CN109685785A - A kind of image quality measure method, apparatus and electronic equipment - Google Patents
A kind of image quality measure method, apparatus and electronic equipment Download PDFInfo
- Publication number
- CN109685785A CN109685785A CN201811563932.1A CN201811563932A CN109685785A CN 109685785 A CN109685785 A CN 109685785A CN 201811563932 A CN201811563932 A CN 201811563932A CN 109685785 A CN109685785 A CN 109685785A
- Authority
- CN
- China
- Prior art keywords
- image
- parameter
- training
- sample
- assessed
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a kind of image quality measure method, apparatus and electronic equipments, can determine image to be assessed；Image to be assessed is inputted into assessment network model, obtains image to be assessed for the assessed value of each image parameter in preset quantity image parameter；Wherein, assessment network model is the model obtained according to training set training, and training set includes: the sample value of multiple sample images and each sample image for each image parameter in preset quantity image parameter.In training pattern, it is contemplated that multiple images parameter, and joint training is carried out for multiple images parameter, to improve the accuracy of image quality measure.
Description
Technical field
The present invention relates to computer application technologies, more particularly to a kind of image quality measure method, apparatus and electricity
Sub- equipment.
Background technique
Current most of video softwares or video website all support user's uploaded videos, and determine from uploaded videos
Video surface plot.However, the picture quality in user's uploaded videos is generally poor, to obtain the preferable image of quality as video
Surface plot needs to carry out quality evaluation to multiple images for including in video.
Currently, carrying out quality evaluation primarily directed to two kinds of acutance, brightness image parameters to image.Specifically, to image
When carrying out quality evaluation, the acutance of image is extracted according to the acutance algorithm of setting, extracts image according to the brightness algorithm of setting
Brightness, sharpness of selection height and image of high brightness from multiple images, as video surface plot.Here, the acutance of image and bright
Degree is extracted respectively by different algorithms, not in view of the relationship of the two, so that image cannot be evaluated effectively
Quality also just can not accurately determine the preferable image of mass.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of image quality measure method, apparatus and electronic equipment, to improve
The accuracy of image quality measure.Specific technical solution is as follows:
To achieve the above object, the embodiment of the invention provides a kind of image quality measure methods, which comprises
Determine image to be assessed；
By the image input assessment network model to be assessed, the image to be assessed is obtained for preset quantity image
The assessed value of each image parameter in parameter；
Wherein, the assessment network model is the model obtained according to training set training, and the training set includes: multiple samples
This image and each sample image are directed to the sample value of each image parameter in the preset quantity image parameter.
Optionally, described image parameter include: image definition, color saturation, it is one or more in default evaluation；Institute
Stating default evaluation includes the evaluation of image Attraction Degree and image scene evaluation.
Optionally, the assessment network model includes assessing network correspondingly with the preset quantity image parameter
Submodel；
It is described that network model is assessed into the image input to be assessed, the image to be assessed is obtained for preset quantity
In image parameter the step of the assessed value of each image parameter, comprising:
The image to be assessed is inputted into each assessment network submodel in the assessment network submodel respectively, is obtained
Assessed value of the image to be assessed for each image parameter in the preset quantity image parameter.
Optionally, the assessment network model is obtained using following steps training:
Obtain preset neural network model and the training set；
The multiple sample image is inputted into the neural network model, each sample image is obtained and joins for each image
Several assessed values；
Each image parameter is directed to according to each sample image for including in obtained assessed value and the training set
Sample value calculates training penalty values；
Determine whether the neural network model restrains according to the trained penalty values；
If it is not, then adjusting the parameter value in the neural network model, and return described that the multiple sample image is defeated
Enter the neural network model, obtains the step of each sample image is directed to the assessed value of each image parameter；
If so, current neural network model is determined as to assess network model.
Optionally, the basis obtains assessed value and each sample image for including in the training set are for every
The sample value of one image parameter calculates the step of training penalty values, comprising:
The each sample image for including in the obtained assessed value and the training set is directed to each image parameter
Sample value input preset loss function, obtain training penalty values.
Optionally, the basis obtains assessed value and each sample image for including in the training set are for every
The sample value of one image parameter calculates the step of training penalty values, comprising:
For each image parameter, the assessed value of the image parameter, Yi Jisuo are directed to according to obtained each sample image
The sample value that each sample image for including in training set is directed to the image parameter is stated, determines the loss for being directed to the image parameter
Value；
According to the training weight of penalty values and preset each image parameter for each image parameter, training damage is determined
Mistake value.
Optionally, the basis is for the penalty values of each image parameter and the training power of preset each image parameter
Weight determines the step of training penalty values, comprising:
According to following formula, training penalty values s is determined:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N
For the number of image parameter.
