CN107025457A - A kind of image processing method and device - Google Patents
A kind of image processing method and device Download PDFInfo
- Publication number
- CN107025457A CN107025457A CN201710199165.XA CN201710199165A CN107025457A CN 107025457 A CN107025457 A CN 107025457A CN 201710199165 A CN201710199165 A CN 201710199165A CN 107025457 A CN107025457 A CN 107025457A
- Authority
- CN
- China
- Prior art keywords
- image
- probability
- color
- segmentation
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of image processing method and device；The embodiment of the present invention is after image processing requests are received, semantic segmentation model corresponding with the element type for needing to replace can be obtained according to the instruction of the request, belong to the probability of the element type according to each pixel in the model prediction image, to obtain probability figure, then, the probability figure is optimized based on condition random field, and merged image with predicted elemental material using the segmentation effect figure obtained after optimization, so as to reach the purpose by a certain element type partial replacement in image for predicted elemental material；The program can reduce flase drop and the probability of missing inspection, greatly improve the accuracy of segmentation, and improve the syncretizing effect of image.
Description
Technical field
The present invention relates to field of computer technology, and in particular to a kind of image processing method and device.
Background technology
With the popularization of intelligent mobile terminal, carry out shooting one kind that record has been increasingly becoming people's life whenever and wherever possible
Mode, with this simultaneously, image procossing are such as beautified or the processing such as special efficacy also more and more welcomed by the people to image.
In special effect processing, it is one of most commonly seen technology that element, which is replaced,.Exemplified by replacing sky element, in existing skill
In art, the information such as the color of sky and position in image is may be generally based upon, threshold decision is carried out, then, according to judged result
Sky segmentation is carried out to image, and the sky areas obtained after segmentation is replaced with into other elements, such as pyrotechnics, reinder or
Quadratic Finite Element space, etc., so that the image after processing can reach a kind of special effect.
In the research and practice process to prior art, it was found by the inventors of the present invention that because existing scheme is to figure
As carrying out during region division, it is mainly based upon the information such as color and position and carries out threshold decision, therefore, easily cause flase drop and leakage
Inspection, largely effects on the accuracy of segmentation and the syncretizing effect of image, such as produces distortion or not smooth, etc. enough.
The content of the invention
The embodiment of the present invention provides a kind of image processing method and device；The accuracy of segmentation can be improved, and is improved
Syncretizing effect.
The embodiment of the present invention provides a kind of image processing method, including：
Image processing requests are received, described image processing request indicates the member for needing image to be processed and needing to replace
Plain type；
Semantic segmentation model corresponding with the element type is obtained, the semantic segmentation model is instructed by deep neural network
White silk is formed；
According to the semantic segmentation model, the probability that the element type is belonged to each pixel in described image is carried out in advance
Survey, obtain probability figure；
The probability figure is optimized based on condition random field, segmentation effect figure is obtained；
Described image is merged with predicted elemental material according to the segmentation effect figure, image after being handled.
Accordingly, the embodiment of the present invention also provides a kind of image processing apparatus, including：
Receiving unit, for receiving image processing requests, described image processing request indicate to need image to be processed and
Need the element type replaced；
Acquiring unit, for obtaining corresponding with element type semantic segmentation model, the semantic segmentation model by
Deep neural network training is formed；
Predicting unit, for according to the semantic segmentation model, belonging to the element class to each pixel in described image
The probability of type is predicted, and obtains probability figure；
Optimize unit, for being optimized based on condition random field to the probability figure, obtain segmentation effect figure；
Integrated unit, for being merged described image with predicted elemental material according to the segmentation effect figure, is obtained
Image after processing.
The embodiment of the present invention can be obtained with needing to replace after image processing requests are received according to the instruction of the request
The corresponding semantic segmentation model of element type, the general of the element type is belonged to according to each pixel in the model prediction image
Rate, to obtain probability figure, then, is optimized based on condition random field to the probability figure, and after utilization optimization
To segmentation effect figure image is merged with predicted elemental material, so as to reach a certain element type part in image
Replace with the purpose of predicted elemental material；Due to the semantic segmentation model in the program mainly by deep neural network training and
Into, and be not to be based only on the information such as color and position when carrying out semantic segmentation to image using the model, but it is logical
Cross and the probability that each pixel belongs to the element type is predicted, for existing scheme, can greatly reduce
Flase drop and the probability of missing inspection；Further, since the program can also be carried out using condition random field to the probability figure after segmentation
Optimization, therefore, it can obtain more fine segmentation result, greatly improves the accuracy of segmentation, advantageously reduce image fault
Situation, improve image syncretizing effect.
Brief description of the drawings
Technical scheme in order to illustrate the embodiments of the present invention more clearly, makes required in being described below to embodiment
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those skilled in the art, on the premise of not paying creative work, it can also be obtained according to these accompanying drawings other attached
Figure.
Fig. 1 a are the schematic diagram of a scenario of image processing method provided in an embodiment of the present invention；
Fig. 1 b are the flow charts of image processing method provided in an embodiment of the present invention；
Fig. 2 a are another flow charts of image processing method provided in an embodiment of the present invention；
Fig. 2 b are the examples of interfaces figures of image processing requests in image processing method provided in an embodiment of the present invention；
Fig. 2 c are the exemplary plots of sky segmentation in image processing method provided in an embodiment of the present invention；
Fig. 2 d are the handling process frame diagrams of image processing method provided in an embodiment of the present invention；
Fig. 3 a are the structural representations of image processing apparatus provided in an embodiment of the present invention；
Fig. 3 b are another structural representations of image processing apparatus provided in an embodiment of the present invention；
Fig. 4 is the structural representation of server provided in an embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, the every other implementation that those skilled in the art are obtained under the premise of creative work is not made
Example, belongs to the scope of protection of the invention.
The embodiment of the present invention provides a kind of image processing method and device, wherein, the image processing apparatus can specifically collect
Into in the equipment such as server.
For example, with reference to Fig. 1 a, when user needs to handle certain image, it can be sent by terminal to server
Image processing requests, wherein, the image processing requests indicate to need image to be processed and need element type replaced etc. to believe
Breath.Server can obtain semantic segmentation model (semanteme corresponding with the element type after image processing requests are received
Parted pattern is formed by deep neural network training), then, according to the semantic segmentation model, each pixel in the image is belonged to
It is predicted in the probability of the element type, obtains a segmentation probability graph, for convenience, in embodiments of the present invention,
The segmentation probability graph is referred to as probability figure.Hereafter, server can also be initial general to this using modes such as condition random fields
Rate figure is optimized, to obtain finer segmentation result (obtaining segmentation effect figure), then, according to segmentation result by the figure
As being merged with predicted elemental material, such as, the first color part in segmentation effect figure (can be compared by blending algorithm
Such as white portion) it is combined with interchangeable element material, and by the second color part (such as black in segmentation effect figure
Part) it is combined with image, then, two basic change result is synthesized, and image is supplied to after the processing that synthesis is obtained
Terminal, etc..
It is described in detail individually below.It should be noted that, the sequence number of following examples is not as preferably suitable to embodiment
The restriction of sequence.
Embodiment one,
The present embodiment will be described from the angle of image processing apparatus, and the image processing apparatus can specifically be integrated in clothes
It is engaged in the equipment such as device.
A kind of image processing method, including：Image processing requests are received, the image processing requests indicate to need figure to be processed
As and need the element type replaced, obtain corresponding with element type semantic segmentation model, the semantic segmentation model by
Deep neural network training is formed, and according to the semantic segmentation model, belongs to the general of the element type to each pixel in the image
Rate is predicted, and obtains probability figure, and the probability figure is optimized based on condition random field, segmentation effect is obtained
Figure, is merged the image with predicted elemental material according to the segmentation effect figure, image after being handled.
As shown in Figure 1 b, the idiographic flow of the image processing method can be as follows：
101st, image processing requests are received.
For example, image processing requests that can be specifically sent with receiving terminal or other network side equipments, etc..Wherein, should
Image processing requests can indicate the information such as the element type for needing image to be processed and needing replacement.
