CN107845072B - Image generating method, device, storage medium and terminal device - Google Patents
Image generating method, device, storage medium and terminal device Download PDFInfo
- Publication number
- CN107845072B CN107845072B CN201710952803.0A CN201710952803A CN107845072B CN 107845072 B CN107845072 B CN 107845072B CN 201710952803 A CN201710952803 A CN 201710952803A CN 107845072 B CN107845072 B CN 107845072B
- Authority
- CN
- China
- Prior art keywords
- image
- style
- picture
- information
- neural network
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 42
- 238000004458 analytical method Methods 0.000 claims abstract description 41
- 238000010348 incorporation Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims description 40
- 238000013135 deep learning Methods 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 13
- 230000015572 biosynthetic process Effects 0.000 claims description 8
- 238000003786 synthesis reaction Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 description 14
- 241000208818 Helianthus Species 0.000 description 8
- 235000003222 Helianthus annuus Nutrition 0.000 description 8
- 238000010422 painting Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 5
- 239000013589 supplement Substances 0.000 description 4
- 238000009826 distribution Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 241000533950 Leucojum Species 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 239000000470 constituent Substances 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 235000020238 sunflower seed Nutrition 0.000 description 2
- 230000017260 vegetative to reproductive phase transition of meristem Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G06T5/77—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/001—Texturing; Colouring; Generation of texture or colour
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Abstract
The invention discloses a kind of image generating method, device, storage medium and electronic equipments, the described method includes: being analyzed characteristic information to obtain processing strategie according to content analysis model and default style by the characteristic information in pre-set image and default style and pre-set image;Characteristic information is handled according to processing strategie to obtain multiple fixed reference features;It by the fixed reference feature random incorporation of corresponding different characteristic information, and is synthesized with pre-set image, obtains multiple images to be selected;Multiple images to be selected are handled according to picture structure preference pattern, obtain the target image for meeting preset structure layout.The new windy table images transform method of profound level is provided.
Description
Technical field
The invention belongs to fields of communication technology more particularly to a kind of image generating method, device, storage medium and terminal to set
It is standby.
Background technique
Under current self-timer, the environmental background for sharing the common requirements for becoming Internet user, to user's different-style
Photo beautification in the APP of mobile terminal be it is unquestionable, have extremely extensive demand.But the photo of each company is beautiful at present
Change major part to be all confined to directly carry out pixel scale processing to photo, or select template, photo face is directly merged.Also
Content element is not obtained out of photo, then carries out the application that new photo generates.
Traditional technical solution specifically includes that pure pixel scale beautification, is specifically that user uploads a photo, passes through
Different convolution kernels handles each pixel, or carries out parameter regulation to image contrast degree etc., obtains mill skin, element
It retouches, the different-effects such as black and white;American-European old master's style conversion is specifically that user uploads a photo, mentions to this photo
The high-level characteristic for taking deep learning model describes the main structured message of image, then to built-in different-style
Photo, the Gram matrix obtained based on each layer of feature calculation of CNN can be very good to capture the style information of image again
(style of writing and texture etc.).Define loss function in conjunction with both information, guide image from some starting point (such as: random noise
Or content images itself) start, continuous iteration optimization is gradually converted into the transformed image of style;The conversion of animation style, specifically
It is to be detected to sky in photo, white clouds is added, tone adjusting then is carried out to photo other parts.
Summary of the invention
The present invention provides a kind of image generating method, device, storage medium and terminal device, is capable of providing new profound level
Windy table images transform method.
The embodiment of the present invention provides a kind of image generating method, is applied to terminal device, the method includes the steps:
Pre-set image and default style are obtained, and obtains the characteristic information in the pre-set image;
Content analysis model is obtained, the characteristic information is carried out according to the content analysis model and the default style
Analysis obtains processing strategie;
The characteristic information is handled according to the processing strategie to obtain multiple fixed reference features;
It by the fixed reference feature random incorporation of the corresponding different characteristic informations, and synthesizes, obtains with the pre-set image
To multiple images to be selected;
The multiple image to be selected is handled according to picture structure preference pattern, obtains meeting preset structure layout
Target image.
Further, the step of characteristic information obtained in the pre-set image, including the pre-set image is passed through
Scene Recognition deep learning network is crossed to obtain scene characteristic, obtain the pre-set image by object identification deep learning network
The pre-set image is obtained at least one in face characteristic by recognition of face by object features.