To achieve the above object, the embodiment of the invention also provides a kind of image quality measure device, described device includes:
Determining module, for determining image to be assessed；
Evaluation module, for obtaining the image input assessment network model to be assessed the image to be assessed and being directed to
The assessed value of each image parameter in preset quantity image parameter；
Wherein, the assessment network model is the model obtained according to training set training, and the training set includes: multiple samples
This image and each sample image are directed to the sample value of each image parameter in the preset quantity image parameter.
Optionally, described image parameter include: image definition, color saturation, it is one or more in default evaluation；Institute
Stating default evaluation includes the evaluation of image Attraction Degree and image scene evaluation.
Optionally, the assessment network model includes assessing network correspondingly with the preset quantity image parameter
Submodel；
The evaluation module, is specifically used for:
The image to be assessed is inputted into each assessment network submodel in the assessment network submodel respectively, is obtained
Assessed value of the image to be assessed for each image parameter in the preset quantity image parameter.
Optionally, described device further include:
Training module, for training the assessment network model；
The training module, is specifically used for:
Obtain preset neural network model and the training set；
The multiple sample image is inputted into the neural network model, each sample image is obtained and joins for each image
Several assessed values；
Each image parameter is directed to according to each sample image for including in obtained assessed value and the training set
Sample value calculates training penalty values；
Determine whether the neural network model restrains according to the trained penalty values；
If it is not, then adjusting the parameter value in the neural network model, and return described that the multiple sample image is defeated
Enter the neural network model, obtains the step of each sample image is directed to the assessed value of each image parameter；
If so, current neural network model is determined as to assess network model.
Optionally, described device further includes penalty values determining module,
The penalty values determining module, is specifically used for:
The each sample image for including in the obtained assessed value and the training set is directed to each image parameter
Sample value input preset loss function, obtain training penalty values.
Optionally, the penalty values determining module, is specifically used for:
For each image parameter, the assessed value of the image parameter, Yi Jisuo are directed to according to obtained each sample image
The sample value that each sample image for including in training set is directed to the image parameter is stated, determines the loss for being directed to the image parameter
Value；
According to the training weight of penalty values and preset each image parameter for each image parameter, training damage is determined
Mistake value.
Optionally, the penalty values determining module, is specifically used for:
According to following formula, training penalty values s is determined:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N
For the number of image parameter.
To achieve the above object, the embodiment of the invention also provides a kind of electronic equipment, including processor, communication interface,
Memory and communication bus, wherein processor, communication interface, memory complete mutual communication by bus, memory,
For storing computer program；Processor when for executing the program stored on memory, realizes any of the above-described method step
Suddenly.
To achieve the above object, computer-readable the embodiment of the invention also provides a kind of computer readable storage medium
Computer program is stored in storage medium, the computer program realizes any of the above-described method and step when being executed by processor.
As it can be seen that the embodiment of the invention provides a kind of image quality measure method, apparatus and electronic equipment, can determine to
Assess image；Image to be assessed is inputted into assessment network model, obtains image to be assessed in preset quantity image parameter
The assessed value of each image parameter；Wherein, assessment network model is the model obtained according to training set training, and training set includes:
Multiple sample images and each sample image are directed to the sample value of each image parameter in preset quantity image parameter.It can
See, in the embodiment of the present invention, in training pattern, it is contemplated that multiple images parameter, and combined for multiple images parameter
Training, to improve the accuracy of image quality measure.
Certainly, it implements any of the products of the present invention or method must be not necessarily required to reach all the above excellent simultaneously
Point.
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.
Fig. 1 is a kind of flow diagram of image quality measure method provided in an embodiment of the present invention；
Fig. 2 is another flow diagram of image quality measure method provided in an embodiment of the present invention；
Fig. 3 is a kind of schematic diagram of image quality measure provided in an embodiment of the present invention；
Fig. 4 is a kind of structural schematic diagram of image quality measure device provided in an embodiment of the present invention；
Fig. 5 is a kind of structural schematic diagram of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.
In order to solve the existing assessment to video cover plot quality, image sharpness, brightness are only considered, to can not have
The problem of effect evaluates picture quality, the embodiment of the invention provides a kind of image quality measure method, this method can be applied
In electronic equipment or server, the accuracy of image quality measure can be improved.
Combined with specific embodiments below, above-mentioned image quality measure method is illustrated.
Referring to Fig. 1, Fig. 1 is a kind of flow chart of image quality measure method provided in an embodiment of the present invention, and method includes
Following steps:
Step S101: image to be assessed is determined.
In embodiments of the present invention, image to be assessed can be the image that user uploads to video software or video website,
Or the image obtained from network.
S102: by the characteristic input assessment network model of image to be assessed, image to be assessed is obtained for present count
Measure the assessed value of each image parameter in an image parameter.Wherein, assessment network model is the mould obtained according to training set training
Type, training set include: multiple sample images and each sample image for image ginseng each in preset quantity image parameter
Several sample values.