So-called element type, refers to the classification of element, and element refers to that the fundamental of visual information can be carried, than
Such as, if the image processing requests indicate to need the element type replaced to be " sky ", show, it is necessary to all in the image
Sky portion is replaced；Again such as, if the image processing requests indicate to need the element type replaced to be " portrait ", table
It is bright, it is necessary to the owner in the image as part is replaced, by that analogy, etc..
102nd, semantic segmentation model corresponding with the element type is obtained, the semantic segmentation model is instructed by deep neural network
White silk is formed.
If for example, in a step 101, received image processing requests indicate to need the element type replaced for " my god
It is empty ", then at this point it is possible to obtain semantic segmentation model corresponding with " sky ", and if in a step 101, received image
Processing request indicates to need the element type replaced to be " portrait ", then at this point it is possible to obtain semantic segmentation corresponding with " portrait "
Model, etc..
Optionally, the semantic segmentation model can be pre-stored in the image processing apparatus or other storage devices,
When needing to use, obtained by the image processing apparatus, or, the semantic segmentation model can also be by the image processing apparatus
Voluntarily set up and form, i.e. before step " obtaining semantic segmentation model corresponding with the element type ", the image processing method
It can also include：
Set up the corresponding semantic segmentation model of the element type, such as, specifically can be as follows：
The training data for including the element type is obtained, according to the training data, using deep neural network to default
Semantic segmentation initial model be trained, obtain the corresponding semantic segmentation model of the element type.
For example, exemplified by setting up " sky " corresponding semantic segmentation model, can specifically collect certain amount (such as 8000
Etc.) the picture for including sky, it is then, initial to default semantic segmentation using deep neural network according to these pictures
Model is adjusted (fine tune), and the model finally given is " sky " corresponding semantic segmentation model.
It should be noted that, the default semantic segmentation initial model can in advance be set according to the demand of practical application
Put, such as, can using training in advance it is good for semantic segmentation model of 20 classifications of general scene, etc..
103rd, according to the semantic segmentation model, the probability that the element type is belonged to each pixel in the image is carried out in advance
Survey, obtain probability figure；For example, specifically can be as follows：
(1) image is imported into the semantic segmentation model, to predict that each pixel belongs to the element type in the image
Probability.
For example, so that the element type is " sky " as an example, then at this point it is possible to which the image is imported into " sky " corresponding semanteme
Parted pattern, to predict that each pixel in the image belongs to the probability of " sky ".
In another example, so that the element type is " portrait " as an example, then at this point it is possible to which the image is imported into " portrait " corresponding language
Adopted parted pattern, to predict that each pixel in the image belongs to the probability of " portrait ", by that analogy, etc..
(2) color of the respective pixel in default masking-out is configured according to the probability, obtains probability figure.
For example, can specifically determine whether the probability is more than predetermined threshold value, if so, then by respective pixel in default masking-out
Color be set to the first color, if it is not, color of the respective pixel in default masking-out then is set into the second color, it is determined that
Color of all pixels in default masking-out is respectively provided with the image finish after, output sets the default masking-out after color, obtains
Probability figure.
A masking-out comprising the first color and the second color can be now obtained, wherein, the first face in the masking-out
Color table show respective pixel belong to the element type probability it is larger, and the second color represents that respective pixel belongs to the element type
Probability is smaller, therefore, for convenience, in embodiments of the present invention, and the masking-out of the output is referred to as into probability figure.
Wherein, the predetermined threshold value can be configured according to the demand of practical application, such as, be specially with the predetermined threshold value
80%, and the element type be " sky " exemplified by, if some pixel A belong to " sky " probability be more than 80%, can be by picture
Colors of the plain A in default masking-out is set to the first color, otherwise, if the probability that pixel A belongs to " sky " is less than or equal to 80%,
Then color of the pixel A in default masking-out can be set to second color, etc..
Wherein, the first color and the second color can also be depending on the demands according to practical application, such as, can be by the first face
Color is set to white, and the second color is set into black, or, the first color can also be set to pink colour, and by the second face
Color is set to green, etc..For convenience, in embodiments of the present invention, incite somebody to action using the first color as white, and the second face
Color be black exemplified by illustrate.
104th, based on condition random field (CRF or CRFs, Conditional Random Fields, also referred to as condition random
Domain) the probability figure is optimized, obtain segmentation effect figure.
For example, node that specifically can be by the pixel-map in the probability figure into condition random field, determines node
Between side constraint similitude, and the similitude constrained according to side adjusts to the segmentation result of pixel in the probability figure
It is whole, obtain segmentation effect figure.
Wherein, condition random field is a kind of discriminate probabilistic model, is one kind of random field.As markov is random
, condition random field is that with undirected graph model, the node (i.e. summit) in graph model represents the company between stochastic variable, node
Line represents the dependence relation between stochastic variable.Condition random field has the ability of expression long-distance dependence and overlapping property feature,
The advantage for the problems such as can preferably solving to mark (classification) biasing, and all features can carry out global normalization, can
The optimal solution of the overall situation is tried to achieve, therefore, it can optimize the probability figure using condition random field, to reach Optimized Segmentation result
Purpose.
It should be noted that, because segmentation effect figure is obtained by the optimization of probability figure, therefore, the segmentation effect figure is same
Sample is also a masking-out comprising the first color and the second color.
105th, the image is merged with predicted elemental material according to the segmentation effect figure, image after being handled；Example
Such as, specifically can be as follows：
(1) interchangeable element material is obtained according to preset strategy.
Wherein, the preset strategy can be configured according to the demand of practical application, such as, can receive user's triggering
Selection of materials is instructed, then, is obtained corresponding material from material database according to selection of materials instruction, is used as interchangeable element element
Material, etc..
Optionally, in order to increase the diversity of the element material, this yuan can also can be obtained by the way of intercepting at random
Material, i.e. step " obtaining interchangeable element material according to preset strategy " can also include：
Candidate image is obtained, the candidate image is intercepted at random, and regard the image being truncated to as interchangeable member
Material, etc..
Wherein, the candidate image can by being obtained on network, or, can also be uploaded by user, very
Extremely, the image processing apparatus, etc. can also be provided to, herein by user's directly sectional drawing on terminal screen or webpage
Repeat no more.
(2) the first color part in the segmentation effect figure and the element material that gets mutually are tied by blending algorithm
Close, obtain first and combine figure.
, therefore, now, can because the probability that the pixel of the first color part belongs to the element type of needs replacement is higher
With by blending algorithm, the part and the element material that gets are combined, you can so that the pixel of the part to be replaced
For the element material got.
(3) the second color part in the segmentation effect figure is combined with the image by blending algorithm, obtains second
With reference to figure.
, therefore, now, can because the probability that the pixel of the second color part belongs to the element type of needs replacement is relatively low
So that by blending algorithm, the part and original image to be combined, that is, retain the pixel of the part.
It should be noted that, optionally, in order to improve syncretizing effect, or other special effects are realized, by the second face
Color part can also necessarily be pre-processed to the image before the image is combined, such as carry out color transformed, contrast
Adjustment, brightness adjustment, saturation degree adjust, and/or add other special efficacy masking-outs etc., then, then by blending algorithm, by the second face
Color part is combined with the pretreated image, and figure is combined to obtain second.
(4) combine figure and second by first to be synthesized with reference to figure, image after being handled.
So, the element for needing to be replaced in image just can be replaced with to the element material, such as by image
" sky " replaces with " space ", etc., will not be repeated here.
Optionally, in order that fusion results are more true, it is to avoid noise caused by probabilistic forecasting is inaccurate or
Missing, the segmentation effect figure can also necessarily be handled before fusion, with cause its partitioning boundary it is more smooth, with
And the color transition of the junction in replacement region can be more natural；I.e. step " according to the segmentation effect figure by the image with
Predicted elemental material is merged, image after being handled " before, the image processing method can also include：
The segmentation effect figure is carried out at display model (Appearance Model) algorithm and/or morphological image operation
Reason, segmentation effect figure after being handled.
