CN107886491A - A kind of image combining method based on pixel arest neighbors - Google Patents
A kind of image combining method based on pixel arest neighbors Download PDFInfo
- Publication number
- CN107886491A CN107886491A CN201711206725.6A CN201711206725A CN107886491A CN 107886491 A CN107886491 A CN 107886491A CN 201711206725 A CN201711206725 A CN 201711206725A CN 107886491 A CN107886491 A CN 107886491A
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- output
- training
- arest neighbors
- 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.)
- Withdrawn
Links
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 29
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 24
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 24
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000011218 segmentation Effects 0.000 claims description 9
- 230000013011 mating Effects 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 239000000470 constituent Substances 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000009977 dual effect Effects 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 claims description 3
- 230000007935 neutral effect Effects 0.000 claims description 3
- 230000037303 wrinkles Effects 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010189 synthetic method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of image combining method based on pixel arest neighbors proposed in the present invention, its main contents include:The synthesis of convolutional neural networks (CNNs), pixel corresponds to and pixel arest neighbors:One-to-many mapping, its process is, training one is initial to return device-convolutional neural networks (CNN), incomplete input is mapped to single output image, then K-NN search is performed to the pixel from this recurrence output, the training examples of matched pixel, efficiently match index is then carried out using the multiple dimensioned dramatic symbol for catching appropriate context level, finally exported from training set to synthesis, it is corresponding to generate intensive Pixel-level.The present invention can naturally generate multiple outputs, while can explain and obey the constraint of user so that faster, the image of synthesis is closer to original image for picture search and matching speed.
Description
Technical field
The present invention relates to image to synthesize field, more particularly, to a kind of image combining method based on pixel arest neighbors.
Background technology
With popularization of the digital product in people live, digital picture becomes more and more important information carrier.Have
Caused digital picture is it is impossible to meet the subjective esthetic requirement of people from natural scene a bit, or in order to the original such as entertain
Cause, it is desirable to arbitrarily change some contents in picture, artificially synthesize some new pictures true to nature.Image synthesizes
Technology can be applied to be made and the research of micro- expression and dynamic in virtual cartoon scene, the picture editting of mobile device, mankind's fine motion
Draw the fields such as teaching;The technologies such as picture editting can be combined simultaneously, realize user's clothing needed for autonomous editor in shopping at network,
So as to be easier to search customer satisfaction system end article;The correlation circumstance for predicting environment in advance can also be synthesized by image,
So as to the offer facility such as the operations on the sea such as marine traffic control, fishing and marine regatta.However, due to mode issue
And a large amount of different outputs can not be produced, while also it is difficult to control synthesis output;And the data and bright of lacking training in practice
Aobvious distance metric, it is also difficult to by search extension into large-scale training set.
The present invention proposes a kind of image combining method based on pixel arest neighbors, first trains an initial recurrence device-volume
Product neutral net (CNN), incomplete input is mapped to single output image, then to the pixel from this recurrence output
K-NN search is performed, then carrys out matched pixel using the multiple dimensioned dramatic symbol for catching appropriate context level, efficiently
The training examples of ground match index, finally exported from training set to synthesis, it is corresponding to generate intensive Pixel-level.The present invention can
Multiple outputs are naturally generated, while can explain and obey the constraint of user so that picture search and matching speed faster, close
Into image closer to original image.
The content of the invention
The problems such as unmanageable synthesis output, it is an object of the invention to provide a kind of figure based on pixel arest neighbors
As synthetic method, an initial recurrence device-convolutional neural networks (CNN) is first trained, incomplete input is mapped to single defeated
Go out image, K-NN search then is performed to the pixel from this recurrence output, then using the appropriate context level of seizure
Other multiple dimensioned dramatic symbol carrys out the training examples of matched pixel, efficiently match index, finally from training set to synthesis
Output, it is corresponding to generate intensive Pixel-level.
To solve the above problems, the present invention provides a kind of image combining method based on pixel arest neighbors, its main contents
Including:
(1) synthesis of convolutional neural networks (CNNs);
(2) pixel is corresponding;
(3) pixel arest neighbors:One-to-many mapping.
Wherein, described image combining method, an initial recurrence device-convolutional neural networks (CNN) is trained first, will
Incomplete input is mapped to single output image;This output image is restricted, and is a single output;Then to coming
K-NN search is performed from the pixel of this recurrence output;Accorded with using the multiple dimensioned dramatic for catching appropriate context level
Carry out matched pixel (the recurrence output from training data);The efficiently training examples of match index, finally from training set to
Synthesis output, it is corresponding to generate intensive Pixel-level.