In embodiments of the present invention, image parameter both may include the characterisitic parameter of image inherently, such as image is clear
Clear degree, color saturation can also include the default evaluation to image, wherein default evaluation can give in advance for personnel such as users
Evaluation out, such as the Attraction Degree of image is evaluated, image scene evaluation.
Above-mentioned preset quantity image parameter is used equally for assessment picture quality.For example, by the prediction of multiple images parameter
The higher image of assessed value, is determined as the higher image of quality.In specific implementation, one in above-mentioned image parameter can be chosen
Kind or a variety of pairs of images carry out quality evaluations.
In embodiments of the present invention, each image parameter for the sample image for including in training set is known, each sample
The sample value of the image parameter of this image can be sample image is analyzed so that it is determined that, be also possible to artificial determination
, specifically it is determined and can be selected according to the actual situation using which kind of mode.For example, for image definition or color
Color saturation degree can analyze sample image, so that it is full to obtain the true image definition of the sample image or color
And degree.Attraction Degree evaluation for sample image, then can be taking human as being determined, if attractive scene in the sample image
Or personage, then the Attraction Degree of the sample image is had higher rating.For sample image scene evaluation can also taking human as determination, if
The scene for including in the sample image is compared with horn of plenty, then image scene is had higher rating.
It in embodiments of the present invention, can be each image parameter in order to intuitively embody the quality of each image parameter
Multiple grades are set.For example, three grades is arranged for " attractive degree ", with grade 1, grade 2 by taking " attractive degree " as an example
It is marked with grade 3, grade 1 indicates " not attractive ", and grade 2 indicates " more attractive ", and the expression of grade 3 " attracts very much
People ".
In embodiments of the present invention, assessment network model may include preset quantity assessment network submodel, the assessment
The quantity of network submodel is identical as the quantity for the image parameter that needs are assessed.The corresponding image of one assessment network submodel
Parameter.Image to be assessed, can will be to for the assessed value of each image parameter in preset quantity image parameter in order to obtain
Assessment image inputs each assessment network submodel in preset quantity assessment network submodel respectively, and then obtains to be assessed
Assessed value of the image for each image parameter in preset quantity image parameter.It is pre- according to training set for assessing network submodel
It is first trained.
As an example, it may refer to Fig. 3, image parameter may include image definition, color saturation, image suction
Degree of drawing evaluation and image scene evaluate this 4 kinds, then assess in network model comprising 4 assessment network submodels, correspond respectively to
First assessment network submodel of image definition assesses network submodel corresponding to the second of color saturation, corresponds to figure
The third assessment network submodel evaluated as Attraction Degree and the 4th assessment network submodel evaluated corresponding to image scene.
After image to be assessed is inputted above-mentioned 4 trained assessment network submodels respectively, the first assessment network submodule
Type can export the assessed value for image definition；Second assessment network submodel can export commenting for color saturation
Valuation；Third assessment network submodel can export the assessed value for the evaluation of image Attraction Degree；4th assessment network submodel
The assessed value for image scene evaluation can be exported.
In embodiments of the present invention, above-mentioned neural network submodel can be Recognition with Recurrent Neural Network model, convolutional Neural net
Network model, cyclic convolution neural network model, deep neural network model etc..It is not limited in the embodiment of the present invention.
As it can be seen that image quality measure method provided in an embodiment of the present invention, can determine image to be assessed, by figure to be assessed
As input assessment network model, image to be assessed is obtained for the pre- assessment of each image parameter in preset quantity image parameter
Valuation.Wherein, assessment network model is the model obtained according to training set training, and training set includes: multiple sample images, and
Sample value of each sample image for each image parameter in preset quantity image parameter.As it can be seen that in the embodiment of the present invention,
In training pattern, it is contemplated that multiple images parameter, wherein both include the characterisitic parameter of image inherently, also include default
Evaluation parameter, and joint training is carried out for multiple images parameter, it is trained respectively compared to single image parameter, it can
Improve the accuracy of image quality measure.
In embodiments of the present invention, referring to fig. 2, the training process for assessing network model may refer to following steps:
Step S201: preset neural network model and training set are obtained.
In embodiments of the present invention, it may include preset quantity neural network submodule in preset neural network model
Type, neural network submodel and image parameter correspond, and the process of model training is to update to join in neural network submodel
Several processes.
It include multiple sample images and each sample image in training set for each in preset quantity image parameter
The sample value of image parameter.
Step S202: inputting neural network model for multiple sample images, obtains each sample image for each image
The assessed value of parameter.
In this step, if including preset quantity neural network submodel in preset neural network model, by sample
Image is inputted respectively in the multiple neural network submodels for including in preset neural network model, obtains joining for multiple images
Several assessed values.Detailed process and step S102 in embodiment illustrated in fig. 1 are essentially identical, may refer to step S102, herein not
It repeats.