Then now, the image " is merged, obtained after processing by step according to the segmentation effect figure with predicted elemental material
Image " can include：According to segmentation effect figure after processing, the image is merged with predicted elemental material, such as carried out saturating
Lightness (Alpha) is merged, image after being handled.
Wherein, display model algorithm is a kind of Feature Points Extraction for being widely used in area of pattern recognition, and it can be with
Statistical modeling is carried out to texture, and two statistical models of shape and texture are further fused to apparent model.And image aspects
The processing such as noise reduction process and/or connected domain analysis can be included by learning operation processing, pass through display model algorithm or morphological image
Operation etc. processing after segmentation effect figure, its partitioning boundary can it is more smooth and replace region junction color transition
Can be more natural.
It should be noted that, " Alpha fusions " described in the embodiment of the present invention refers to being melted based on Alpha values
Close, wherein, Alpha is primarily used to the transparency level of specified pixel.General, it can be protected for the alpha parts of each pixel
8 are stayed, alpha virtual value is in the range of [0,255], and [0,255] represents opacity [0%, 100%].Therefore, pixel
When alpha is 0, represent fully transparent, when the alpha of pixel is 128, represent that 50% is transparent, when the alpha of pixel is 255, table
Show completely opaque.
From the foregoing, it will be observed that the present embodiment is after image processing requests are received, instruction acquisition that can be according to the request and need
The corresponding semantic segmentation model of the element type to be replaced, belongs to the element type according to each pixel in the model prediction image
Probability, to obtain probability figure, then, the probability figure is optimized based on condition random field, and using optimization
The segmentation effect figure obtained afterwards is merged image with predicted elemental material, so as to reach a certain element type in image
Partial replacement is the purpose of predicted elemental material；Because the semantic segmentation model in the program is mainly instructed by deep neural network
Practice, and be not to be based only on the information such as color and position when carrying out semantic segmentation to image using the model, and
It is to be predicted by belonging to the probability of the element type to each pixel, can be significantly for existing scheme
Reduce flase drop and the probability of missing inspection；Further, since the program can also be using condition random field to the probability figure after segmentation
Optimize, therefore, it can obtain more fine segmentation result, greatly improve the accuracy of segmentation, advantageously reduce image
The situation of distortion, improves the syncretizing effect of image.
Embodiment two,
Citing, is described in further detail by the method according to described by embodiment one below.
In the present embodiment, server will be specifically integrated in the image processing apparatus, and the element that the needs are replaced is
Illustrated exemplified by " sky ".
As shown in Fig. 2 a and 2d, a kind of image processing method, idiographic flow can be as follows：
201st, terminal to server sends image processing requests, wherein, the image processing requests can indicate to need processing
The information such as element type replaced of image and needing.
Wherein, the triggering mode of the image processing requests can have a variety of, such as, can by click on or slide webpage or
Triggering key on client end interface is triggered, or, can also carry out triggering by inputting preset instructions, etc..
For example, exemplified by being triggered by clicking trigger key, referring to Fig. 2 b, when user is needed the sky portion in picture A
Divide and replace with other elements, such as, can be by uploading pictures A when replacing with " space " element or increase " cloud ", and click on
Triggering key " playing once " sends the image processing requests to trigger generation image processing requests to server, wherein, the image
Processing request indicates to need image to be processed to be image A, and needs the element type replaced to be " sky ".
It should be noted that, in the present embodiment, it will be illustrated so that the element for needing to replace is " sky " as an example, should
Understand, the element type that the needs are replaced can also be other types, such as " portrait ", " eyes " or " plant ", etc., its
Realize similar, will not be repeated here.
202nd, server is received after the image processing requests, obtains semantic segmentation model corresponding with " sky ", the language
Adopted parted pattern is formed by deep neural network training.
Optionally, the semantic segmentation model can be pre-stored in the image processing apparatus or other storage devices,
When needing to use, obtained by the image processing apparatus, or, the semantic segmentation model can also be by the image processing apparatus
Voluntarily set up and form, for example, the training data for including the element type can be obtained, such as, collect a number of include
The picture of sky, then, according to the training data (i.e. the picture comprising sky), using deep neural network to default semanteme
Segmentation initial model is trained, and is somebody's turn to do " sky " corresponding semantic segmentation model.
It should be noted that, the default semantic segmentation initial model can in advance be set according to the demand of practical application
Put, such as, can using training in advance it is good for semantic segmentation model of 20 classifications of general scene, etc..
203rd, the image is imported the semantic segmentation model by server, with predict in the image each pixel belong to this " my god
The probability of sky ".
If for example, in step 202., in received image processing requests, indicating to need image to be processed to be picture
A, then at this point it is possible to which picture A is imported into the way of Three Channel Color image in " sky " corresponding semantic segmentation model,
To predict that each pixel in image A belongs to the probability of " sky ", then, step 204 is performed.
204th, server is configured according to the probability to color of the respective pixel in default masking-out, obtains probability
Figure.
For example, can specifically determine whether the probability is more than predetermined threshold value, if so, then by respective pixel in default masking-out
Color be set to the first color, if it is not, color of the respective pixel in default masking-out then is set into the second color, it is determined that
Color of all pixels in default masking-out is respectively provided with the image finish after, output sets the default masking-out after color, obtains
Probability figure.
Wherein, the predetermined threshold value can be configured according to the demand of practical application, such as, be specially with the predetermined threshold value
Exemplified by 80%, if the probability that some pixel K belongs to " sky " is more than 80%, pixel K can preset the face in masking-out
Color is set to the first color, otherwise, can be by pixel K if the probability that some pixel K belongs to " sky " is less than or equal to 80%
Color in default masking-out is set to second color, etc..
Wherein, the first color and the second color can also be depending on the demands according to practical application, such as, can be by the first face
Color is set to white, and the second color is set into black, or, the first color can also be set to pink colour, and by the second face
Color is set to green, etc..
Such as, white is set to the first color, the second color is set to exemplified by black, then picture A is imported into the language
After adopted parted pattern, probability figure as shown in Figure 2 c can be obtained.
205th, server is optimized based on condition random field to the probability figure, obtains segmentation effect figure.
For example, server can be by the pixel-map in the probability figure into condition random field node, it is determined that section
The similitude of side constraint between point, and segmentation result progress of the similitude constrained according to side to pixel in the probability figure
Adjustment, obtains segmentation effect figure.
Because condition random field is a kind of undirected graph model, it therefore, it can each pixel in image corresponding to bar
A node in part random field, and it is default include the prior information of the parameters such as color, texture and position, so, node it
Between side constrain similar pixel just there is similar segmentation result, so, the similitude that can be constrained according to side is initial to this
The segmentation result of pixel is adjusted in probability graph so that sky segmentation result is more fine, for example, Fig. 2 c are participated in, based on bar
After part random field is optimized to the probability figure, segmentation result more fine segmentation effect figure can be obtained.
206th, server carries out display model algorithm to the segmentation effect figure and/or morphological image operation is handled, and obtains
Segmentation effect figure after processing, then, performs step 207.
Wherein, morphological image operation processing can include the processing such as noise reduction process and/or connected domain analysis.Pass through outward appearance
Segmentation effect figure after the processing such as model algorithm or morphological image operation, its partitioning boundary can more smooth and replacement area
The color transition of the junction in domain can be more natural.
It should be noted that, step 206 is optional step,, can after step 205 is finished if not performing step 206
Directly to perform step 207, and in a step 208, by blending algorithm by the segmentation effect figure, image and element material
Merged, image after being handled.
207th, server obtains interchangeable element material according to preset strategy.
Wherein, the preset strategy can be configured according to the demand of practical application, such as, can receive user's triggering
Selection of materials is instructed, then, is obtained corresponding material from material database according to selection of materials instruction, is used as interchangeable element element
Material, etc..
Optionally, in order to increase the diversity of the element material, this yuan can also can be obtained by the way of intercepting at random
Material, such as, server can obtain candidate image, then, the candidate image is intercepted at random, and will be truncated to
Image is used as interchangeable element material, etc..
Wherein, the candidate image can by being obtained on network, or, can also be uploaded by user, very
Extremely, the image processing apparatus, etc. can also be provided to, herein by user's directly sectional drawing on terminal screen or webpage
Repeat no more.