Wherein, the synthesis of described convolutional neural networks (CNNs), CNNs are applied in segmentation, deep learning and surface normal
Estimation, semantic border detection etc.;These networks usually using image tag data to upper standard loss (such as softmax or
l2Return) it is trained;However, such network generally can not handle what the image from (imperfect) label synthesized well
Inverse problem;One main innovation is the introduction of the generation network (GAN) of dual training;This expression formula is in computer vision
Power is had a very big impact, generates task applied to various images, to low-resolution image, segmentation masking-out, surface normal
Figure and other inputs are handled.
Wherein, described pixel is corresponding, and an important results of pixel orientation arest neighbors are in synthesis output and training sample
It is corresponding that pixel is generated between example;The semantic corresponding relation between inquiry and training image pixel is established, can be from training sample
Middle extraction high-frequency information, a new image is synthesized in the input given from one.
Wherein, described pixel arest neighbors:One-to-many mapping, the problem of information drawing picture is synthesized, are defined as follows:It is given defeated
Enter x condition (such as edge graph, normal depth figure or low-resolution image), synthesize the output image of high quality;Assuming that input/
The training pair of output, is designated as (xn,yn);Simplest method is exactly using this task as (non-linear) regression problem:
Wherein, f (xn;ω) refer to being returned the output of device with any (being probably nonlinear) of ω parametrizations;In public affairs
Full convolutional neural networks are used in formula, particularly pixel network is as nonlinear regression device;Pixel arest neighbors include frequency analysis,
Example matching, mating chemical composition, pixel representation and effectively search.
Further, described frequency analysis, prediction output f (x) are directly analyzed in the case of super-resolution, its
In the case that conditional input x is low-resolution image;The low-resolution image of given face, it is understood that there may be output life can be used as
Into multiple textures (such as wrinkle) or minute shapes clue (such as local feature of nose);In practice, this group output is past
Toward by " fuzzy " the single output for by returning return;This can be with input, output and the frequency analysis of original target image
It can be clearly seen;Assuming that single output is sufficiently used for intermediate frequency output, but multiple outputs are needed to catch possible high frequency texture
Space.
Further, described example matching, in order to catch multiple possible outputs, classics are used in computer vision
Nonparametric technique;Simple K- arest neighbors (KNN) algorithm can return to report K output;However, it is possible to predict f (x) with it
(multiple possible) high frequency imaging lost, rather than return to whole image with KNN models:
Global (x)=f (x)+(yk-f(xk)) (2)
Wherein,Dist be measure two (intermediate frequencies) rebuild between similitude away from
From function;Multiple outputs are generated, K best match, rather than overall best match can be reported from training set.
Further, described mating chemical composition, it is more to synthesize by the way that (high frequency) patch is replicated and pasted from training set
Output;It is in reconstruction image to allow such composition matching, i.e. simple match single pixel rather than global image
Ith pixel writes fi(x), final synthesis output can be written as:
Compi(x)=fi(x)+(yjk-fi(xk) (3)
Wherein,yjkRefer to the output pixel j in training example k.
Further, described pixel representation, if distance function only considers global information, combinations matches are reduced to
Global (example) matching;On the contrary, the different levels of deep layer network tend to capture the spatial context of varying number (due to difference
Acceptance region);The multiple dimensioned pixel that descriptor aggregates into these information across many levels one high precision represents;Construction one
Individual pixel descriptor, using from conv- { 12,22,33,43,53Feature train the pixel network mould for semantic segmentation
Type;In order to assess pixel similarity, the COS distance between two descriptors is calculated.
Further, described effective search, reconstructed image f (x) is given, global K- is found first by conv-5 features
Arest neighbors, then T × T pixel window search pixels level matching around the pixel i only in this group of K image;In practice,
Change K from { 1,2 ..., 10 }, change T from { 1,3,5,10,96 }, and 72 candidate's outputs are generated for given input;By
It is 96 × 96 in the size of composograph, search parameter includes full constituent output (K=10, T=96) and global sample matches (K
=1, T=1), exported them as candidate.
Brief description of the drawings
Fig. 1 is a kind of system framework figure of the image combining method based on pixel arest neighbors of the present invention.
Fig. 2 is a kind of frequency analysis of the image combining method based on pixel arest neighbors of the present invention.