Further, the acquisition content analysis model, according to the content analysis model and the default style to institute
It states characteristic information and is analyzed the step of obtaining processing strategie, comprising:
The content analysis model includes style policy library, and the style policy library includes knowledge base and policy library;
The knowledge base includes the description information of multiple pictures and the corresponding picture, the description of each picture
Information includes fixed reference feature information and style information;
The policy library includes the image style switching strategy of corresponding each picture and/or the switching strategy of face part;
The fixed reference feature information of picture in the characteristic information and the default style, with the knowledge base and style are believed
Breath is compared, and meets if both comparing, and obtains the fixed reference feature information of the picture and the figure of the corresponding picture
As style switching strategy and/or the switching strategy of face part.
Further, described that the multiple image to be selected is screened according to picture structure preference pattern, met
The step of target image of preset structure layout, comprising:
Described image structure choice model includes the first deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the target image for meeting preset structure layout is obtained.
Further, described that the multiple image to be selected is screened according to picture structure preference pattern, met
The step of target image of preset structure layout, comprising:
Described image structure choice model includes the first deep neural network and the second deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, is obtained each described
The unified target image of target signature style.
The embodiment of the present invention also provides a kind of video generation device, is applied to terminal device, and described device includes:
First acquisition unit for obtaining pre-set image and default style, and obtains the letter of the feature in the pre-set image
Breath;
Second acquisition unit, for obtaining content analysis model, according to the content analysis model and the default style
The characteristic information is analyzed to obtain processing strategie;
First processing units, it is multiple with reference to special for being handled to obtain to the characteristic information according to the processing strategie
Sign;
Synthesis unit, for that will correspond to the fixed reference feature random incorporation of the different characteristic informations, and with it is described pre-
If image synthesizes, multiple images to be selected are obtained;
The second processing unit is obtained for being handled according to picture structure preference pattern the multiple image to be selected
Meet the target image of preset structure layout.
Further, the first acquisition unit is specifically used for:
Including the pre-set image is obtained scene characteristic by scene Recognition deep learning network, by the pre-set image
Object features are obtained by object identification deep learning network, obtain the pre-set image in face characteristic by recognition of face
At least one of.
Further, the content analysis model includes style policy library, and the style policy library includes knowledge base and plan
Slightly library;The knowledge base includes the description information of multiple pictures and the corresponding picture, the description information of each picture
Including fixed reference feature information and style information;The policy library include corresponding each picture image style switching strategy and/or
The switching strategy of face part;
The second acquisition unit is specifically used for:
The fixed reference feature information of picture in the characteristic information and the default style, with the knowledge base and style are believed
Breath is compared, and meets if both comparing, and obtains the fixed reference feature information of the picture and the figure of the corresponding picture
As style switching strategy and/or the switching strategy of face part.
Further, described image structure choice model includes the first deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in first deep neural network, the target figure for meeting preset structure layout is obtained
Picture.
Further, described image structure choice model includes the first deep neural network and the second deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, is obtained each described
The unified target image of target signature style.
The embodiment of the present invention also provides a kind of storage medium, is stored thereon with computer program, when the computer program
When running on computers, so that the computer executes as above described in any item image generating methods.
The embodiment of the present invention also provides a kind of terminal device, including processor and memory, and the memory has computer
Program, the processor is by calling the computer program, for executing described in any item image generating methods as above.
Image generating method, device, storage medium and terminal device provided in an embodiment of the present invention, by pre-set image and
Characteristic information in default style and pre-set image divides characteristic information according to content analysis model and default style
Analysis obtains processing strategie;Characteristic information is handled according to processing strategie to obtain multiple fixed reference features;It will corresponding different characteristic
The fixed reference feature random incorporation of information, and synthesized with pre-set image, obtain multiple images to be selected;According to picture structure preference pattern
Multiple images to be selected are handled, the target image for meeting preset structure layout is obtained.According to default style, pass through content point
Model is analysed, pre-set image different characteristic information is analyzed, different needs is provided and strengthens and weaken stylization strategy, and mend
Fill the multiple fixed reference features for needing to express.Then will corresponding different characteristic information fixed reference feature random incorporation, and with default figure
As synthesis, multiple images to be selected are obtained, image to be selected is laid out again, finally carries out the selection of image to be selected.New deep layer is provided
Secondary windy table images transform method.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, drawings in the following description are only some embodiments of the invention, skill common for this field
For art personnel, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is the flow diagram of image generating method provided in an embodiment of the present invention;
Fig. 2 is another flow diagram of image generating method provided in an embodiment of the present invention;
Fig. 3 is that the characteristic information of image generating method provided in an embodiment of the present invention identifies schematic diagram;
Fig. 4 is the another flow diagram of image generating method provided in an embodiment of the present invention;
Fig. 5 is the policy selection schematic diagram of image generating method provided in an embodiment of the present invention;
Fig. 6 is that the image of image generating method provided in an embodiment of the present invention synthesizes schematic diagram;
Fig. 7 is the structural schematic diagram of video generation device 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.