208th, server by blending algorithm by this handle after segmentation effect figure, image and element material merged,
Image after being handled.
For example, still using the first color as white, the second color is illustrates exemplified by black, then now, server can
To be combined the white portion in the segmentation effect figure with the element material got by blending algorithm, the first combination is obtained
Figure, and, the black portions in the segmentation effect figure are combined with image A by blending algorithm, second is obtained and combines figure,
Then, combine figure and second by first to be synthesized with reference to figure, image after being handled.
Because the probability that the pixel of white portion belongs to " sky " is higher, therefore, at this point it is possible to by blending algorithm,
The pixel of the part replaces with to the element material got, and due to the pixel of black portions belong to the probability of " sky " compared with
It is low, therefore, at this point it is possible to by blending algorithm, the pixel of the part is combined with original image A, that is, retains the part
Pixel, so, first is combined after scheming to be synthesized with reference to figure with second, just can replace with " sky " in original image A
Corresponding element material, such as replace with " night sky on Christmas ", etc. by " sky " in image A, referring to Fig. 2 d, herein no longer
Repeat.
It should be noted that, optionally, as shown in Figure 2 d, in order to improve syncretizing effect, or the other special effects of realization,
Black portions (i.e. the second color part) before image A is combined, can also be subjected to certain pre- place to image A
Reason, such as carry out it is color transformed, setting contrast, brightness adjustment, saturation degree adjustment, and/or add other special efficacy masking-outs etc., so
Afterwards, then by blending algorithm, black portions are combined with the pretreated image A, figure are combined to obtain second, herein
Repeat no more.
209th, image after processing is sent to terminal by server.
Such as, image after the processing can be shown on the interface of relative client.Optionally, the server can also be carried
Approach and/or sharing interface are preserved for corresponding, so that user is protected and/or is shared, such as, is schemed after can this be handled
As preserving beyond the clouds or locally (i.e. in terminal), and by this handle after images share to microblogging, circle of friends, and/or be inserted into
In the chat conversations interface of instant messenger, etc., it will not be repeated here.
From the foregoing, it will be observed that the present embodiment is after image processing requests are received, can be obtained according to the instruction of the request with " my god
It is empty " corresponding semantic segmentation model, the probability of " sky " is belonged to according to each pixel in the model prediction image, it is initial to obtain
Probability graph, then, is optimized, and utilize the segmentation effect figure obtained after optimization based on condition random field to the probability figure
Image is merged with predicted elemental material, so as to reach that by " sky " partial replacement in image be predicted elemental material
Purpose；Because the semantic segmentation model in the program is mainly what is trained by deep neural network, and utilizing the mould
It is not to be based only on the information such as color and position when type carries out semantic segmentation to image, but by belonging to this to each pixel
The probability of element type is predicted, for existing scheme, can greatly reduce flase drop and the probability of missing inspection；
Further, since the program can also optimize the probability figure after segmentation using condition random field, it therefore, it can obtain
More fine segmentation result, greatly improves the accuracy of segmentation, advantageously reduces the situation of image fault, improves melting for image
Close effect.
Embodiment three,
In order to preferably implement above method, the embodiment of the present invention also provides a kind of image processing apparatus, the image procossing
Device can be specifically integrated in the equipment such as server.
As shown in Figure 3 a, the image processing apparatus includes receiving unit 301, acquiring unit 302, predicting unit 303, optimization
Unit 304 and integrated unit 305, it is as follows：
(1) receiving unit 301；
Receiving unit 301, for receiving image processing requests, the image processing requests indicate to need image to be processed, with
And need the information such as the element type of replacement.
(2) acquiring unit 302；
Acquiring unit 302, for obtaining semantic segmentation model corresponding with the element type, the semantic segmentation model is by depth
Degree neural metwork training is formed.
If for example, image processing requests received by receiving unit 301 indicate to need the element type replaced for " my god
It is empty ", then now, acquiring unit 302 can obtain semantic segmentation model corresponding with " sky ", and if receiving unit 301 is connect
The image processing requests that receive indicate to need the element type replaced to be " portrait ", then now, acquiring unit 302 can obtain with
" portrait " corresponding semantic segmentation model, etc., is no longer enumerated herein.
Optionally, the semantic segmentation model can be pre-stored in the image processing apparatus or other storage devices,
When needing to use, obtained by the image processing apparatus, or, the semantic segmentation model can also be by the image processing apparatus
Voluntarily set up and form, i.e., as shown in Figure 3 b, the image processing apparatus can also set up unit 306 including model, as follows：
The model sets up unit 306, can be used for setting up the corresponding semantic segmentation model of the element type, such as, specifically
Can be as follows：
The training data for including the element type is obtained, according to the training data, using deep neural network to default
Semantic segmentation initial model be trained, obtain the corresponding semantic segmentation model of the element type.
Wherein, the default semantic segmentation initial model can be in advance configured according to the demand of practical application, such as,
Can using training in advance it is good for semantic segmentation model of 20 classifications of general scene, etc..
(3) predicting unit 303；
Predicting unit 303, for according to the semantic segmentation model, belonging to the element type to each pixel in the image
Probability is predicted, and obtains probability figure.
For example, the predicting unit 303 can include prediction subelement and set subelement, it is as follows：
Subelement is predicted, can be used for the image importing the semantic segmentation model, to predict each pixel in the image
Belong to the probability of the element type.
For example, so that the element type is " sky " as an example, then now, the image can be imported " sky " by prediction subelement
Corresponding semantic segmentation model, to predict that each pixel in the image belongs to the probability of " sky ".
Subelement is set, can be used for being configured color of the respective pixel in default masking-out according to the probability, obtain
To probability figure.
Such as, the setting subelement, is specifically determined for whether the probability is more than predetermined threshold value, if so, then by phase
Color of the pixel in default masking-out is answered to be set to the first color；If it is not, then color of the respective pixel in default masking-out is set
It is set to the second color；After it is determined that color of all pixels in default masking-out is respectively provided with and finished in the image, output sets face
Default masking-out after color, obtains probability figure.
Wherein, the predetermined threshold value can be configured according to the demand of practical application, and the first color and the second color
Such as, the first color can be set to white, the second color is set to black depending on demand according to practical application
Color, etc..
(4) unit 304 is optimized；
Optimize unit 304, for being optimized based on condition random field to the probability figure, obtain segmentation effect figure.
For example, the optimization unit 304, specifically can be used for the pixel-map in the probability figure to condition random field
In node, determine between node side constraint similitude, the similitude constrained according to side is to pixel in the probability figure
Segmentation result be adjusted, obtain segmentation effect figure.
(5) integrated unit 305；
Integrated unit 305, for being merged the image with predicted elemental material according to the segmentation effect figure, is obtained everywhere
Image after reason.
For example, the integrated unit 305 can include material obtaining subelement, the first fusion subelement, the second fusant list
Member and synthesis subelement, it is as follows：
The material obtaining subelement, for obtaining interchangeable element material according to preset strategy.
Wherein, the preset strategy can be configured according to the demand of practical application, such as, the material obtaining subelement,
The specific selection of materials instruction that can be used for receiving user's triggering, corresponding element is obtained according to selection of materials instruction from material database
Material, is used as interchangeable element material, etc..
Optionally, in order to increase the diversity of the element material, this yuan can also can be obtained by the way of intercepting at random
Material, i.e.,：
The material obtaining subelement, specifically for obtaining candidate image, is intercepted at random to the candidate image, and will be cut
The image got is used as interchangeable element material.
Wherein, the candidate image can by being obtained on network, or, can also be uploaded by user, very
Extremely, the image processing apparatus, etc. can also be provided to, herein by user's directly sectional drawing on terminal screen or webpage
Repeat no more.
This first fusion subelement, can be used for by blending algorithm by the first color part in the segmentation effect figure with
The element material got is combined, and is obtained first and is combined figure.
This second fusion subelement, can be used for by blending algorithm by the second color part in the segmentation effect figure with
The image is combined, and is obtained second and is combined figure.