Fig. 3 is a kind of pixel representation of the image combining method based on pixel arest neighbors of the present invention.
Fig. 4 is a kind of effective search of the image combining method based on pixel arest neighbors of the present invention.
Embodiment
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combine, the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 is a kind of system framework figure of the image combining method based on pixel arest neighbors of the present invention.Mainly include convolution
The synthesis of neutral net (CNNs), pixel corresponds to and pixel arest neighbors:One-to-many mapping.
Image combining method, an initial recurrence device-convolutional neural networks (CNN) is trained first, by incomplete input
It is mapped to single output image;This output image is restricted, and is a single output;Then to defeated from this recurrence
The pixel gone out performs K-NN search;Carry out matched pixel using the multiple dimensioned dramatic symbol for catching appropriate context level
(the recurrence output from training data);The efficiently training examples of match index, finally exported from training set to synthesis, it is raw
It is corresponding into intensive Pixel-level.
The synthesis of convolutional neural networks (CNNs), CNNs are applied on segmentation, deep learning and surface normal estimation, semantic side
Boundary's detection etc.;These networks are usually using image tag data to upper standard loss (such as softmax or l2Return) carry out
Training;However, such network generally can not handle the inverse problem that the image from (imperfect) label synthesizes well;One
Main innovation is the introduction of the generation network (GAN) of dual training;This expression formula has very big in computer vision
Influence power, task is generated applied to various images, to low-resolution image, segmentation masking-out, surface normal figure and other are defeated
Enter to be handled.
Pixel is corresponding, and an important results of pixel orientation arest neighbors are to generate picture between synthesis output and training examples
Element is corresponding;The semantic corresponding relation between inquiry and training image pixel is established, high frequency letter can be extracted from training sample
Breath, a new image is synthesized in the input given from one.
Pixel arest neighbors:One-to-many mapping, the problem of information drawing picture is synthesized, are defined as follows:Given input x condition (example
Such as edge graph, normal depth figure or low-resolution image), synthesize the output image of high quality;Assuming that the training of input/output
It is right, it is designated as (xn,yn);Simplest method is exactly using this task as (non-linear) regression problem:
Wherein, f (xn;ω) refer to being returned the output of device with any (being probably nonlinear) of ω parametrizations;In public affairs
Full convolutional neural networks are used in formula, particularly pixel network is as nonlinear regression device;Pixel arest neighbors include frequency analysis,
Example matching, mating chemical composition, pixel representation and effectively search.
Example is matched, and in order to catch multiple possible outputs, classical nonparametric technique is used in computer vision;Letter
Single K- arest neighbors (KNN) algorithm can return to report K output;However, it is possible to (multiple possibility of f (x) loss are predicted with it
) high frequency imaging, rather than return to whole image with KNN models:
Global (x)=f (x)+(yk-f(xk)) (2)
Wherein,Dist be measure two (intermediate frequencies) rebuild between similitude away from
From function;Multiple outputs are generated, K best match, rather than overall best match can be reported from training set.
Mating chemical composition, more outputs are synthesized by being replicated from training set and pasting (high frequency) patch;In order to allow
Such composition matching, i.e. simple match single pixel rather than global image, it is that the ith pixel in reconstruction image writes fi
(x), final synthesis output can be written as:
Compi(x)=fi(x)+(yjk-fi(xk) (3)
Wherein,yjkRefer to the output pixel j in training example k.
Fig. 2 is a kind of frequency analysis of the image combining method based on pixel arest neighbors of the present invention.Prediction output f (x) exists
Directly analyzed in the case of super-resolution, in the case that wherein condition entry x is low-resolution image;Give the low of face
Image in different resolution, it is understood that there may be can be as the multiple textures (such as wrinkle) or minute shapes clue (such as nose of output generation
Local feature);In practice, this group output is often by " fuzzy " for by returning the single output returned;This is in input, output
With in the frequency analysis of original target image it will be clear that;Assuming that single output is sufficiently used for intermediate frequency output, but need
It is multiple to export to catch the space of possible high frequency texture.
Fig. 3 is a kind of pixel representation of the image combining method based on pixel arest neighbors of the present invention.This figure shows right
The output of various input patterns.If distance function only considers global information, combinations matches are reduced to global (example) matching;
On the contrary, the different levels of deep layer network tend to capture the spatial context of varying number (due to different acceptance regions);Description
The multiple dimensioned pixel that symbol aggregates into these information across many levels one high precision represents;A pixel descriptor is constructed,
Using from conv- { 12,22,33,43,53Feature train the pixel network model for semantic segmentation;In order to assess picture
Plain similitude, calculate the COS distance between two descriptors.