As shown in FIG. 1, FIG. 1 is the flow diagrams of image generating method provided in an embodiment of the present invention.The present invention is implemented
Example provides a kind of image generating method, is applied to terminal device, the method comprising the steps of:
Step S101 obtains pre-set image and default style, and obtains the characteristic information in the pre-set image.It is wherein pre-
If image can be photo, drawing, picture etc..Default style includes that ink, science fiction, abstract, realistic, sketch, cartoon etc. are a variety of
Different styles.
Step S102 obtains content analysis model, according to the content analysis model and the default style to the spy
Reference breath is analyzed to obtain processing strategie.
Step S103 handles the characteristic information according to the processing strategie to obtain multiple fixed reference features.
Step S104, will the corresponding different characteristic informations the fixed reference feature random incorporation, and with the default figure
As synthesis, multiple images to be selected are obtained.
Step S105 handles the multiple image to be selected according to picture structure preference pattern, obtains meeting default
The target image of topology layout.
Pre-set image different characteristic information is analyzed, difference is provided by content analysis model according to default style
Needs strengthen and reduction stylization strategy, and supplement and need multiple fixed reference features for expressing.It then will corresponding different characteristic letter
The fixed reference feature random incorporation of breath, and synthesized with pre-set image, multiple images to be selected are obtained, image to be selected is laid out again, finally
Carry out the selection of image to be selected.The new windy table images transform method of profound level is provided.
As shown in Fig. 2, Fig. 2 is another flow diagram of image generating method provided in an embodiment of the present invention.Image is raw
It can be to obtain a photo on the terminal device at method, such as shooting is obtained or obtained from other equipment, and one selected
After style, photo is passed through to multiple deep learning networks parallel, photo content information is obtained, such as the information of people in photo, scene
Information, the object information etc. for including.Then pass through content analysis model, determine which part different-style should strengthen, weaken
Which part also lacks which element.Subsequently into the image processing module of different-style, at the content of different piece
Reason.Then the multiple portions that will have been handled generate model into image, carry out image in conjunction with the style that original image and user select
Drafting and colouring processing.
Further, the step of characteristic information obtained in the pre-set image, including the pre-set image is passed through
Scene Recognition deep learning network is crossed to obtain scene characteristic, obtain the pre-set image by object identification deep learning network
The pre-set image is obtained at least one in face characteristic by recognition of face by object features.
As shown in figure 3, the characteristic information that Fig. 3 is image generating method provided in an embodiment of the present invention identifies schematic diagram.Tool
Body, pre-set image can be subjected to content analysis by multiple trained deep learning networks simultaneously, by pre-set image
Deep learning network by scene Recognition, the scene characteristic information being currently located, such as: indoor, outdoor, coffee shop, sea
The different characteristics information such as side;Deep learning network by pre-set image Jing Guo object identification, obtains primary objects list in photo
Characteristic information;Pre-set image is passed through into Face datection, whether is had the characteristic information of face, if there is face and obtains people
The multiple characteristic points of face.It is also possible that then marking out the face, and carry out the age for the face, gender identification if there is face.
All the elements export together.
As shown in Figure 4 and Figure 5, Fig. 4 is the another flow diagram of image generating method provided in an embodiment of the present invention;Figure
5 be the policy selection schematic diagram of image generating method provided in an embodiment of the present invention.The acquisition content analysis model, according to institute
It states content analysis model and the default style and the step of obtaining processing strategie is analyzed to the characteristic information, comprising:
Sub-step 1021, the content analysis model include style policy library, the style policy library include knowledge base and
Policy library;
Sub-step 1022, the knowledge base includes the description information of multiple pictures and the corresponding picture, each described
The description information of picture includes fixed reference feature information and style information;
Sub-step 1023, the policy library include image style switching strategy and/or the face part of corresponding each picture
Switching strategy;
Sub-step 1024, by the fixed reference feature of picture in the characteristic information and the default style, with the knowledge base
Information and style information are compared, and meet if both comparing, and obtain the fixed reference feature information of the picture and correspond to
The image style switching strategy of the picture and/or the switching strategy of face part.
Specifically, content analysis model, receives and decompose after being analyzed in the default style and pre-set image of user's selection
The characteristic information list of corresponding article is inquired under corresponding style strategy, the feature of each article in corresponding style policy library
The decision that information should change
Wherein, the knowledge base includes the description information of multiple pictures and the corresponding picture, the description of each picture
Information includes fixed reference feature information and style information.Picture can be the famous painting of various countries, or welcome figure on network
It draws, is also possible to photo or picture that user voluntarily obtains, the picture drawn such as oneself.Fixed reference feature information includes in picture
Including Item Information, scene information, people information.Item Information may include the specific article in picture, such as sunflower, pen
Cylinder, cloud, ship etc., can also include the Item Information of such as steamer, pleasure boat, speedboat, raft more refinement, and Item Information may be used also
To include the description of other features, as sunflower seeds stage, sunflower development stage, sunflower flowering phase, sunflower are withered
Stage etc..
Knowledge base can also be counted by natural language processing and be constructed by the description information of different-style picture, each
A picture, according to article, picture style, picture author and the time occurred in picture, construct a label vector w1 ...,
wm}。
The policy library can also include statistics strategy, and statistics strategy is the paintings weighting that composition is counted by a large amount of paintings
Distribution.
In order to facilitate inventive concept of the invention is understood, following example is please referred to:
The genre labels K and photo content output S that user provides are obtained, is scanned in knowledge base W, which is answered
The article for including is ranked up, and obtains five top ranked label vectors.
Then according to tactful and artificial strategy progress strategy generating is counted, image procossing is then carried out according to strategy, is executed
Corresponding style work, generates corresponding style to each article and converts image.Artificial strategy is the image provided by designer
Style switching strategy, face part switching strategy.Statistics strategy is the paintings weighting distribution that composition is counted by a large amount of paintings.
Such as: it obtains user styles label { ink }, obtains user picture scene information { snowfield }, obtain user picture object
Product information { people, tree, mountain },
Search knowledge base obtains { snowflake, people, tree, mountain, red plum }, by statistics strategy, obtains snowflake, set, mountain-letter
Change, people-refinement, the strategy of red plum picture library load.
Due to the drafting of some special details, artificial strategy is needed to participate in, thus obtain face part, ink converts plan
Slightly.
It should be noted that this example is only an example, do not limit the invention, as photo can use drawing
Substitution etc..
Further, described that the multiple image to be selected is screened according to picture structure preference pattern, met
The step of target image of preset structure layout, comprising:
Described image structure choice model includes the first deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the target image for meeting preset structure layout is obtained.
As shown in fig. 6, the image that Fig. 6 is image generating method provided in an embodiment of the present invention synthesizes schematic diagram.Described
The multiple image to be selected is screened according to picture structure preference pattern, obtains the target image for meeting preset structure layout
Step, comprising:
Described image structure choice model includes the first deep neural network and the second deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, is obtained each described
The unified target image of target signature style.
Image generation is to receive to provide various image processing functions, receives the list of style switch decision, executes corresponding turn
Operation is changed, corresponding style image is generated to each article.
Then multiple objects random combine is generated into multiple new images, is uniformly input to the deep neural network of the style
In, export structure is one most reasonable.
Then by this image all constituents of output, second deep neural network is inputted, is finely adjusted.So that
Style guarantees unified between various pieces, then reconfigures by what first network exported.
The image generating method that the embodiment of the present invention is proposed obtains Item Information, scene information, people from pre-set image
The characteristic informations such as information, then characteristic information is handled to obtain multiple fixed reference features according to corresponding processing strategie, again
The new multiple fixed reference features of combination obtain new construction and the unified image of style, are users' with the method that image generates
Photo provides the new windy table images transform method of profound level, provides a high-quality user experience.
The analysis of multiple dimensions has been carried out to the image obtained from terminal device, carries out element extraction, including pass through scene
Multiple depth models such as identification, article identification, Face datection, gender identification, age identification, obtain individual features information, then
The demand such as default style proposed according to user converts, and by content analysis model, analyzes new image difference article,
It provides different needs and strengthens and weaken stylization strategy, and supplement the article for needing the user demand expressed.Then pass through figure
It is laid out again as generating, carries out the selection and stylization unification of image generation respectively by two deep neural network models.
As shown in fig. 7, Fig. 7 is the structural schematic diagram of video generation device provided in an embodiment of the present invention.The present invention is implemented
Example also provides a kind of video generation device, is applied to terminal device, and described image generating means 200 include first acquisition unit
201, second acquisition unit 202, first processing units 203, synthesis unit 204 and the second processing unit 205.
First acquisition unit 201 for obtaining pre-set image and default style, and obtains the feature in the pre-set image
Information.Wherein pre-set image can be photo, drawing, picture etc..Default style include ink, science fiction, abstract, realistic, sketch,
A variety of different styles such as cartoon.
Second acquisition unit 202, for obtaining content analysis model, according to the content analysis model and the default wind
Lattice are analyzed to obtain processing strategie to the characteristic information.
First processing units 203 obtain multiple ginsengs for being handled according to the processing strategie the characteristic information
Examine feature.
Synthesis unit 204, for that will correspond to the fixed reference feature random incorporation of the different characteristic informations, and with it is described
Pre-set image synthesis, obtains multiple images to be selected.
The second processing unit 205 is obtained for being handled according to picture structure preference pattern the multiple image to be selected
To the target image for meeting preset structure layout.
Pre-set image different characteristic information is analyzed, difference is provided by content analysis model according to default style
Needs strengthen and reduction stylization strategy, and supplement and need multiple fixed reference features for expressing.It then will corresponding different characteristic letter
The fixed reference feature random incorporation of breath, and synthesized with pre-set image, multiple images to be selected are obtained, image to be selected is laid out again, finally
Carry out the selection of image to be selected.The new windy table images transform method of profound level is provided.
Further, the first acquisition unit is specifically used for:
Including the pre-set image is obtained scene characteristic by scene Recognition deep learning network, by the pre-set image
Object features are obtained by object identification deep learning network, obtain the pre-set image in face characteristic by recognition of face
At least one of.
Specifically, pre-set image can be subjected to content analysis by multiple trained deep learning networks simultaneously,
Deep learning network by pre-set image Jing Guo scene Recognition, the scene characteristic information being currently located, such as: it is indoor, it is outdoor,
Coffee shop, the different characteristics information such as seashore;Deep learning network by pre-set image Jing Guo object identification obtains main in photo
The characteristic information of object list;Pre-set image is passed through into Face datection, whether is had the characteristic information of face, if there is face
And obtain the multiple characteristic points of face.It is also possible that then marking out the face, and carry out age, property for the face if there is face
It does not identify.All the elements export together.
Specifically, content analysis model, receives and decompose after being analyzed in the default style and pre-set image of user's selection
The characteristic information list of corresponding article is inquired under corresponding style strategy, the feature of each article in corresponding style policy library
The decision that information should change
Wherein, the knowledge base includes the description information of multiple pictures and the corresponding picture, the description of each picture
Information includes fixed reference feature information and style information.Picture can be the famous painting of various countries, or welcome figure on network
It draws, is also possible to photo or picture that user voluntarily obtains, the picture drawn such as oneself.Fixed reference feature information includes in picture
Including Item Information, scene information, people information.Item Information may include the specific article in picture, such as sunflower, pen
Cylinder, cloud, ship etc., can also include the Item Information of such as steamer, pleasure boat, speedboat, raft more refinement, and Item Information may be used also
To include the description of other features, as sunflower seeds stage, sunflower development stage, sunflower flowering phase, sunflower are withered
Stage etc..
Knowledge base can also be counted by natural language processing and be constructed by the description information of different-style picture, each
A picture, according to article, picture style, picture author and the time occurred in picture, construct a label vector w1 ...,
wm}。
The policy library can also include statistics strategy, and statistics strategy is the paintings weighting that composition is counted by a large amount of paintings
Distribution.
Further, the content analysis model includes style policy library, and the style policy library includes knowledge base and plan
Slightly library;The knowledge base includes the description information of multiple pictures and the corresponding picture, the description information of each picture
Including fixed reference feature information and style information;The policy library include corresponding each picture image style switching strategy and/or
The switching strategy of face part;
The second acquisition unit is specifically used for:
The fixed reference feature information of picture in the characteristic information and the default style, with the knowledge base and style are believed
Breath is compared, and meets if both comparing, and obtains the fixed reference feature information of the picture and the figure of the corresponding picture
As style switching strategy and/or the switching strategy of face part.
Further, described image structure choice model includes the first deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in first deep neural network, the target figure for meeting preset structure layout is obtained
Picture.
Further, described image structure choice model includes the first deep neural network and the second deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, is obtained each described
The unified target image of target signature style.
Image generation is to receive to provide various image processing functions, receives the list of style switch decision, executes corresponding turn
Operation is changed, corresponding style image is generated to each article.
Then multiple objects random combine is generated into multiple new images, is uniformly input to the deep neural network of the style
In, export structure is one most reasonable.
Then by this image all constituents of output, second deep neural network is inputted, is finely adjusted.So that
Style guarantees unified between various pieces, then reconfigures by what first network exported.
The video generation device that the embodiment of the present invention is proposed obtains Item Information, scene information, people from pre-set image
The characteristic informations such as information, then characteristic information is handled to obtain multiple fixed reference features according to corresponding processing strategie, again
The new multiple fixed reference features of combination obtain new construction and the unified image of style, are users' with the method that image generates
Photo provides the new windy table images transform method of profound level, provides a high-quality user experience.
The analysis of multiple dimensions has been carried out to the image obtained from terminal device, carries out element extraction, including pass through scene
Multiple depth models such as identification, article identification, Face datection, gender identification, age identification, obtain individual features information, then
The demand such as default style proposed according to user converts, and by content analysis model, analyzes new image difference article,
It provides different needs and strengthens and weaken stylization strategy, and supplement the article for needing the user demand expressed.Then pass through figure
It is laid out again as generating, carries out the selection and stylization unification of image generation respectively by two deep neural network models.
When it is implemented, the above modules can be used as independent entity to realize, any combination can also be carried out, is made
It is realized for same or several entities.
Above-mentioned all technical solutions can form alternative embodiment of the invention using any combination, not another herein
One repeats.
In the embodiment of the present invention, the image generating method in described image generating means and foregoing embodiments belongs to same structure
Think, either offer method in described image generation method embodiment can be provided in described image generating means, it is specific
Realization process is detailed in the embodiment of described image generation method, and details are not described herein again.
The embodiment of the present invention also provides a kind of terminal device, including processor and memory, and the memory has computer
Program, the processor is by calling the computer program, for executing described in any item image generating methods as above.
Wherein, the terminal device can be smart phone, tablet computer, desktop computer, laptop or palm
Apparatus such as computer.
The embodiment of the present invention also provides a kind of storage medium, and the storage medium is stored with computer program, when the meter
When calculation machine program is run on computers, so that the computer executes the image generating method in any of the above-described embodiment, than
Such as: obtaining pre-set image and default style, and obtain the characteristic information in the pre-set image;Obtain content analysis model, root
The characteristic information is analyzed to obtain processing strategie according to the content analysis model and the default style;According to the place
Reason strategy handles the characteristic information to obtain multiple fixed reference features;By the reference of the corresponding different characteristic informations
Feature random incorporation, and synthesized with the pre-set image, obtain multiple images to be selected;According to picture structure preference pattern to described
Multiple images to be selected are handled, and the target image for meeting preset structure layout is obtained.
In embodiments of the present invention, the storage medium can be magnetic disk, CD, read-only memory (Read Only
Memory, ROM) or random access memory (Random Access Memory, RAM) etc..
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, there is no the portion being described in detail in some embodiment
Point, reference can be made to the related descriptions of other embodiments.
It should be noted that for the described image generation method of the embodiment of the present invention, this field common test personnel
It is understood that realize all or part of the process of described image of embodiment of the present invention generation method, is that can pass through computer program
It is completed to control relevant hardware, the computer program can be stored in a computer-readable storage medium, such as store
It is executed in the memory of electronic equipment, and by least one processor in the electronic equipment, may include in the process of implementation
Such as the process of the embodiment of described image generation method.Wherein, the storage medium can for magnetic disk, CD, read-only memory,
Random access memory etc..
For the described image generating means of the embodiment of the present invention, each functional module be can integrate in a processing core
In piece, it is also possible to modules and physically exists alone, can also be integrated in two or more modules in a module.On
It states integrated module both and can take the form of hardware realization, can also be realized in the form of software function module.The collection
If at module realized in the form of software function module and when sold or used as an independent product, also can store
In one computer-readable storage medium, the storage medium is for example read-only memory, disk or CD etc..
Be provided for the embodiments of the invention above a kind of image generating method, device, storage medium and electronic equipment into
It has gone and has been discussed in detail, used herein a specific example illustrates the principle and implementation of the invention, the above implementation
The explanation of example is merely used to help understand method and its core concept of the invention;Meanwhile for those skilled in the art, according to
According to thought of the invention, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification
It should not be construed as limiting the invention.
Claims (10)
1. a kind of image generating method is applied to terminal device, which is characterized in that the method includes the steps:
Pre-set image and default style are obtained, and obtains the characteristic information in the pre-set image;
Content analysis model is obtained, the characteristic information is analyzed according to the content analysis model and the default style
Obtain processing strategie;
The characteristic information is handled according to the processing strategie to obtain multiple fixed reference features;
It by the fixed reference feature random incorporation of the corresponding different characteristic informations, and synthesizes, obtains more with the pre-set image
A image to be selected;
The multiple image to be selected is handled according to picture structure preference pattern, obtains the target for meeting preset structure layout
Image;
Wherein, the acquisition content analysis model, believes the feature according to the content analysis model and the default style
Breath is analyzed the step of obtaining processing strategie, comprising:
The content analysis model includes style policy library, and the style policy library includes knowledge base and policy library;
The knowledge base includes the description information of multiple pictures and the corresponding picture, the description information of each picture
Including fixed reference feature information and style information;
The policy library includes the image style switching strategy of corresponding each picture and/or the switching strategy of face part;
By the fixed reference feature information of picture in the characteristic information and the default style, with the knowledge base and style information into
Row compares, and meets if both comparing, and obtains the fixed reference feature information of the picture and the image wind of the corresponding picture
Lattice switching strategy and/or the switching strategy of face part.