The synthesis subelement, can be used for combining first and schemes to be synthesized with reference to figure with second, image after being handled.
Optionally, in order that fusion results are more true, it is to avoid noise caused by probabilistic forecasting is inaccurate or
Missing, the segmentation effect figure can also necessarily be handled before fusion, with cause its partitioning boundary it is more smooth, with
And the color transition of the junction in replacement region can be more natural；I.e. as shown in Figure 3 b, the image processing apparatus can also be wrapped
Pretreatment unit 307 is included, it is as follows：
The pretreatment unit 307, can be used for carrying out display model algorithm and/or morphological image to the segmentation effect figure
Operation is handled, segmentation effect figure after being handled.
Then now, integrated unit 305, specifically can be used for according to segmentation effect figure after processing, by the image and default member
Material is merged, image after being handled.
Wherein, morphological image operation processing can include the processing such as noise reduction process and/or connected domain analysis, herein no longer
Repeat.
It when it is implemented, above unit can be realized as independent entity, can also be combined, be made
Realized for same or several entities, the specific implementation of above unit can be found in embodiment of the method above, herein not
Repeat again.
From the foregoing, it will be observed that the present embodiment is after image processing requests are received, can be by acquiring unit 302 according to the request
Indicate to obtain semantic segmentation model corresponding with the element type that needs are replaced, and by predicting unit 303 according to the model prediction
Each pixel belongs to the probability of the element type in image, to obtain probability figure, then, and bar is based on by optimization unit 304
Part random field is optimized to the probability figure, and will be schemed using the segmentation effect figure obtained after optimization by integrated unit 305
As being merged with predicted elemental material, so as to reach that by a certain element type partial replacement in image be predicted elemental material
Purpose；Because the semantic segmentation model in the program is mainly what is trained by deep neural network, and should utilizing
It is not to be based only on the information such as color and position when model carries out semantic segmentation to image, but by belonging to each pixel
The probability of the element type is predicted, for existing scheme, can greatly reduce the general of flase drop and missing inspection
Rate；Further, since the program can also be optimized using condition random field to the probability figure after segmentation, it therefore, it can
More fine segmentation result is obtained, the accuracy of segmentation is greatly improved, the situation of image fault is advantageously reduced, improves image
Syncretizing effect.
Example IV,
The embodiment of the present invention also provides a kind of server, as shown in figure 4, it illustrates the clothes involved by the embodiment of the present invention
The structural representation of business device, specifically：
The server can include one or processor 401, one or more meters of more than one processing core
The parts such as memory 402, power supply 403 and the input block 404 of calculation machine readable storage medium storing program for executing.Those skilled in the art can manage
The server architecture shown in solution, Fig. 4 does not constitute the restriction to server, can include than illustrating more or less portions
Part, either combines some parts or different parts arrangement.Wherein：
Processor 401 is the control centre of the server, utilizes each of various interfaces and the whole server of connection
Part, by operation or performs and is stored in software program and/or module in memory 402, and calls and be stored in memory
Data in 402, the various functions and processing data of execute server, so as to carry out integral monitoring to server.Optionally, locate
Reason device 401 may include one or more processing cores；It is preferred that, processor 401 can integrated application processor and modulatedemodulate mediate
Device is managed, wherein, application processor mainly handles operating system, user interface and application program etc., and modem processor is main
Handle radio communication.It is understood that above-mentioned modem processor can not also be integrated into processor 401.
Memory 402 can be used for storage software program and module, and processor 401 is stored in memory 402 by operation
Software program and module, so as to perform various function application and data processing.Memory 402 can mainly include storage journey
Sequence area and storage data field, wherein, the application program (ratio that storing program area can be needed for storage program area, at least one function
Such as sound-playing function, image player function) etc.；Storage data field can be stored uses created data according to server
Deng.In addition, memory 402 can include high-speed random access memory, nonvolatile memory can also be included, for example, at least
One disk memory, flush memory device or other volatile solid-state parts.Correspondingly, memory 402 can also include
Memory Controller, to provide access of the processor 401 to memory 402.
Server also includes the power supply 403 powered to all parts, it is preferred that power supply 403 can pass through power management system
System is logically contiguous with processor 401, so as to realize the work(such as management charging, electric discharge and power managed by power-supply management system
Energy.Power supply 403 can also include one or more direct current or AC power, recharging system, power failure monitor electricity
The random component such as road, power supply changeover device or inverter, power supply status indicator.
The server may also include input block 404, and the input block 404 can be used for the numeral for receiving input or character letter
Breath, and generation is set with user and function control is relevant keyboard, mouse, action bars, optics or trace ball signal are defeated
Enter.
Although not shown, server can also will not be repeated here including display unit etc..Specifically in the present embodiment,
Processor 401 in server can according to following instruction, by the process of one or more application program is corresponding can
Perform file to be loaded into memory 402, and the application program being stored in memory 402 is run by processor 401, so that
Various functions are realized, it is as follows：
Image processing requests are received, the image processing requests indicate the element for needing image to be processed and needing to replace
Type, obtains semantic segmentation model corresponding with the element type, and the semantic segmentation model is formed by deep neural network training,
According to the semantic segmentation model, the probability that each pixel in the image belongs to the element type is predicted, obtains initial general
Rate figure, is optimized to the probability figure based on condition random field, obtains segmentation effect figure, should according to the segmentation effect figure
Image is merged with predicted elemental material, image after being handled.
For example, specifically interchangeable element material can be obtained according to preset strategy, then, by blending algorithm by this point
Element material of first color part with getting cut in design sketch is combined, and is obtained first and is combined figure, and passes through fusion
The second color part in the segmentation effect figure is combined by algorithm with the image, is obtained second and is combined figure, subsequently, by first
Synthesized with reference to figure and second with reference to figure, image after being handled.
Optionally, the semantic segmentation model can be pre-stored in the image processing apparatus or other storage devices,
When needing to use, obtained by the image processing apparatus, or, the semantic segmentation model can also be by the image processing apparatus
Voluntarily set up and form, i.e., processor 401 can also run the application program being stored in memory 402, so as to realize following work(
Energy：
The training data for including the element type is obtained, according to the training data, using deep neural network to default
Semantic segmentation initial model be trained, obtain the corresponding semantic segmentation model of the element type.
Wherein, the default semantic segmentation initial model can be in advance configured according to the demand of practical application, such as,
Can using training in advance it is good for semantic segmentation model of 20 classifications of general scene, etc..
Optionally, in order that fusion results are more true, it is to avoid noise caused by probabilistic forecasting is inaccurate or
Missing, the segmentation effect figure can also necessarily be handled before fusion, with cause its partitioning boundary it is more smooth, with
And the color transition of the junction in replacement region can be more natural；I.e. processor 401, which can also be run, is stored in memory 402
In application program, so as to implement function such as：
Display model algorithm and/or morphological image operation processing are carried out to the segmentation effect figure, is split after being handled
Design sketch, so, just can be according to segmentation effect figure after the processing, by the image and predicted elemental material subsequently in fusion
Merged, image after being handled refers to embodiment above, will not be repeated here.
The specific implementation of each operation can be found in embodiment above above, will not be repeated here.
, can be according to the instruction of the request from the foregoing, it will be observed that the server of the present embodiment is after image processing requests are received
Semantic segmentation model corresponding with the element type for needing to replace is obtained, this is belonged to according to each pixel in the model prediction image
The probability of element type, to obtain probability figure, then, is optimized based on condition random field to the probability figure, and
Image is merged with predicted elemental material using the segmentation effect figure obtained after optimization, so that reach will be a certain in image
Element type partial replacement is the purpose of predicted elemental material；Because the semantic segmentation model in the program is mainly by depth god
Through network training, and it is not to be based only on color and position when carrying out semantic segmentation to image using the model
Etc. information, but it is predicted by the probability for belonging to the element type to each pixel, accordingly, with respect to existing scheme
Speech, can greatly reduce flase drop and the probability of missing inspection；Further, since after the program can also utilize condition random field to segmentation
Probability figure is optimized, and therefore, it can obtain more fine segmentation result, greatly improves the accuracy of segmentation, favorably
In the situation for reducing image fault, improve the syncretizing effect of image.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can
To instruct the hardware of correlation to complete by program, the program can be stored in a computer-readable recording medium, storage
Medium can include：Read-only storage (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
A kind of image processing method and device provided above the embodiment of the present invention is described in detail, herein
Apply specific case to be set forth the principle and embodiment of the present invention, the explanation of above example is only intended to help
Understand the method and its core concept of the present invention；Simultaneously for those skilled in the art, according to the thought of the present invention, in tool
It will change in body embodiment and application, in summary, this specification content should not be construed as to the present invention
Limitation.