Step S203: each image is directed to according to each sample image for including in obtained assessed value and training set
The sample value of parameter calculates training penalty values.
In embodiments of the present invention, the penalty values of entire neural network model can be by the damage of multiple neural network submodels
Mistake value determines.
Above-mentioned steps S203 can specifically include following refinement step:
Step S203a: for each image parameter, the assessment of the image parameter is directed to according to obtained each sample image
The each sample image for including in value and training set is directed to the sample value of the image parameter, determines for the image parameter
Penalty values.
In this step, it can be first directed to each image parameter, calculate a penalty values.Specifically, in step S202,
The available assessed value for the image parameter.In conjunction with the sample value for the image parameter for including in training set, i.e.,
The penalty values for the image parameter can be calculated.
In the embodiment of the present invention, it can will be directed to for include in the assessed value of each image parameter and training set
The sample value of each image parameter substitutes into preset loss function, obtains the penalty values for each image parameter.The present invention is real
It applies in example, including but not limited to, as loss function, is obtained using mean square error (Mean Squared Error, MSE) formula
Penalty values.
Step S203b: according to for each image parameter penalty values and preset each image parameter training weight,
Determine training penalty values.
In embodiments of the present invention, it can will be directed to the penalty values weighted calculation of each image parameter, obtain final instruction
Practice penalty values.The weight of each image parameter can be preset by user.
In the embodiment of the present invention, training penalty values s can be determined according to following formula:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N
For the number of image parameter.
As an example, the pre-set image definition of user, color saturation, the evaluation of image Attraction Degree and image
The weight of scene evaluation is respectively 0.2,0.1,0.5 and 0.2, and is a for the penalty values of image definition, is saturated for color
The penalty values of degree are b, and the penalty values for the evaluation of image Attraction Degree are c, and the penalty values for image scene evaluation are d, then most
The training penalty values of whole quality evaluation are 0.2*a+0.1*b+0.5*c+0.2*d.
Step S204: it determines whether neural network model restrains according to training penalty values, is to then follow the steps S205；Otherwise
Return to step S202.
Step S205: current neural network model is determined as to assess network model.
In embodiments of the present invention, the size of final penalty values and default loss threshold value can be compared, to judge nerve net
Whether network model restrains.If not converged, the parameter value in neural network model is adjusted, S202 is returned to step.If convergence,
Then illustrate that neural network model training is completed, current neural network model can be determined as assessing network model.
As it can be seen that in embodiments of the present invention, for each image parameter, a neural network submodel can be constructed.
And be trained the neural network model of above-mentioned multiple neural network submodels as a whole, it in the training process, can
With the penalty values of each neural network submodel of synthesis, the final trained penalty values of entire neural network model are calculated.And root
The parameter value that each neural network submodel is adjusted according to final training penalty values realizes multiple neural network submodel joint instructions
Practice.It is individually trained compared to for a certain image parameter, the result of joint training is more accurate.
Based on identical inventive concept, according to above-mentioned image quality measure embodiment of the method, the embodiment of the present invention is also provided
A kind of image quality measure device referring to fig. 4 may include with lower module:
Determining module 401, for determining image to be assessed.
Evaluation module 402 obtains image to be assessed for present count for image to be assessed to be inputted assessment network model
Measure the assessed value of each image parameter in an image parameter；
Wherein, assessment network model is the model obtained according to training set training, and training set includes: multiple sample images,
And each sample image is for the sample value of each image parameter in preset quantity image parameter.
In embodiments of the present invention, image parameter include: image definition, color saturation, it is a kind of in default evaluation or
It is a variety of；Default evaluation includes the evaluation of image Attraction Degree and image scene evaluation.
In embodiments of the present invention, assessment network model includes assessing net correspondingly with preset quantity image parameter
String bag model.
Evaluation module 402, is specifically used for: image to be assessed is inputted to each assessment net in assessment network submodel respectively
String bag model obtains image to be assessed for the assessed value of each image parameter in preset quantity image parameter.
In embodiments of the present invention, it can also include: training module on device basic shown in Fig. 4, be used for Training valuation
Network model.
Training module is specifically used for:
Obtain preset neural network model and training set；
Multiple sample images are inputted into neural network model, obtain the assessment that each sample image is directed to each image parameter
Value；
The sample of each image parameter is directed to according to each sample image for including in obtained assessed value and training set
Value calculates training penalty values；
Determine whether neural network model restrains according to training penalty values；
If it is not, then adjusting the parameter value in neural network model, and returns and multiple sample images are inputted into neural network mould
Type obtains the step of each sample image is directed to the assessed value of each image parameter；
If so, current neural network model is determined as to assess network model.