Fig. 4 is a kind of effective search of the image combining method based on pixel arest neighbors of the present invention.This figure shows use
The example for multiple outputs that this method is generated by simply changing these parameters.Given reconstructed image f (x), first by conv-
5 features find global K- arest neighbors, then T × T pixel window search pixels around the pixel i only in this group of K image
Level matching;In practice, change K from { 1,2 ..., 10 }, change T from { 1,3,5,10,96 }, and be given input life
Exported into 72 candidates;Because the size of composograph is 96 × 96, search parameter includes full constituent output (K=10, T=96)
With global sample matches (K=1, T=1), exported them as candidate.
For those skilled in the art, the present invention is not restricted to the details of above-described embodiment, in the essence without departing substantially from the present invention
In the case of refreshing and scope, the present invention can be realized with other concrete forms.In addition, those skilled in the art can be to this hair
Bright to carry out various changes and modification without departing from the spirit and scope of the present invention, these improvement and modification also should be regarded as the present invention's
Protection domain.Therefore, appended claims are intended to be construed to include preferred embodiment and fall into all changes of the scope of the invention
More and change.
Claims (10)
1. a kind of image combining method based on pixel arest neighbors, it is characterised in that mainly including convolutional neural networks (CNNs)
Synthesis (one);Pixel is corresponding (two);Pixel arest neighbors:One-to-many mapping (three).
2. based on the image combining method described in claims 1, it is characterised in that train an initial recurrence device-volume first
Product neutral net (CNN), incomplete input is mapped to single output image;This output image is restricted, and is one
Single output;Then K-NN search is performed to the pixel for carrying out autoregression output;Use the appropriate context level of seizure
Multiple dimensioned dramatic symbol carrys out matched pixel (the recurrence output from training data);The efficiently training examples of match index,
Finally exported from training set to synthesis, it is corresponding to generate intensive Pixel-level.
3. the synthesis (one) based on the convolutional neural networks (CNNs) described in claims 1, it is characterised in that CNNs is applied
Segmentation, deep learning and surface normal estimation, semantic border detection etc.;These networks are usually using image tag data to upper
Standard loss (such as softmax or l2Return) it is trained;Come from however, such network generally can not be handled well
The inverse problem of the image synthesis of (imperfect) label;One main innovation is the introduction of the generation network (GAN) of dual training;
This expression formula has a very big impact power in computer vision, task is generated applied to various images, to low resolution
Rate image, segmentation masking-out, surface normal figure and other inputs are handled.
It is 4. corresponding (two) based on the pixel described in claims 1, it is characterised in that an important knot of pixel orientation arest neighbors
Fruit is to generate pixel between synthesis output and training examples to correspond to;It is semantic right between inquiry and training image pixel to establish
It should be related to, high-frequency information can be extracted from training sample, a new image is synthesized in the input given from one.
5. based on the pixel arest neighbors described in claims 1:One-to-many mapping (three), it is characterised in that close information drawing picture
Into the problem of be defined as follows:Given input x condition (such as edge graph, normal depth figure or low-resolution image), synthesis are high
The output image of quality;Assuming that the training pair of input/output, is designated as (xn,yn);Simplest method is exactly that this task is made
For (non-linear) regression problem:
<mrow>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>&omega;</mi>
</munder>
<mo>|</mo>
<mo>|</mo>
<mi>&omega;</mi>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
<mo>+</mo>
<munder>
<mo>&Sigma;</mo>
<mi>n</mi>
</munder>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>y</mi>
<mi>n</mi>
</msub>
<mo>-</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>n</mi>
</msub>
<mo>;</mo>
<mi>&omega;</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msub>
<mo>|</mo>
<msub>
<mi>l</mi>
<mn>2</mn>
</msub>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, f (xn;ω) refer to being returned the output of device with any (being probably nonlinear) of ω parametrizations;Make in formula
With full convolutional neural networks, particularly pixel network is as nonlinear regression device;Pixel arest neighbors includes frequency analysis, example
Match somebody with somebody, mating chemical composition, pixel representation and effectively search.