Claims (16)
1. a kind of image processing method, it is characterised in that including：
Image processing requests are received, described image processing request indicates the element class for needing image to be processed and needing to replace
Type；
Obtain corresponding with element type semantic segmentation model, the semantic segmentation model is by deep neural network training
Into；
According to the semantic segmentation model, the probability that each pixel in described image belongs to the element type is predicted,
Obtain probability figure；
The probability figure is optimized based on condition random field, segmentation effect figure is obtained；
Described image is merged with predicted elemental material according to the segmentation effect figure, image after being handled.
2. according to the method described in claim 1, it is characterised in that described according to the semantic segmentation model, to described image
In each pixel belong to the probability of the element type and be predicted, obtain probability figure, including：
Described image is imported into the semantic segmentation model, to predict that each pixel belongs to the element type in described image
Probability；
Color of the respective pixel in default masking-out is configured according to the probability, probability figure is obtained.
3. method according to claim 2, it is characterised in that it is described according to the probability to respective pixel in default masking-out
On color be configured, obtain probability figure, including：
Determine whether the probability is more than predetermined threshold value；
If so, color of the respective pixel in default masking-out then is set into the first color；
If it is not, color of the respective pixel in default masking-out then is set into the second color；
After it is determined that color of all pixels in default masking-out is respectively provided with and finished in described image, output sets pre- after color
If masking-out, probability figure is obtained.
4. according to the method described in claim 1, it is characterised in that described that the probability figure is entered based on condition random field
Row optimization, obtains segmentation effect figure, including：
By node of the pixel-map in the probability figure into condition random field；
Determine the similitude of the side constraint between node；
The similitude constrained according to side is adjusted to the segmentation result of pixel in the probability figure, obtains segmentation effect
Figure.
5. the method according to any one of Claims 1-4, it is characterised in that it is described according to the segmentation effect figure by institute
State image to be merged with predicted elemental material, image after being handled, including：
Interchangeable element material is obtained according to preset strategy；
The first color part in the segmentation effect figure is combined with the element material got by blending algorithm, obtained
First combines figure；
The second color part in the segmentation effect figure is combined with described image by blending algorithm, the second combination is obtained
Figure；
Figure and second is combined by first to be synthesized with reference to figure, image after being handled.
6. method according to claim 5, it is characterised in that described to obtain interchangeable element element according to preset strategy
Material, including：
Candidate image is obtained, the candidate image is intercepted at random, and regard the image being truncated to as interchangeable element
Material；Or,
The selection of materials instruction of user's triggering is received, corresponding material is obtained from material database according to selection of materials instruction, made
For interchangeable element material.
7. the method according to any one of Claims 1-4, it is characterised in that it is described according to the segmentation effect figure by institute
Image is stated to be merged with predicted elemental material, after being handled before image, in addition to：
Display model algorithm and/or morphological image operation processing are carried out to the segmentation effect figure, effect is split after being handled
Fruit is schemed；
It is described to be merged described image with predicted elemental material according to the segmentation effect figure, image after being handled, bag
Include：According to segmentation effect figure after processing, described image is merged with predicted elemental material, image after being handled.
8. the method according to any one of Claims 1-4, it is characterised in that the acquisition is corresponding with the element type
Semantic segmentation model before, in addition to：
Obtain the training data for including the element type；
According to the training data, default semantic segmentation initial model is trained using deep neural network, institute is obtained
State the corresponding semantic segmentation model of element type.
9. a kind of image processing apparatus, it is characterised in that including：
Receiving unit, for receiving image processing requests, described image processing request is indicated to need image to be processed and needed
The element type of replacement；
Acquiring unit, for obtaining semantic segmentation model corresponding with the element type, the semantic segmentation model is by depth
Neural metwork training is formed；
Predicting unit, for according to the semantic segmentation model, belonging to the element type to each pixel in described image
Probability is predicted, and obtains probability figure；
Optimize unit, for being optimized based on condition random field to the probability figure, obtain segmentation effect figure；
Integrated unit, for being merged described image with predicted elemental material according to the segmentation effect figure, is handled
Image afterwards.
10. device according to claim 9, it is characterised in that the predicting unit includes prediction subelement and sets son
Unit；
The prediction subelement, for described image to be imported into the semantic segmentation model, to predict each picture in described image
Element belongs to the probability of the element type；
The setting subelement, for being configured according to the probability to color of the respective pixel in default masking-out, is obtained
Probability figure.
11. device according to claim 10, it is characterised in that the setting subelement, specifically for：
Determine whether the probability is more than predetermined threshold value；
If so, color of the respective pixel in default masking-out then is set into the first color；
If it is not, color of the respective pixel in default masking-out then is set into the second color；
After it is determined that color of all pixels in default masking-out is respectively provided with and finished in described image, output sets pre- after color
If masking-out, probability figure is obtained.
12. device according to claim 9, it is characterised in that
The optimization unit, specifically for the node by the pixel-map in the probability figure into condition random field, really
Determine the similitude of the side constraint between node, segmentation result of the similitude constrained according to side to pixel in the probability figure
It is adjusted, obtains segmentation effect figure.
13. the device according to any one of claim 9 to 12, it is characterised in that the integrated unit includes material obtaining
Subelement, the first fusion subelement, the second fusion subelement and synthesis subelement；
The material obtaining subelement, for obtaining interchangeable element material according to preset strategy；
The first fusion subelement, for by the way that blending algorithm is by the first color part in the segmentation effect figure and obtains
To element material be combined, obtain first combine figure；
It is described second fusion subelement, for by blending algorithm by the second color part in the segmentation effect figure with it is described
Image is combined, and is obtained second and is combined figure；
The synthesis subelement, is synthesized, image after being handled for combining figure and second by first with reference to figure.
14. device according to claim 13, it is characterised in that
The material obtaining subelement, specifically for obtaining candidate image, is intercepted at random to the candidate image, and will be cut
The image got is used as interchangeable element material；Or,
The material obtaining subelement, instructs specifically for the selection of materials for receiving user's triggering, is referred to according to the selection of materials
Order obtains corresponding material from material database, is used as interchangeable element material.
15. the device according to any one of claim 9 to 12, it is characterised in that also including pretreatment unit；
The pretreatment unit, for being carried out to the segmentation effect figure at display model algorithm and/or morphological image operation
Reason, segmentation effect figure after being handled；
The integrated unit, specifically for according to segmentation effect figure after processing, described image is melted with predicted elemental material
Close, image after being handled.