In embodiments of the present invention, on device basic shown in Fig. 4, can also include, penalty values determining module, penalty values
Determining module is specifically used for:
Obtained assessed value and each sample image for including in training set are directed to the sample value of each image parameter
Preset loss function is inputted, obtains training penalty values.
In embodiments of the present invention, penalty values determining module specifically can be used for:
For each image parameter, the assessed value of the image parameter, Yi Jixun are directed to according to obtained each sample image
Practice the sample value for concentrating each sample image for including to be directed to the image parameter, determines the penalty values for being directed to the image parameter；
According to the training weight of penalty values and preset each image parameter for each image parameter, training damage is determined
Mistake value.
In embodiments of the present invention, penalty values determining module specifically can be used for:
According to following formula, training penalty values s is determined:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N
For the number of image parameter.
As it can be seen that in embodiments of the present invention, can determine image to be assessed；Image to be assessed is inputted into assessment network mould
Type obtains image to be assessed for the assessed value of each image parameter in preset quantity image parameter；Wherein, network mould is assessed
Type is the model obtained according to training set training, and training set includes: multiple sample images and each sample image for default
The sample value of each image parameter in quantity image parameter.As it can be seen that in the embodiment of the present invention, in training pattern, it is contemplated that
Multiple images parameter, and joint training is carried out for multiple images parameter, to improve the accuracy of image quality measure.
Based on identical inventive concept, according to above-mentioned image quality measure embodiment of the method, the embodiment of the present invention is also provided
A kind of electronic equipment, as shown in figure 5, include processor 501, communication interface 502, memory 503 and communication bus 504,
In, processor 501, communication interface 502, memory 503 completes mutual communication by communication bus 504,
Memory 503, for storing computer program；
Processor 501 when for executing the program stored on memory 503, realizes image matter shown in above-mentioned Fig. 1-4
Measure appraisal procedure embodiment.Wherein, image quality measure method includes:
Determine image to be assessed；Image to be assessed is inputted into assessment network model, obtains image to be assessed for present count
Measure the assessed value of each image parameter in an image parameter；Wherein, assessment network model is the mould obtained according to training set training
Type, training set include: multiple sample images and each sample image for image ginseng each in preset quantity image parameter
Several sample values.
Communication bus 504 can be Peripheral Component Interconnect standard (Peripheral Component Interconnect,
PCI) bus or expanding the industrial standard structure (Extended Industry Standard Architecture, EISA) bus
Deng.The communication bus 504 can be divided into address bus, data/address bus, control bus etc..For convenient for indicating, only with one in Fig. 5
Thick line indicates, it is not intended that an only bus or a type of bus.
Communication interface 502 is for the communication between above-mentioned electronic equipment and other equipment.
Memory 503 may include random access memory (Random Access Memory, RAM), also may include
Nonvolatile memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory
503 can also be that at least one is located remotely from the storage device of aforementioned processor.
Above-mentioned processor 501 can be general processor, including central processing unit (Central Processing
Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be digital signal processor (Digital
Signal Processing, DSP), specific integrated circuit (Application Specific Integrated Circuit,
ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic
Device, discrete gate or transistor logic, discrete hardware components.
Based on identical inventive concept, according to above-mentioned image quality measure embodiment of the method, provided by the invention another
In embodiment, a kind of computer readable storage medium is additionally provided, computer journey is stored in the computer readable storage medium
Sequence realizes any image method for evaluating quality step shown in above-mentioned Fig. 1-4 when computer program is executed by processor.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for image matter
For amount assessment device, electronic equipment and computer readable storage medium embodiment, implement since it is substantially similar to method
Example, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.The above is only it is of the invention compared with
Good embodiment, is not intended to limit the scope of the present invention.Done within the spirit and principles of the present invention
What modification, equivalent replacement, improvement etc., is included within the scope of protection of the present invention.
Claims (15)
1. a kind of image quality measure method, which is characterized in that the described method includes:
Determine image to be assessed；
By the image input assessment network model to be assessed, the image to be assessed is obtained for preset quantity image parameter
In each image parameter assessed value；
Wherein, the assessment network model is the model obtained according to training set training, and the training set includes: multiple sample graphs
Picture and each sample image are directed to the sample value of each image parameter in the preset quantity image parameter.
2. the method according to claim 1, wherein described image parameter includes: image definition, color saturation
It is one or more in degree, default evaluation；The default evaluation includes the evaluation of image Attraction Degree and image scene evaluation.
3. method according to claim 1 or 2, which is characterized in that the assessment network model includes and the present count
It measures an image parameter and assesses network submodel correspondingly；
It is described that network model is assessed into the image input to be assessed, the image to be assessed is obtained for preset quantity image
In parameter the step of the assessed value of each image parameter, comprising:
The image to be assessed is inputted into each assessment network submodel in the assessment network submodel respectively, is obtained described
Assessed value of the image to be assessed for each image parameter in the preset quantity image parameter.