6. based on the frequency analysis described in claims 5, it is characterised in that prediction output f (x) is in the case of super-resolution
Directly analyzed, in the case that wherein condition entry x is low-resolution image;The low-resolution image of given face, may
In the presence of the multiple textures (such as wrinkle) or minute shapes clue (such as local feature of nose) that can be used as output generation;In reality
In trampling, this group output is often by " fuzzy " for by returning the single output returned;This is in input, output and original target image
In frequency analysis it will be clear that;Assuming that single output is sufficiently used for intermediate frequency output, but need multiple outputs can to catch
The space of the high frequency texture of energy.
7. based on the example matching described in claims 5, it is characterised in that in order to catch multiple possible outputs, calculating
Classical nonparametric technique is used in machine vision;Simple K- arest neighbors (KNN) algorithm can return to report K output;However, can
To predict (multiple possible) high frequency imaging that f (x) loses with it, rather than with KNN models return to whole image:
Global (x)=f (x)+(yk-f(xk)) (2)
Wherein,Dist be measure two (intermediate frequencies) rebuild between similitude apart from letter
Number;Multiple outputs are generated, K best match, rather than overall best match can be reported from training set.
8. based on the mating chemical composition described in claims 5, it is characterised in that by being replicated from training set and pasting (high frequency)
Patch synthesizes more outputs;In order to allow such composition matching, i.e. simple match single pixel rather than global image,
F is write for the ith pixel in reconstruction imagei(x), final synthesis output can be written as:
Compi(x)=fi(x)+(yjk-fi(xk) (3)
Wherein,yjkRefer to the output pixel j in training example k.
9. based on the pixel representation described in claims 5, it is characterised in that if distance function only considers global information,
Then combinations matches are reduced to global (example) matching;On the contrary, the different levels of deep layer network tend to capture the sky of varying number
Between context (due to different acceptance regions);Descriptor aggregates into these information across many levels more chis of one high precision
Pixel is spent to represent;A pixel descriptor is constructed, using from conv- { 12,22,33,43,53Feature come train be used for semanteme
The pixel network model of segmentation;In order to assess pixel similarity, the COS distance between two descriptors is calculated.
10. based on effective search described in claims 5, it is characterised in that given reconstructed image f (x), first by
Conv-5 features find global K- arest neighbors, then T × T pixel window search around the pixel i only in this group of K image
Pixel matching;In practice, change K from { 1,2 ..., 10 } are middle, change T from { 1,3,5,10,96 }, and it is defeated for what is given
Enter 72 candidate's outputs of generation;Because the size of composograph is 96 × 96, search parameter includes full constituent output (K=10, T
=96) and global sample matches (K=1, T=1), exported them as candidate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711206725.6A CN107886491A (en) | 2017-11-27 | 2017-11-27 | A kind of image combining method based on pixel arest neighbors |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711206725.6A CN107886491A (en) | 2017-11-27 | 2017-11-27 | A kind of image combining method based on pixel arest neighbors |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107886491A true CN107886491A (en) | 2018-04-06 |
Family
ID=61775376
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711206725.6A Withdrawn CN107886491A (en) | 2017-11-27 | 2017-11-27 | A kind of image combining method based on pixel arest neighbors |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107886491A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190684A (en) * | 2018-08-15 | 2019-01-11 | 西安电子科技大学 | SAR image sample generating method based on sketch and structural generation confrontation network |
CN109361934A (en) * | 2018-11-30 | 2019-02-19 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment and storage medium |
CN111798935A (en) * | 2019-04-09 | 2020-10-20 | 南京药石科技股份有限公司 | Universal compound structure-property correlation prediction method based on neural network |
CN112365533A (en) * | 2020-10-15 | 2021-02-12 | 浙江大华技术股份有限公司 | Coal flow monitoring method and device based on image segmentation and electronic device |
CN113627341A (en) * | 2021-08-11 | 2021-11-09 | 人民中科(济南)智能技术有限公司 | Method, system, equipment and storage medium for comparing video samples |
CN115359261A (en) * | 2022-10-21 | 2022-11-18 | 阿里巴巴(中国)有限公司 | Image recognition method, computer-readable storage medium, and electronic device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102231281A (en) * | 2011-07-18 | 2011-11-02 | 渤海大学 | Voice visualization method based on integration characteristic and neural network |
WO2017021322A1 (en) * | 2015-07-31 | 2017-02-09 | Eberhard Karls Universität Tübingen | Method and device for image synthesis |
CN106778928A (en) * | 2016-12-21 | 2017-05-31 | 广州华多网络科技有限公司 | Image processing method and device |
-
2017
- 2017-11-27 CN CN201711206725.6A patent/CN107886491A/en not_active Withdrawn
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102231281A (en) * | 2011-07-18 | 2011-11-02 | 渤海大学 | Voice visualization method based on integration characteristic and neural network |
WO2017021322A1 (en) * | 2015-07-31 | 2017-02-09 | Eberhard Karls Universität Tübingen | Method and device for image synthesis |
CN106778928A (en) * | 2016-12-21 | 2017-05-31 | 广州华多网络科技有限公司 | Image processing method and device |
Non-Patent Citations (1)
Title |
---|
AAYUSH BANSAL ET.AL: "PixelNN: Example-based Image Synthesis", 《ARXIV:1708.05349V1 [CS.CV]》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190684A (en) * | 2018-08-15 | 2019-01-11 | 西安电子科技大学 | SAR image sample generating method based on sketch and structural generation confrontation network |
CN109190684B (en) * | 2018-08-15 | 2022-03-04 | 西安电子科技大学 | SAR image sample generation method based on sketch and structure generation countermeasure network |
CN109361934A (en) * | 2018-11-30 | 2019-02-19 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment and storage medium |
CN109361934B (en) * | 2018-11-30 | 2021-10-08 | 腾讯科技(深圳)有限公司 | Image processing method, device, equipment and storage medium |
US11798145B2 (en) | 2018-11-30 | 2023-10-24 | Tencent Technology (Shenzhen) Company Limited | Image processing method and apparatus, device, and storage medium |
CN111798935A (en) * | 2019-04-09 | 2020-10-20 | 南京药石科技股份有限公司 | Universal compound structure-property correlation prediction method based on neural network |
CN112365533A (en) * | 2020-10-15 | 2021-02-12 | 浙江大华技术股份有限公司 | Coal flow monitoring method and device based on image segmentation and electronic device |
CN113627341A (en) * | 2021-08-11 | 2021-11-09 | 人民中科(济南)智能技术有限公司 | Method, system, equipment and storage medium for comparing video samples |
CN113627341B (en) * | 2021-08-11 | 2024-04-12 | 人民中科(济南)智能技术有限公司 | Video sample comparison method, system, equipment and storage medium |
CN115359261A (en) * | 2022-10-21 | 2022-11-18 | 阿里巴巴(中国)有限公司 | Image recognition method, computer-readable storage medium, and electronic device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Deep attention-based classification network for robust depth prediction | |
CN107886491A (en) | A kind of image combining method based on pixel arest neighbors | |
Wang et al. | Towards unified depth and semantic prediction from a single image | |
Eslami et al. | Attend, infer, repeat: Fast scene understanding with generative models | |
Pfister et al. | Deep convolutional neural networks for efficient pose estimation in gesture videos | |
CN109919209B (en) | Domain self-adaptive deep learning method and readable storage medium | |
Kondapally et al. | Towards a Transitional Weather Scene Recognition Approach for Autonomous Vehicles | |
Zhang et al. | Spike transformer: Monocular depth estimation for spiking camera | |
Bešić et al. | Dynamic object removal and spatio-temporal RGB-D inpainting via geometry-aware adversarial learning | |
Chen et al. | Unpaired deep image dehazing using contrastive disentanglement learning | |
Adate et al. | A survey on deep learning methodologies of recent applications | |
Hua et al. | Depth estimation with convolutional conditional random field network | |
Cheng et al. | S 3 Net: Semantic-aware self-supervised depth estimation with monocular videos and synthetic data | |
CN111742345A (en) | Visual tracking by coloring | |
Logacheva et al. | Deeplandscape: Adversarial modeling of landscape videos | |
Salem et al. | Semantic image inpainting using self-learning encoder-decoder and adversarial loss | |
Choo et al. | Learning background subtraction by video synthesis and multi-scale recurrent networks | |
Ben‐Zvi et al. | Line‐drawing video stylization | |
Zhu et al. | To see in the dark: N2DGAN for background modeling in nighttime scene | |
Wang et al. | A coarse-to-fine approach for dynamic-to-static image translation | |
Zhong et al. | Background subtraction driven seeds selection for moving objects segmentation and matting | |
Vijayalakshmi K et al. | Copy-paste forgery detection using deep learning with error level analysis | |
Uddin et al. | Depth guided attention for person re-identification | |
Saif et al. | Aggressive action estimation: a comprehensive review on neural network based human segmentation and action recognition | |
Metri et al. | Image generation using generative adversarial networks |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180406 |