16. the device according to any one of claim 9 to 12, it is characterised in that also set up unit including model；
The model sets up unit, and the training data of the element type is included for obtaining, according to the training data, profit
Default semantic segmentation initial model is trained with deep neural network, the corresponding semantic segmentation of the element type is obtained
Model.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710199165.XA CN107025457B (en) | 2017-03-29 | 2017-03-29 | Image processing method and device |
| PCT/CN2018/080446 WO2018177237A1 (en) | 2017-03-29 | 2018-03-26 | Image processing method and device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710199165.XA CN107025457B (en) | 2017-03-29 | 2017-03-29 | Image processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107025457A true CN107025457A (en) | 2017-08-08 |
| CN107025457B CN107025457B (en) | 2022-03-08 |
Family
ID=59525827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710199165.XA Active CN107025457B (en) | 2017-03-29 | 2017-03-29 | Image processing method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN107025457B (en) |
| WO (1) | WO2018177237A1 (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507201A (en) * | 2017-09-22 | 2017-12-22 | 深圳天琴医疗科技有限公司 | A kind of medical image cutting method and device |
| CN107705334A (en) * | 2017-08-25 | 2018-02-16 | 北京图森未来科技有限公司 | A kind of video camera method for detecting abnormality and device |
| CN107993191A (en) * | 2017-11-30 | 2018-05-04 | 腾讯科技（深圳）有限公司 | A kind of image processing method and device |
| CN108305260A (en) * | 2018-03-02 | 2018-07-20 | 苏州大学 | Detection method, device and the equipment of angle point in a kind of image |
| WO2018177237A1 (en) * | 2017-03-29 | 2018-10-04 | 腾讯科技（深圳）有限公司 | Image processing method and device, and storage medium |
| CN108764143A (en) * | 2018-05-29 | 2018-11-06 | 北京字节跳动网络技术有限公司 | Image processing method, device, computer equipment and storage medium |
| CN109741347A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | A kind of image partition method of the iterative learning based on convolutional neural networks |
| WO2019109524A1 (en) * | 2017-12-07 | 2019-06-13 | 平安科技（深圳）有限公司 | Foreign object detection method, application server, and computer readable storage medium |
| CN110163862A (en) * | 2018-10-22 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Image, semantic dividing method, device and computer equipment |
| CN110310222A (en) * | 2019-06-20 | 2019-10-08 | 北京奇艺世纪科技有限公司 | A kind of image Style Transfer method, apparatus, electronic equipment and storage medium |
| CN110544218A (en) * | 2019-09-03 | 2019-12-06 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN110610495A (en) * | 2018-06-15 | 2019-12-24 | 北京京东尚科信息技术有限公司 | Image processing method and system and electronic equipment |
| CN110910334A (en) * | 2018-09-15 | 2020-03-24 | 北京市商汤科技开发有限公司 | Instance segmentation method, image processing device and computer readable storage medium |
| CN110930296A (en) * | 2019-11-20 | 2020-03-27 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN110956221A (en) * | 2019-12-17 | 2020-04-03 | 北京化工大学 | Small sample polarization synthetic aperture radar image classification method based on deep recursive network |
| CN111052028A (en) * | 2018-01-23 | 2020-04-21 | 深圳市大疆创新科技有限公司 | System and method for automatic surface and sky detection |
| CN111210434A (en) * | 2019-12-19 | 2020-05-29 | 上海艾麒信息科技有限公司 | Image replacement method and system based on sky identification |
| CN111354059A (en) * | 2020-02-26 | 2020-06-30 | 北京三快在线科技有限公司 | Image processing method and device |
| CN111445486A (en) * | 2020-03-25 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Image processing method, device, equipment and computer readable storage medium |
| CN111507946A (en) * | 2020-04-02 | 2020-08-07 | 浙江工业大学之江学院 | Element data driven flower type pattern rapid generation method based on similarity sample |
| CN111862045A (en) * | 2020-07-21 | 2020-10-30 | 上海杏脉信息科技有限公司 | Method and device for generating blood vessel model |
| CN111915636A (en) * | 2020-07-03 | 2020-11-10 | 闽江学院 | Method and device for positioning and dividing waste target |
| CN112866573A (en) * | 2021-01-13 | 2021-05-28 | 京东方科技集团股份有限公司 | Display, fusion display system and image processing method |
Families Citing this family (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109598678B (en) * | 2018-12-25 | 2023-12-12 | 维沃移动通信有限公司 | Image processing method and device and terminal equipment |
| CN111489359B (en) * | 2019-01-25 | 2023-05-30 | 银河水滴科技（北京）有限公司 | Image segmentation method and device |
| CN111832587B (en) * | 2019-04-18 | 2023-11-14 | 北京四维图新科技股份有限公司 | Image semantic annotation method, device and storage medium |
| CN110992371B (en) * | 2019-11-20 | 2023-10-27 | 北京奇艺世纪科技有限公司 | Portrait segmentation method and device based on priori information and electronic equipment |
| CN111461996B (en) * | 2020-03-06 | 2023-08-29 | 合肥师范学院 | Quick intelligent color matching method for image |
| CN113554658B (en) * | 2020-04-23 | 2024-06-14 | 北京达佳互联信息技术有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN111598902B (en) * | 2020-05-20 | 2023-05-30 | 抖音视界有限公司 | Image segmentation method, device, electronic equipment and computer readable medium |
| CN111832745B (en) * | 2020-06-12 | 2023-08-01 | 北京百度网讯科技有限公司 | Data augmentation method and device and electronic equipment |
| CN112037142B (en) * | 2020-08-24 | 2024-02-13 | 腾讯科技（深圳）有限公司 | Image denoising method, device, computer and readable storage medium |
| CN112508964B (en) * | 2020-11-30 | 2024-02-20 | 北京百度网讯科技有限公司 | Image segmentation method, device, electronic equipment and storage medium |
| CN112800499B (en) * | 2020-12-02 | 2023-12-26 | 杭州群核信息技术有限公司 | Diatom ooze pattern high-order design method based on image processing and real-time material generation capability |
| CN112633142A (en) * | 2020-12-21 | 2021-04-09 | 广东电网有限责任公司电力科学研究院 | Power transmission line violation building identification method and related device |
| CN112819741B (en) * | 2021-02-03 | 2024-03-08 | 四川大学 | Image fusion method and device, electronic equipment and storage medium |
| CN112861885B (en) * | 2021-03-25 | 2023-09-22 | 北京百度网讯科技有限公司 | Image recognition method, device, electronic equipment and storage medium |
| CN113129319B (en) * | 2021-04-29 | 2023-06-23 | 北京市商汤科技开发有限公司 | Image processing method, device, computer equipment and storage medium |
| CN113657401B (en) * | 2021-08-24 | 2024-02-06 | 凌云光技术股份有限公司 | Probability map visualization method and device for defect detection |
| CN114445313A (en) * | 2022-01-28 | 2022-05-06 | 北京百度网讯科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN117437338A (en) * | 2023-10-08 | 2024-01-23 | 书行科技(北京)有限公司 | Special effect generation method, device and computer readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101777180A (en) * | 2009-12-23 | 2010-07-14 | 中国科学院自动化研究所 | Complex background real-time alternating method based on background modeling and energy minimization |
| CN104133956A (en) * | 2014-07-25 | 2014-11-05 | 小米科技有限责任公司 | Method and device for processing pictures |
| CN104463843A (en) * | 2014-10-31 | 2015-03-25 | 南京邮电大学 | Interactive image segmentation method of android system |
| CN104636761A (en) * | 2015-03-12 | 2015-05-20 | 华东理工大学 | Image semantic annotation method based on hierarchical segmentation |
| EP2996085A1 (en) * | 2014-09-09 | 2016-03-16 | icoMetrix NV | Method and system for analyzing image data |
| CN105574513A (en) * | 2015-12-22 | 2016-05-11 | 北京旷视科技有限公司 | Character detection method and device |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8014590B2 (en) * | 2005-12-07 | 2011-09-06 | Drvision Technologies Llc | Method of directed pattern enhancement for flexible recognition |
| CN102486827B (en) * | 2010-12-03 | 2014-11-05 | 中兴通讯股份有限公司 | Extraction method of foreground object in complex background environment and apparatus thereof |
| CN103116754B (en) * | 2013-01-24 | 2016-05-18 | 浙江大学 | Batch images dividing method and system based on model of cognition |
| CN107025457B (en) * | 2017-03-29 | 2022-03-08 | 腾讯科技（深圳）有限公司 | Image processing method and device |
-
2017
- 2017-03-29 CN CN201710199165.XA patent/CN107025457B/en active Active
-
2018