4. method according to claim 1 or 2, which is characterized in that the assessment network model is using following steps training
It obtains:
Obtain preset neural network model and the training set；
The multiple sample image is inputted into the neural network model, obtains each sample image for each image parameter
Assessed value；
The sample of each image parameter is directed to according to each sample image for including in obtained assessed value and the training set
Value calculates training penalty values；
Determine whether the neural network model restrains according to the trained penalty values；
If it is not, then adjusting the parameter value in the neural network model, and return described by the multiple sample image input institute
Neural network model is stated, the step of each sample image is directed to the assessed value of each image parameter is obtained；
If so, current neural network model is determined as to assess network model.
5. according to the method described in claim 4, it is characterized in that, assessed value that the basis obtains and the training set
In include each sample image be directed to each image parameter sample value, calculate training penalty values the step of, comprising:
The each sample image for including in the obtained assessed value and the training set is directed to the sample of each image parameter
This value inputs preset loss function, obtains training penalty values.
6. according to the method described in claim 5, it is characterized in that, assessed value that the basis obtains and the training set
In include each sample image be directed to each image parameter sample value, calculate training penalty values the step of, comprising:
For each image parameter, the assessed value and the instruction of the image parameter are directed to according to obtained each sample image
Practice the sample value for concentrating each sample image for including to be directed to the image parameter, determines the penalty values for being directed to the image parameter；
According to the training weight of penalty values and preset each image parameter for each image parameter, training loss is determined
Value.
7. according to the method described in claim 6, it is characterized in that, the basis is for the penalty values of each image parameter and pre-
If each image parameter training weight, determine training penalty values the step of, comprising:
According to following formula, training penalty values s is determined:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N is figure
As the number of parameter.
8. a kind of image quality measure device, which is characterized in that described device includes:
Determining module, for determining image to be assessed；
Evaluation module, for obtaining the image to be assessed for default for the image input assessment network model to be assessed
The assessed value of each image parameter in quantity image parameter；
Wherein, the assessment network model is the model obtained according to training set training, and the training set includes: multiple sample graphs
Picture and each sample image are directed to the sample value of each image parameter in the preset quantity image parameter.
9. device according to claim 8, which is characterized in that described image parameter includes: image definition, color saturation
It is one or more in degree, default evaluation；The default evaluation includes the evaluation of image Attraction Degree and image scene evaluation.
10. device according to claim 8 or claim 9, which is characterized in that the assessment network model includes and the present count
It measures an image parameter and assesses network submodel correspondingly；
The evaluation module, is specifically used for:
The image to be assessed is inputted into each assessment network submodel in the assessment network submodel respectively, is obtained described
Assessed value of the image to be assessed for each image parameter in the preset quantity image parameter.
11. device according to claim 8 or claim 9, which is characterized in that described device further include:
Training module, for training the assessment network model；
The training module, is specifically used for:
Obtain preset neural network model and the training set；
The multiple sample image is inputted into the neural network model, obtains each sample image for each image parameter
Assessed value；
The sample of each image parameter is directed to according to each sample image for including in obtained assessed value and the training set
Value calculates training penalty values；
Determine whether the neural network model restrains according to the trained penalty values；
If it is not, then adjusting the parameter value in the neural network model, and return described by the multiple sample image input institute
Neural network model is stated, the step of each sample image is directed to the assessed value of each image parameter is obtained；
If so, current neural network model is determined as to assess network model.
12. device according to claim 11, which is characterized in that described device further includes penalty values determining module,
The penalty values determining module, is specifically used for:
The each sample image for including in the obtained assessed value and the training set is directed to the sample of each image parameter
This value inputs preset loss function, obtains training penalty values.
13. device according to claim 12, which is characterized in that the penalty values determining module is specifically used for:
For each image parameter, the assessed value and the instruction of the image parameter are directed to according to obtained each sample image
Practice the sample value for concentrating each sample image for including to be directed to the image parameter, determines the penalty values for being directed to the image parameter；
According to the training weight of penalty values and preset each image parameter for each image parameter, training loss is determined
Value.
14. device according to claim 13, which is characterized in that the penalty values determining module is specifically used for:
According to following formula, training penalty values s is determined:
Wherein, siFor the penalty values for i-th of image parameter, wiFor the training weight of preset i-th of image parameter, N is figure
As the number of parameter.
15. a kind of electronic equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein processing
Device, communication interface, memory complete mutual communication by bus,
Memory, for storing computer program；
Processor when for executing the program stored on memory, realizes method and step as claimed in claim 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811563932.1A CN109685785A (en) | 2018-12-20 | 2018-12-20 | A kind of image quality measure method, apparatus and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811563932.1A CN109685785A (en) | 2018-12-20 | 2018-12-20 | A kind of image quality measure method, apparatus and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109685785A true CN109685785A (en) | 2019-04-26 |
Family
ID=66188006
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811563932.1A Pending CN109685785A (en) | 2018-12-20 | 2018-12-20 | A kind of image quality measure method, apparatus and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685785A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211119A (en) * | 2019-06-04 | 2019-09-06 | 厦门美图之家科技有限公司 | Image quality measure method, apparatus, electronic equipment and readable storage medium storing program for executing |
| CN110378883A (en) * | 2019-07-11 | 2019-10-25 | 北京奇艺世纪科技有限公司 | Picture appraisal model generating method, image processing method, device, computer equipment and storage medium |
| CN110807476A (en) * | 2019-10-17 | 2020-02-18 | 新华三信息安全技术有限公司 | Password security level classification method and device and electronic equipment |
| CN110838106A (en) * | 2019-10-31 | 2020-02-25 | 国网河北省电力有限公司电力科学研究院 | Multi-dimensional evaluation method for image recognition software of secondary equipment of transformer substation |
| CN110956615A (en) * | 2019-11-15 | 2020-04-03 | 北京金山云网络技术有限公司 | Image quality evaluation model training method and device, electronic equipment and storage medium |
| CN110996169A (en) * | 2019-07-12 | 2020-04-10 | 北京达佳互联信息技术有限公司 | Method, device, electronic equipment and computer-readable storage medium for clipping video |
| CN111915595A (en) * | 2020-08-06 | 2020-11-10 | 北京金山云网络技术有限公司 | Image quality evaluation method, and training method and device of image quality evaluation model |
| EP3817392A1 (en) * | 2019-12-18 | 2021-05-05 | Beijing Baidu Netcom Science Technology Co., Ltd. | Video jitter detection method and apparatus |
| WO2021082819A1 (en) * | 2019-10-31 | 2021-05-06 | 北京金山云网络技术有限公司 | Image generation method and apparatus, and electronic device |
| CN112950581A (en) * | 2021-02-25 | 2021-06-11 | 北京金山云网络技术有限公司 | Quality evaluation method and device and electronic equipment |
| CN112950579A (en) * | 2021-02-26 | 2021-06-11 | 北京金山云网络技术有限公司 | Image quality evaluation method and device and electronic equipment |
| CN113011468A (en) * | 2021-02-25 | 2021-06-22 | 上海皓桦科技股份有限公司 | Image feature extraction method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133948A (en) * | 2017-05-09 | 2017-09-05 | 电子科技大学 | Image blurring and noise evaluating method based on multitask convolutional neural networks |
| CN108446651A (en) * | 2018-03-27 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | Face identification method and device |
| CN108960087A (en) * | 2018-06-20 | 2018-12-07 | 中国科学院重庆绿色智能技术研究院 | A kind of quality of human face image appraisal procedure and system based on various dimensions evaluation criteria |
| CN109002812A (en) * | 2018-08-08 | 2018-12-14 | 北京未来媒体科技股份有限公司 | A kind of method and device of intelligent recognition video cover |
-
2018
- 2018-12-20 CN CN201811563932.1A patent/CN109685785A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107133948A (en) * | 2017-05-09 | 2017-09-05 | 电子科技大学 | Image blurring and noise evaluating method based on multitask convolutional neural networks |
| CN108446651A (en) * | 2018-03-27 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | Face identification method and device |
| CN108960087A (en) * | 2018-06-20 | 2018-12-07 | 中国科学院重庆绿色智能技术研究院 | A kind of quality of human face image appraisal procedure and system based on various dimensions evaluation criteria |
| CN109002812A (en) * | 2018-08-08 | 2018-12-14 | 北京未来媒体科技股份有限公司 | A kind of method and device of intelligent recognition video cover |
Non-Patent Citations (3)
| Title |
|---|
| WEILONG HOU 等: "《Blind Image Quality Assessment via Deep Learning》", 《 IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS》 * |
| 金鑫: "《图像美学质量评价技术发展趋势》", 《科技导报》 * |
| 陈汝洪: "《影像构成基础》", 30 April 2016 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211119A (en) * | 2019-06-04 | 2019-09-06 | 厦门美图之家科技有限公司 | Image quality measure method, apparatus, electronic equipment and readable storage medium storing program for executing |
| CN110378883A (en) * | 2019-07-11 | 2019-10-25 | 北京奇艺世纪科技有限公司 | Picture appraisal model generating method, image processing method, device, computer equipment and storage medium |
| CN110996169A (en) * | 2019-07-12 | 2020-04-10 | 北京达佳互联信息技术有限公司 | Method, device, electronic equipment and computer-readable storage medium for clipping video |
| CN110807476A (en) * | 2019-10-17 | 2020-02-18 | 新华三信息安全技术有限公司 | Password security level classification method and device and electronic equipment |
| CN110807476B (en) * | 2019-10-17 | 2022-11-18 | 新华三信息安全技术有限公司 | Password security level classification method and device and electronic equipment |
| WO2021082819A1 (en) * | 2019-10-31 | 2021-05-06 | 北京金山云网络技术有限公司 | Image generation method and apparatus, and electronic device |
| CN110838106B (en) * | 2019-10-31 | 2023-04-14 | 国网河北省电力有限公司电力科学研究院 | Multi-dimensional evaluation method for image recognition software of secondary equipment of transformer substation |
| US11836898B2 (en) | 2019-10-31 | 2023-12-05 | Beijing Kingsoft Cloud Network Technology Co., Ltd. | Method and apparatus for generating image, and electronic device |
| CN110838106A (en) * | 2019-10-31 | 2020-02-25 | 国网河北省电力有限公司电力科学研究院 | Multi-dimensional evaluation method for image recognition software of secondary equipment of transformer substation |
| CN110956615A (en) * | 2019-11-15 | 2020-04-03 | 北京金山云网络技术有限公司 | Image quality evaluation model training method and device, electronic equipment and storage medium |
| CN110956615B (en) * | 2019-11-15 | 2023-04-07 | 北京金山云网络技术有限公司 | Image quality evaluation model training method and device, electronic equipment and storage medium |
| US11546577B2 (en) | 2019-12-18 | 2023-01-03 | Beijing Baidu Netcom Science Technology Co., Ltd. | Video jitter detection method and apparatus |
| EP3817392A1 (en) * | 2019-12-18 | 2021-05-05 | Beijing Baidu Netcom Science Technology Co., Ltd. | Video jitter detection method and apparatus |
| CN111915595A (en) * | 2020-08-06 | 2020-11-10 | 北京金山云网络技术有限公司 | Image quality evaluation method, and training method and device of image quality evaluation model |
| CN112950581A (en) * | 2021-02-25 | 2021-06-11 | 北京金山云网络技术有限公司 | Quality evaluation method and device and electronic equipment |
| CN113011468A (en) * | 2021-02-25 | 2021-06-22 | 上海皓桦科技股份有限公司 | Image feature extraction method and device |
| CN112950581B (en) * | 2021-02-25 | 2024-06-21 | 北京金山云网络技术有限公司 | Quality evaluation method and device and electronic equipment |
| CN112950579A (en) * | 2021-02-26 | 2021-06-11 | 北京金山云网络技术有限公司 | Image quality evaluation method and device and electronic equipment |
| CN112950579B (en) * | 2021-02-26 | 2024-05-31 | 北京金山云网络技术有限公司 | Image quality evaluation method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685785A (en) | A kind of image quality measure method, apparatus and electronic equipment | |
| CN110019896A (en) | A kind of image search method, device and electronic equipment | |
| CN109492577A (en) | A kind of gesture identification method, device and electronic equipment | |
| CN109754391A (en) | A kind of image quality evaluating method, device and electronic equipment | |
| CN109146892A (en) | A kind of image cropping method and device based on aesthetics | |
| CN110059750A (en) | House type shape recognition process, device and equipment | |
| CN109858563B (en) | Self-supervision characterization learning method and device based on transformation recognition | |
| CN110781976B (en) | Extension method of training image, training method and related device | |
| CN109522775A (en) | Face character detection method, device and electronic equipment | |
| CN109284735B (en) | Mouse feelings monitoring method, device and storage medium | |
| CN107330750A (en) | A kind of recommended products figure method and device, electronic equipment | |
| CN110363135A (en) | Determination method, eyes control method and the device of human eye closure degree | |
| CN105469376A (en) | Method and device for determining picture similarity | |
| CN109919252A (en) | The method for generating classifier using a small number of mark images | |
| CN107729901A (en) | Method for building up, device and the image processing method and system of image processing model | |
| CN110443212A (en) | Positive sample acquisition methods, device, equipment and storage medium for target detection | |
| CN106326632A (en) | Object analyzing method and object analyzing system | |
| CN110390261A (en) | Object detection method, device, computer readable storage medium and electronic equipment | |
| CN108121943A (en) | Method of discrimination and device and computing device based on picture | |
| CN110458875A (en) | Detection method, image split-joint method, related device and the equipment of abnormal point pair | |
| Bastomi et al. | Object detection and distance estimation tool for blind people using convolutional methods with stereovision | |
| CN106028088A (en) | Insertion method and device of media data | |
| CN115311452A (en) | Precise glue filling evaluation method and system based on machine vision | |
| CN110427998A (en) | Model training, object detection method and device, electronic equipment, storage medium | |
| CN109754077A (en) | Network model compression method, device and the computer equipment of deep neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190426 |