2. image generating method according to claim 1, which is characterized in that the feature obtained in the pre-set image
The step of information, including by the pre-set image by scene Recognition deep learning network obtain scene characteristic, will it is described preset
Image obtains object features by object identification deep learning network, the pre-set image is obtained face spy by recognition of face
At least one of in sign.
3. image generating method according to claim 1, which is characterized in that it is described according to picture structure preference pattern to institute
It states multiple images to be selected to be screened, obtains the step of meeting the target image of preset structure layout, comprising:
Described image structure choice model includes the first deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the target image for meeting preset structure layout is obtained.
4. image generating method according to claim 1, which is characterized in that it is described according to picture structure preference pattern to institute
It states multiple images to be selected to be screened, obtains the step of meeting the target image of preset structure layout, comprising:
Described image structure choice model includes the first deep neural network and the second deep neural network;
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, each target signature is obtained
The unified target image of style.
5. a kind of video generation device, it is applied to terminal device, which is characterized in that described device includes:
First acquisition unit for obtaining pre-set image and default style, and obtains the characteristic information in the pre-set image;
Second acquisition unit, for obtaining content analysis model, according to the content analysis model and the default style to institute
Characteristic information is stated to be analyzed to obtain processing strategie;
First processing units obtain multiple fixed reference features for being handled according to the processing strategie the characteristic information;
Synthesis unit, for that will correspond to the fixed reference feature random incorporation of the different characteristic informations, and with the default figure
As synthesis, multiple images to be selected are obtained;
The second processing unit is met for being handled according to picture structure preference pattern the multiple image to be selected
The target image of preset structure layout;
Wherein, the content analysis model includes style policy library, and the style policy library includes knowledge base and policy library;It is described
Knowledge base includes the description information of multiple pictures and the corresponding picture, and the description information of each picture includes reference
Characteristic information and style information;The policy library includes image style switching strategy and/or the face part of corresponding each picture
Switching strategy;
The second acquisition unit is specifically used for:
By the fixed reference feature information of picture in the characteristic information and the default style, with the knowledge base and style information into
Row compares, and meets if both comparing, and obtains the fixed reference feature information of the picture and the image wind of the corresponding picture
Lattice switching strategy and/or the switching strategy of face part.
6. video generation device according to claim 5, which is characterized in that the first acquisition unit is specifically used for:
Including the pre-set image is obtained scene characteristic by scene Recognition deep learning network, passes through the pre-set image
Object identification deep learning network obtains object features, obtains the pre-set image in face characteristic extremely by recognition of face
One item missing.
7. video generation device according to claim 5, which is characterized in that described image structure choice model includes first
Deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in first deep neural network, the target image for meeting preset structure layout is obtained.
8. video generation device according to claim 5, which is characterized in that described image structure choice model includes first
Deep neural network and the second deep neural network;
Described the second processing unit is specifically used for:
Multiple images to be selected are inputted in the first deep neural network, the image to be selected for meeting preset structure layout is obtained,
The image to be selected for meeting preset structure layout is inputted in the second deep neural network and is adjusted, each target signature is obtained
The unified target image of style.
9. a kind of storage medium, is stored thereon with computer program, which is characterized in that when the computer program on computers
When operation, so that the computer executes such as the described in any item image generating methods of Claims 1-4.
10. a kind of terminal device, including processor and memory, the memory have computer program, which is characterized in that described
Processor is by calling the computer program, for executing such as the described in any item image generating methods of Claims 1-4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710952803.0A CN107845072B (en) | 2017-10-13 | 2017-10-13 | Image generating method, device, storage medium and terminal device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710952803.0A CN107845072B (en) | 2017-10-13 | 2017-10-13 | Image generating method, device, storage medium and terminal device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107845072A CN107845072A (en) | 2018-03-27 |
CN107845072B true CN107845072B (en) | 2019-03-12 |
Family
ID=61661350
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710952803.0A Active CN107845072B (en) | 2017-10-13 | 2017-10-13 | Image generating method, device, storage medium and terminal device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107845072B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533744B (en) * | 2018-05-25 | 2023-02-28 | 阿里巴巴集团控股有限公司 | Method, device and system for synthesizing picture and mobile terminal |
CN108961157B (en) * | 2018-06-19 | 2021-06-01 | Oppo广东移动通信有限公司 | Picture processing method, picture processing device and terminal equipment |
CN108898587A (en) * | 2018-06-19 | 2018-11-27 | Oppo广东移动通信有限公司 | Image processing method, picture processing unit and terminal device |
CN108898169B (en) * | 2018-06-19 | 2021-06-01 | Oppo广东移动通信有限公司 | Picture processing method, picture processing device and terminal equipment |
CN108805095A (en) * | 2018-06-19 | 2018-11-13 | Oppo广东移动通信有限公司 | image processing method, device, mobile terminal and computer readable storage medium |
CN109308681B (en) * | 2018-09-29 | 2023-11-24 | 北京字节跳动网络技术有限公司 | Image processing method and device |
CN109146825B (en) * | 2018-10-12 | 2020-11-27 | 深圳美图创新科技有限公司 | Photography style conversion method, device and readable storage medium |
CN112819685B (en) * | 2019-11-15 | 2022-11-04 | 青岛海信移动通信技术股份有限公司 | Image style mode recommendation method and terminal |
CN111046975B (en) * | 2019-12-27 | 2023-05-12 | 深圳云天励飞技术有限公司 | Portrait generation method, device and system, electronic equipment and storage medium |
CN115841432B (en) * | 2023-02-09 | 2023-08-08 | 北京达佳互联信息技术有限公司 | Method, device, equipment and medium for determining and training beauty special effect data |
CN117036203B (en) * | 2023-10-08 | 2024-01-26 | 杭州黑岩网络科技有限公司 | Intelligent drawing method and system |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103997593A (en) * | 2013-02-18 | 2014-08-20 | 卡西欧计算机株式会社 | Image creating device, image creating method and recording medium storing program |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9159147B2 (en) * | 2013-03-15 | 2015-10-13 | Airdrawer Llc | Method and apparatus for personalized handwriting avatar |
US10726560B2 (en) * | 2014-10-31 | 2020-07-28 | Fyusion, Inc. | Real-time mobile device capture and generation of art-styled AR/VR content |
GB2542890A (en) * | 2015-10-02 | 2017-04-05 | Adobe Systems Inc | Searching using specific attributes found in images |
CN106920208A (en) * | 2015-12-24 | 2017-07-04 | 掌赢信息科技(上海)有限公司 | A kind of face stylizing method and electronic equipment based on image procossing |
CN107180410A (en) * | 2017-04-11 | 2017-09-19 | 中国农业大学 | The stylized method for reconstructing and device of a kind of image |
CN107171932B (en) * | 2017-04-27 | 2021-06-08 | 腾讯科技(深圳)有限公司 | Picture style conversion method, device and system |
CN107240085A (en) * | 2017-05-08 | 2017-10-10 | 广州智慧城市发展研究院 | A kind of image interfusion method and system based on convolutional neural networks model |
-
2017
- 2017-10-13 CN CN201710952803.0A patent/CN107845072B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103997593A (en) * | 2013-02-18 | 2014-08-20 | 卡西欧计算机株式会社 | Image creating device, image creating method and recording medium storing program |
Also Published As
Publication number | Publication date |
---|---|
CN107845072A (en) | 2018-03-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107845072B (en) | Image generating method, device, storage medium and terminal device | |
Lin et al. | Probabilistic color-by-numbers: Suggesting pattern colorizations using factor graphs | |
CN109919869A (en) | A kind of image enchancing method, device and storage medium | |
CN110222722A (en) | Interactive image stylization processing method, calculates equipment and storage medium at system | |
Zhang et al. | Compositional model-based sketch generator in facial entertainment | |
WO2021164550A1 (en) | Image classification method and apparatus | |
CN111967533B (en) | Sketch image translation method based on scene recognition | |
CN111161137A (en) | Multi-style Chinese painting flower generation method based on neural network | |
Liu et al. | Name your style: An arbitrary artist-aware image style transfer | |
CN110415184A (en) | A kind of multi-modality images Enhancement Method based on orthogonal first space | |
Zhang et al. | A comprehensive survey on computational aesthetic evaluation of visual art images: Metrics and challenges | |
Xiao et al. | Vismantic: Meaning-making with Images. | |
Wu et al. | StyleMe: Towards Intelligent Fashion Generation with Designer Style | |
CN112862672B (en) | Liu-bang generation method, device, computer equipment and storage medium | |
KR20160074958A (en) | Apparatus and method for generating motion effects by analyzing motion of object | |
Weng et al. | Data augmentation computing model based on generative adversarial network | |
CN113838158B (en) | Image and video reconstruction method and device, terminal equipment and storage medium | |
Oliva et al. | Deep mean maps | |
CN106469437B (en) | Image processing method and image processing apparatus | |
Togo et al. | Text-guided style transfer-based image manipulation using multimodal generative models | |
Li et al. | Chinese painting style transfer system based on machine learning | |
Wu et al. | Preserving structural consistency in arbitrary artist and artwork style transfer | |
CN113553633A (en) | Data generation method and device, electronic equipment and computer storage medium | |
Fan et al. | Deep Facial Synthesis: A New Challenge | |
Talasila et al. | Enhancing Text-to-Image Synthesis with an Improved Semi-Supervised Image Generation Model Incorporating N-Gram, Enhanced TF-IDF, and BOW Techniques |
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 |