- 2018-03-26 WO PCT/CN2018/080446 patent/WO2018177237A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101777180A (en) * | 2009-12-23 | 2010-07-14 | 中国科学院自动化研究所 | Complex background real-time alternating method based on background modeling and energy minimization |
| CN104133956A (en) * | 2014-07-25 | 2014-11-05 | 小米科技有限责任公司 | Method and device for processing pictures |
| EP2996085A1 (en) * | 2014-09-09 | 2016-03-16 | icoMetrix NV | Method and system for analyzing image data |
| CN104463843A (en) * | 2014-10-31 | 2015-03-25 | 南京邮电大学 | Interactive image segmentation method of android system |
| CN104636761A (en) * | 2015-03-12 | 2015-05-20 | 华东理工大学 | Image semantic annotation method based on hierarchical segmentation |
| CN105574513A (en) * | 2015-12-22 | 2016-05-11 | 北京旷视科技有限公司 | Character detection method and device |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018177237A1 (en) * | 2017-03-29 | 2018-10-04 | 腾讯科技（深圳）有限公司 | Image processing method and device, and storage medium |
| CN107705334A (en) * | 2017-08-25 | 2018-02-16 | 北京图森未来科技有限公司 | A kind of video camera method for detecting abnormality and device |
| CN107705334B (en) * | 2017-08-25 | 2020-08-25 | 北京图森智途科技有限公司 | Camera abnormity detection method and device |
| CN107507201A (en) * | 2017-09-22 | 2017-12-22 | 深圳天琴医疗科技有限公司 | A kind of medical image cutting method and device |
| CN107993191B (en) * | 2017-11-30 | 2023-03-21 | 腾讯科技（深圳）有限公司 | Image processing method and device |
| CN107993191A (en) * | 2017-11-30 | 2018-05-04 | 腾讯科技（深圳）有限公司 | A kind of image processing method and device |
| WO2019109524A1 (en) * | 2017-12-07 | 2019-06-13 | 平安科技（深圳）有限公司 | Foreign object detection method, application server, and computer readable storage medium |
| CN111052028B (en) * | 2018-01-23 | 2022-04-05 | 深圳市大疆创新科技有限公司 | System and method for automatic surface and sky detection |
| CN111052028A (en) * | 2018-01-23 | 2020-04-21 | 深圳市大疆创新科技有限公司 | System and method for automatic surface and sky detection |
| CN108305260A (en) * | 2018-03-02 | 2018-07-20 | 苏州大学 | Detection method, device and the equipment of angle point in a kind of image |
| CN108305260B (en) * | 2018-03-02 | 2022-04-12 | 苏州大学 | Method, device and equipment for detecting angular points in image |
| CN108764143A (en) * | 2018-05-29 | 2018-11-06 | 北京字节跳动网络技术有限公司 | Image processing method, device, computer equipment and storage medium |
| CN108764143B (en) * | 2018-05-29 | 2020-11-24 | 北京字节跳动网络技术有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN110610495A (en) * | 2018-06-15 | 2019-12-24 | 北京京东尚科信息技术有限公司 | Image processing method and system and electronic equipment |
| CN110610495B (en) * | 2018-06-15 | 2022-06-07 | 北京京东尚科信息技术有限公司 | Image processing method and system and electronic equipment |
| CN110910334A (en) * | 2018-09-15 | 2020-03-24 | 北京市商汤科技开发有限公司 | Instance segmentation method, image processing device and computer readable storage medium |
| CN110910334B (en) * | 2018-09-15 | 2023-03-21 | 北京市商汤科技开发有限公司 | Instance segmentation method, image processing device and computer readable storage medium |
| CN110163862B (en) * | 2018-10-22 | 2023-08-25 | 腾讯科技（深圳）有限公司 | Image semantic segmentation method and device and computer equipment |
| CN110163862A (en) * | 2018-10-22 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Image, semantic dividing method, device and computer equipment |
| CN109741347A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | A kind of image partition method of the iterative learning based on convolutional neural networks |
| CN109741347B (en) * | 2018-12-30 | 2021-03-16 | 北京工业大学 | Iterative learning image segmentation method based on convolutional neural network |
| CN110310222A (en) * | 2019-06-20 | 2019-10-08 | 北京奇艺世纪科技有限公司 | A kind of image Style Transfer method, apparatus, electronic equipment and storage medium |
| CN110544218B (en) * | 2019-09-03 | 2024-02-13 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN110544218A (en) * | 2019-09-03 | 2019-12-06 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN110930296A (en) * | 2019-11-20 | 2020-03-27 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN110930296B (en) * | 2019-11-20 | 2023-08-08 | Oppo广东移动通信有限公司 | Image processing method, device, equipment and storage medium |
| CN110956221A (en) * | 2019-12-17 | 2020-04-03 | 北京化工大学 | Small sample polarization synthetic aperture radar image classification method based on deep recursive network |
| CN111210434A (en) * | 2019-12-19 | 2020-05-29 | 上海艾麒信息科技有限公司 | Image replacement method and system based on sky identification |
| CN111354059A (en) * | 2020-02-26 | 2020-06-30 | 北京三快在线科技有限公司 | Image processing method and device |
| CN111354059B (en) * | 2020-02-26 | 2023-04-28 | 北京三快在线科技有限公司 | Image processing method and device |
| CN111445486A (en) * | 2020-03-25 | 2020-07-24 | 腾讯科技（深圳）有限公司 | Image processing method, device, equipment and computer readable storage medium |
| CN111445486B (en) * | 2020-03-25 | 2023-10-03 | 腾讯科技（深圳）有限公司 | Image processing method, device, equipment and computer readable storage medium |
| CN111507946A (en) * | 2020-04-02 | 2020-08-07 | 浙江工业大学之江学院 | Element data driven flower type pattern rapid generation method based on similarity sample |
| CN111915636A (en) * | 2020-07-03 | 2020-11-10 | 闽江学院 | Method and device for positioning and dividing waste target |
| CN111915636B (en) * | 2020-07-03 | 2023-10-24 | 闽江学院 | Method and device for positioning and dividing waste targets |
| CN111862045A (en) * | 2020-07-21 | 2020-10-30 | 上海杏脉信息科技有限公司 | Method and device for generating blood vessel model |
| CN112866573B (en) * | 2021-01-13 | 2022-11-04 | 京东方科技集团股份有限公司 | Display, fusion display system and image processing method |
| CN112866573A (en) * | 2021-01-13 | 2021-05-28 | 京东方科技集团股份有限公司 | Display, fusion display system and image processing method |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2018177237A1 (en) | 2018-10-04 |
| CN107025457B (en) | 2022-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107025457A (en) | A kind of image processing method and device | |
| CN109032720A (en) | Folder icon display methods, system, readable storage medium storing program for executing and terminal | |
| US20130021361A1 (en) | Method and device for rendering user interface font | |
| CN107292886A (en) | Object intrusion detection method and device based on mesh generation and neutral net | |
| CN109101899B (en) | Face detection method and system based on convolutional neural network | |
| CN106056606A (en) | Image processing method and device | |
| CN107680042A (en) | Rendering intent, device, engine and storage medium | |
| CN110706151B (en) | Video-oriented non-uniform style migration method | |
| CN104751507A (en) | Method and device for rendering pattern contents | |
| CN105138802A (en) | Intelligent design system and method for gun barrel | |
| CN107680003A (en) | The node tree generation method and device of project supervision task | |
| CN107527322A (en) | Rendering intent, device, engine and storage medium | |
| CN114936019A (en) | Component and strategy linkage method, device, equipment, system and storage medium | |
| CN108961267A (en) | Image processing method, picture processing unit and terminal device | |
| CN108197203A (en) | A kind of shop front head figure selection method, device, server and storage medium | |
| CN108875759A (en) | A kind of image processing method, device and server | |
| CN112288397A (en) | Flow template configuration method, flow execution method and device and electronic equipment | |
| CN106708623A (en) | Object resource processing method, apparatus and system | |
| CN108153572A (en) | GIS display methods, device and system | |
| CN103838861B (en) | A kind of method and display system based on three-dimension GIS dynamic display information | |
| CN104239212B (en) | Query method and device of test case and establishment method and device of test case | |
| CN105978712A (en) | Method for cloud computation management system | |
| CN109376981A (en) | Determination method, apparatus, server and the data processing method of data processing method | |
| CN108010126A (en) | Method and system based on voxel structure large-scale complex landform | |
| CN107396142A (en) | Interface navigation data processing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |