between the target image characteristics information and historical image characteristic information
When Euclidean distance is greater than preset value, then the target image characteristics information and the historical image characteristic information mismatch, when not
When matching, then detection request is generated, picture to be detected is detected.
In a specific example, the method be can be applied in auditing system, be compared when by selected characteristic
It is right, when confirming in history picture in the presence of the picture high with picture similarity to be detected, by having invoked historical record as detection
As a result, skipping the detecting step to picture to be detected, avoid carrying out repeating detection.
When carrying out similarity detection to image to be detected and history image, the knowledge of image detection can be increased by training
Other ability, refers to Fig. 9, and described Fig. 9 is the training pattern of image detection.Training data is obtained first, passes through feature extraction mould
Type extracts the feature of each sample data in training data, is then clustered by preset algorithm to each data sample, passes through
It crosses in the sample data after clustering, the sample data for belonging to the same category should have similitude.It is each by manually marking
Sample data in class further excludes dissimilar sample, finally obtains multiple classes.Sample data in each class is similar,
Sample data between each class is dissimilar.
When carrying out model training, feature extraction is carried out to sample data, Analysis On Multi-scale Features can be used in the feature extraction
The model of extraction can carry out Multi resolution feature extraction using VGG network.The Analysis On Multi-scale Features of 1472 dimension of output after feature extraction
Information carries out dimensionality reduction to full articulamentum, in full articulamentum.The dimensionality reduction can using PCA method to Analysis On Multi-scale Features information into
Row dimensionality reduction obtains the target image characteristics information of 64 dimensions.It finally can be using ternary loss function (triplet loss) to institute
It states target image information to optimize, preset threshold can be obtained by the method for triplet loss, referring to Figure 10, it is described
The calculation method of preset threshold includes:
S1010. target sample information, positive sample information and negative sample information are obtained；
S1020. according to the target sample information, the positive sample information and the negative sample information, mesh is successively obtained
Mark sampling feature vectors, positive sample feature vector and negative sample feature vector；
S1030. calculate the Euclidean distance between target sample feature vector and positive sample feature vector, with obtain it is similar away from
From；
S1040. calculate the Euclidean distance between target sample feature vector and negative sample feature vector, with obtain foreign peoples away from
From；
S1050. according to the similar distance and foreign peoples's distance, preset threshold is calculated.
Specifically, when carrying out triplet loss optimization, the formula of triplet loss is as follows:
Wherein:The feature vector of positive sample a is represented,The feature vector of positive sample p is represented,
Represent the feature vector of negative sample n.The target of triplet loss optimization is that triplet loss is the smaller the better, i.e. positive sample a and just
The distance of sample pIt is the smaller the better, the distance of positive sample a and negative sample n
It is the bigger the better.
When using triplet loss training, the training data used is triple<A, B, C>, A and B is similar diagram in fact
Piece, A and C are dissimilar pictures, i.e.,<A, B>=1,<A, C>=0, stochastic gradient descent algorithm (Stochastic can be used
Gradient Descent, SGD) it optimizes.
Triplet loss is introduced in the training pattern of image detection, and the model learning of image detection can be made to more
Character representation with distinction improves the accuracy of image detection.
In a specific example, a kind of image detecting method described in the present embodiment can be applied to image audit system
In system.In described image auditing system, the side for providing image may provide same or similar image, if to institute
It states the same or similar image to carry out repeating audit, system resource can be wasted, by described image detection method, obtain to be detected
The characteristic information of image is compared with the history feature information in historical record, when such a match occurs, then before directly extracting
Audit logging, no longer audited.
A kind of image detecting method that the embodiment of the present invention proposes, the method extract the multiple dimensioned spy of image to be detected
Sign, and by Analysis On Multi-scale Features dimensionality reduction to obtain target image characteristics information.Search whether exist and target image characteristics information
The historical image characteristic information matched, then skip pictures detecting step, is examined with the corresponding history of historical image characteristic information if it exists
The testing result that record is surveyed as image to be detected exports.The method can obtain more horn of plenty by Multi resolution feature extraction
Feature, while avoid repetition detection, reduce calculation amount, avoid detection resource waste.
The embodiment of the invention also provides a kind of image detection device, referring to Figure 11, described device includes: mapping to be checked
As input module 1110, image characteristics extraction module 1120, characteristics of image dimensionality reduction module 1130, image duplicate checking module 1140, go through
History detection record calling module 1150 and detection request generation module 1160；
Image to be detected input module 1110 is for obtaining image to be detected；
Image characteristics extraction module 1120 is used to extract the mapping to be checked according to preset Multi resolution feature extraction model
The Analysis On Multi-scale Features information of picture；
Characteristics of image dimensionality reduction module 1130 is used to the Analysis On Multi-scale Features information carrying out dimensionality reduction, to obtain target image spy
Reference breath；
Image duplicate checking module 1140 is used to search historical image characteristic information according to the target image characteristics information, with
Obtain lookup result；
If history detection record calling module 1150 in the lookup result for existing and target image characteristics information
The historical image characteristic information matched then calls the corresponding history detection record of historical image characteristic information, and the history is examined
The testing result that record is surveyed as image to be detected exports；
Detection request generation module 1160 is used to be not present and target image characteristics information according in the lookup result
The historical image characteristic information matched then generates the detection request of described image to be detected.
Further, referring to Figure 12, described image characteristic extracting module 1120 include foundation characteristic extraction unit 1210,
Foundation characteristic pond unit 1220 and characteristic information concatenation unit 1230:
The foundation characteristic extraction unit 1210 is used to described image to be detected being input to preset convolutional neural networks
In, according to preset convolutional neural networks, extract the foundation characteristic information of described image to be detected in each convolution group；
Foundation characteristic pond unit 1220 is used to carry out pond to each foundation characteristic information, to obtain each convolution
The pond characteristic information of described image to be detected in group；
The characteristic information concatenation unit 1230 is used to splice the pond of described image to be detected in each convolution group
Characteristic information, to obtain target image characteristics information.
Further, referring to Figure 13, described image Feature Dimension Reduction module 1130 include mean vector computing unit 1310,
Correlation matrix calculation unit 1320, eigenvectors matrix computing unit 1330 and target image characteristics obtaining unit 1340；
The mean vector computing unit 1310 is used to calculate the mean vector of the Analysis On Multi-scale Features information；
The correlation matrix calculation unit 1320 is used to calculate the Analysis On Multi-scale Features information according to the mean vector
Correlation matrix；
Described eigenvector matrix calculation unit 1330 is used to that feature vector square to be calculated according to the correlation matrix
Battle array；
The target image characteristics obtaining unit 1340 is used for according to described eigenvector matrix and preset dimensionality reduction dimension,
Obtain target image characteristics information.
Any embodiment of that present invention institute providing method can be performed in the device provided in above-described embodiment, has execution this method
Corresponding functional module and beneficial effect.The not technical detail of detailed description in the above-described embodiments, reference can be made to the present invention is any
A kind of image detecting method provided by embodiment.
The present embodiment additionally provides a kind of computer readable storage medium, and computer is stored in the storage medium to be held
Row instruction, the computer executable instructions are loaded by processor and execute a kind of above-mentioned image detecting method of the present embodiment.
The present embodiment additionally provides a kind of equipment, and the equipment includes processor and memory, wherein the memory is deposited
Computer program is contained, the computer program is suitable for being loaded by the processor and executing a kind of above-mentioned image of the present embodiment
Detection method.
The equipment can be terminal, mobile terminal or server, and the equipment, which may also participate in, constitutes this hair
Device or system provided by bright embodiment.As shown in figure 14, terminal 1 (or 4 mobile terminals 14 or server 15) can
To include that one or more (using 1402a, 1402b ... ... in figures, 1402n to show) (processor 1402 can for processor 1402
To include but is not limited to the processing unit of Micro-processor MCV or programmable logic device FPGA etc.), storage for storing data
Device 1404 and transmitting device 1406 for communication function.It in addition to this, can also include: that display, input/output connect
Mouth (I/O interface), network interface, power supply and/or camera.It will appreciated by the skilled person that structure shown in Figure 14
Only illustrate, the structure of above-mentioned electronic device is not caused to limit.For example, mobile device 14 may also include than institute in Figure 14
Show more perhaps less component or with the configuration different from shown in Figure 14.
It is to be noted that said one or multiple processors 1402 and/or other data processing circuits lead to herein
Can often " data processing circuit " be referred to as.The data processing circuit all or part of can be presented as software, hardware, firmware
Or any other combination.In addition, data processing circuit for single independent processing module or all or part of can be integrated to meter
In any one in other elements in calculation machine terminal 14 (or mobile device).As involved in the embodiment of the present application,
The data processing circuit controls (such as the selection for the variable resistance end path connecting with interface) as a kind of processor.
Memory 1404 can be used for storing the software program and module of application software, as described in the embodiment of the present invention
Corresponding program instruction/the data storage device of method, the software journey that processor 1402 is stored in memory 1404 by operation
Sequence and module realize that above-mentioned one kind is based on from attention network thereby executing various function application and data processing
Timing behavior capture frame generation method.Memory 1404 may include high speed random access memory, may also include non-volatile memories
Device, such as one or more magnetic storage device, flash memory or other non-volatile solid state memories.In some instances, it deposits
Reservoir 1404 can further comprise the memory remotely located relative to processor 1402, these remote memories can pass through net
Network is connected to mobile device 15.The example of above-mentioned network includes but is not limited to internet, intranet, local area network, moves and lead to
Letter net and combinations thereof.
Transmitting device 1406 is used to that data to be received or sent via a network.Above-mentioned network specific example may include
The wireless network that the communication providers of terminal 14 provide.In an example, transmitting device 1406 includes a network
Adapter (Network Interface Controller, NIC), can be connected by base station with other network equipments so as to
It is communicated with internet.In an example, transmitting device 1406 can be radio frequency (Radio Frequency, RF) module,
It is used to wirelessly be communicated with internet.
Display can such as touch-screen type liquid crystal display (LCD), the liquid crystal display aloow user with
The user interface of terminal 14 (or mobile device) interacts.
Present description provides the method operating procedures as described in embodiment or flow chart, but based on routine or without creation
The labour of property may include more or less operating procedure.The step of enumerating in embodiment and sequence are only numerous steps
One of execution sequence mode, does not represent and unique executes sequence.System in practice or when interrupting product and executing, can be with
It is executed according to embodiment or method shown in the drawings sequence or parallel executes (such as parallel processor or multiple threads
Environment).
Structure shown in the present embodiment, only part-structure relevant to application scheme, is not constituted to this
The restriction for the equipment that application scheme is applied thereon, specific equipment may include more or fewer components than showing,
Perhaps certain components or the arrangement with different components are combined.It is to be understood that method disclosed in the present embodiment,
Device etc., may be implemented in other ways.For example, the apparatus embodiments described above are merely exemplary, for example,
The division of the module is only a kind of division of logic function, and there may be another division manner in actual implementation, such as more
A unit or assembly can be combined or can be integrated into another system, or some features can be ignored or not executed.It is another
Point, shown or discussed mutual coupling, direct-coupling or communication connection can be through some interfaces, device or
The indirect coupling or communication connection of unit module.
Based on this understanding, technical solution of the present invention substantially in other words the part that contributes to existing technology or
The all or part of person's technical solution can be embodied in the form of software products, which is stored in one
In a storage medium, including some instructions are used so that computer equipment (it can be personal computer, server, or
Network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.And storage medium above-mentioned includes:
USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random
Access Memory), the various media that can store program code such as magnetic or disk.
Those skilled in the art further appreciate that, respectively show in conjunction with what embodiment disclosed in this specification described
Example unit and algorithm steps, being implemented in combination with electronic hardware, computer software or the two, in order to clearly demonstrate
The interchangeability of hardware and software generally describes each exemplary composition and step according to function in the above description
Suddenly.These functions are implemented in hardware or software actually, the specific application and design constraint item depending on technical solution
Part.Professional technician can use different methods to achieve the described function each specific application, but this reality
Now it should not be considered as beyond the scope of the present invention.
The above, the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although referring to before
Stating embodiment, invention is explained in detail, those skilled in the art should understand that: it still can be to preceding
Technical solution documented by each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
It modifies or replaces, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of image detecting method, which is characterized in that the described method includes:
Obtain image to be detected；
According to preset Multi resolution feature extraction model, the Analysis On Multi-scale Features information of described image to be detected is extracted；
The Analysis On Multi-scale Features information is subjected to dimensionality reduction, to obtain target image characteristics information；
According to the target image characteristics information, historical image characteristic information is searched, to obtain lookup result；
If there is the historical image characteristic information with the target image characteristics information matches in the lookup result, institute is called
State the corresponding history detection record of historical image characteristic information；
Testing result by history detection record as image to be detected exports.
2. a kind of image detecting method according to claim 1, which is characterized in that described according to preset Analysis On Multi-scale Features
Model is extracted, the Analysis On Multi-scale Features information for extracting described image to be detected includes:
Described image to be detected is input in preset convolutional neural networks；
According to preset convolutional neural networks, the foundation characteristic information of described image to be detected in each convolution group is extracted；
Pond is carried out to each foundation characteristic information, to obtain the pond feature letter of described image to be detected in each convolution group
Breath；
Splice the pond characteristic information of described image to be detected in each convolution group, to obtain target image characteristics information.
3. a kind of image detecting method according to claim 2, which is characterized in that it is described to each foundation characteristic information into
Row pond includes: to obtain the pond characteristic information of described image to be detected in each convolution group
The foundation characteristic information is divided into multiple pond regions；
Obtain the provincial characteristics data in each pond region；
The maximum value of the provincial characteristics data is selected, to obtain pond characteristic；
Splice the pond characteristic in each pond region, to obtain pond characteristic information.
4. a kind of image detecting method according to claim 1, which is characterized in that described by the Analysis On Multi-scale Features information
Dimensionality reduction is carried out, includes: to obtain target image characteristics information
Calculate the mean vector of the Analysis On Multi-scale Features information；
According to the mean vector, the correlation matrix of the Analysis On Multi-scale Features information is calculated；
According to the correlation matrix, eigenvectors matrix is calculated；
According to described eigenvector matrix and preset dimensionality reduction dimension, target image characteristics information is obtained.
5. a kind of image detecting method according to claim 1, which is characterized in that described according to the target image characteristics
Information searches historical image characteristic information, includes: to obtain lookup result
Obtain historical image characteristic information；
Calculate the Euclidean distance between the target image characteristics information and the historical image characteristic information.
6. a kind of image detecting method according to claim 5, which is characterized in that described to calculate the target image characteristics
Euclidean distance between information and the historical image characteristic information includes:
If the Euclidean distance is greater than preset threshold, the detection request of described image to be detected is generated, it is described to be checked to detect
Altimetric image；
If the Euclidean distance is less than preset threshold, the corresponding history detection record of the historical image characteristic information is called；
Testing result by history detection record as image to be detected exports.
7. a kind of image detecting method according to claim 6, which is characterized in that the calculation method packet of the preset threshold
It includes:
Obtain target sample information, positive sample information and negative sample information；
According to the target sample information, the positive sample information and the negative sample information, target sample feature is successively obtained
Vector, positive sample feature vector and negative sample feature vector；
The Euclidean distance between target sample feature vector and positive sample feature vector is calculated, to obtain similar distance；
The Euclidean distance between target sample feature vector and negative sample feature vector is calculated, to obtain foreign peoples's distance；
According to the similar distance and foreign peoples's distance, preset threshold is calculated.
8. a kind of image detection device, which is characterized in that described device includes: that image to be detected input module, characteristics of image mention
Modulus block, characteristics of image dimensionality reduction module, image duplicate checking module, history detection record calling module and detection request generation module；
Image to be detected input module is for obtaining image to be detected；
Image characteristics extraction module is used to extract more rulers of described image to be detected according to preset Multi resolution feature extraction model
Spend characteristic information；
Characteristics of image dimensionality reduction module is used to the Analysis On Multi-scale Features information carrying out dimensionality reduction, to obtain target image characteristics information；
Image duplicate checking module is used to historical image characteristic information is searched, to be searched according to the target image characteristics information
As a result；
If for there is the history with target image characteristics information matches in the lookup result in history detection record calling module
Image feature information then calls the corresponding history detection record of historical image characteristic information, and the history is detected record and is made
It is exported for the testing result of image to be detected；
Detection request generation module is used for according to the history being not present in the lookup result with target image characteristics information matches
Image feature information then generates the detection request of described image to be detected.
9. a kind of image detection device according to claim 8, which is characterized in that described image characteristic extracting module includes
Foundation characteristic extraction unit, foundation characteristic pond unit and characteristic information concatenation unit:
The foundation characteristic extraction unit is for described image to be detected to be input in preset convolutional neural networks, according to pre-
If convolutional neural networks, extract the foundation characteristic information of described image to be detected in each convolution group；
Foundation characteristic pond unit is used to carry out pond to each foundation characteristic information, described in each convolution group to obtain
The pond characteristic information of image to be detected；
The characteristic information concatenation unit is used to splice the pond characteristic information of described image to be detected in each convolution group,
To obtain target image characteristics information.
10. a kind of image detection device according to claim 8, which is characterized in that described image Feature Dimension Reduction module packet
Mean vector computing unit, correlation matrix calculation unit, eigenvectors matrix computing unit and target image characteristics are included to obtain
Unit；
The mean vector computing unit is used to calculate the mean vector of the Analysis On Multi-scale Features information；
The correlation matrix calculation unit is used to calculate the correlation of the Analysis On Multi-scale Features information according to the mean vector
Matrix；
Described eigenvector matrix calculation unit is used to that eigenvectors matrix to be calculated according to the correlation matrix；
The target image characteristics obtaining unit is used to obtain target according to described eigenvector matrix and preset dimensionality reduction dimension
Image feature information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910508310.7A CN110334746A (en) | 2019-06-12 | 2019-06-12 | A kind of image detecting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910508310.7A CN110334746A (en) | 2019-06-12 | 2019-06-12 | A kind of image detecting method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110334746A true CN110334746A (en) | 2019-10-15 |
Family
ID=68140402
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910508310.7A Pending CN110334746A (en) | 2019-06-12 | 2019-06-12 | A kind of image detecting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110334746A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079594A (en) * | 2019-12-04 | 2020-04-28 | 成都考拉悠然科技有限公司 | Video action classification and identification method based on double-current cooperative network |
| CN111710437A (en) * | 2020-05-25 | 2020-09-25 | 广东百慧科技有限公司 | Intelligent inquiry method, system and storage medium based on image processing |
| CN111737512A (en) * | 2020-06-04 | 2020-10-02 | 东华大学 | Silk cultural relic image retrieval method based on depth feature region fusion |
| CN113569909A (en) * | 2021-06-24 | 2021-10-29 | 阿里巴巴新加坡控股有限公司 | Data processing method and device, electronic equipment and storage medium |
| CN113569070A (en) * | 2021-07-24 | 2021-10-29 | 平安科技（深圳）有限公司 | Image detection method and device, electronic equipment and storage medium |
| CN114781548A (en) * | 2022-05-18 | 2022-07-22 | 平安科技（深圳）有限公司 | Image scene classification method, device, equipment and storage medium |
-
2019
- 2019-06-12 CN CN201910508310.7A patent/CN110334746A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079594A (en) * | 2019-12-04 | 2020-04-28 | 成都考拉悠然科技有限公司 | Video action classification and identification method based on double-current cooperative network |
| CN111079594B (en) * | 2019-12-04 | 2023-06-06 | 成都考拉悠然科技有限公司 | Video action classification and identification method based on double-flow cooperative network |
| CN111710437A (en) * | 2020-05-25 | 2020-09-25 | 广东百慧科技有限公司 | Intelligent inquiry method, system and storage medium based on image processing |
| CN111737512A (en) * | 2020-06-04 | 2020-10-02 | 东华大学 | Silk cultural relic image retrieval method based on depth feature region fusion |
| CN111737512B (en) * | 2020-06-04 | 2021-11-12 | 东华大学 | Silk cultural relic image retrieval method based on depth feature region fusion |
| CN113569909A (en) * | 2021-06-24 | 2021-10-29 | 阿里巴巴新加坡控股有限公司 | Data processing method and device, electronic equipment and storage medium |
| CN113569070A (en) * | 2021-07-24 | 2021-10-29 | 平安科技（深圳）有限公司 | Image detection method and device, electronic equipment and storage medium |
| CN114781548A (en) * | 2022-05-18 | 2022-07-22 | 平安科技（深圳）有限公司 | Image scene classification method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110334746A (en) | A kind of image detecting method and device | |
| Li et al. | Instance-level salient object segmentation | |
| US10176364B2 (en) | Media content enrichment using an adapted object detector | |
| US20230119593A1 (en) | Method and apparatus for training facial feature extraction model, method and apparatus for extracting facial features, device, and storage medium | |
| CN105844283B (en) | Method, image search method and the device of image classification ownership for identification | |
| Fang et al. | Bottom-up saliency detection model based on human visual sensitivity and amplitude spectrum | |
| CN106022317A (en) | Face identification method and apparatus | |
| CN106713811B (en) | Video call method and device | |
| CN112270686B (en) | Image segmentation model training method, image segmentation device and electronic equipment | |
| Yan et al. | Multiscale convolutional neural networks for hand detection | |
| CN109598234A (en) | Critical point detection method and apparatus | |
| US11714921B2 (en) | Image processing method with ash code on local feature vectors, image processing device and storage medium | |
| CN110765882B (en) | Video tag determination method, device, server and storage medium | |
| CN110245714B (en) | Image recognition method and device and electronic equipment | |
| CN110866469B (en) | Facial five sense organs identification method, device, equipment and medium | |
| Furnari et al. | Recognizing personal contexts from egocentric images | |
| CN107729848A (en) | Method for checking object and device | |
| CN109993021A (en) | The positive face detecting method of face, device and electronic equipment | |
| CN110598715A (en) | Image recognition method and device, computer equipment and readable storage medium | |
| CN110457677A (en) | Entity-relationship recognition method and device, storage medium, computer equipment | |
| CN112818995B (en) | Image classification method, device, electronic equipment and storage medium | |
| JP2023508704A (en) | Face keypoint detection method, device and electronic device | |
| Sui et al. | ScanDMM: A deep markov model of scanpath prediction for 360deg images | |
| Huo et al. | Semisupervised learning based on a novel iterative optimization model for saliency detection | |
| CN114492601A (en) | Resource classification model training method and device, electronic equipment and storage medium |
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 |