CN110399874A - A kind of synthetic method of Car license recognition training data - Google Patents
A kind of synthetic method of Car license recognition training data Download PDFInfo
- Publication number
- CN110399874A CN110399874A CN201910671343.3A CN201910671343A CN110399874A CN 110399874 A CN110399874 A CN 110399874A CN 201910671343 A CN201910671343 A CN 201910671343A CN 110399874 A CN110399874 A CN 110399874A
- Authority
- CN
- China
- Prior art keywords
- license plate
- image
- vehicle body
- pixel
- training data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 37
- 238000010189 synthetic method Methods 0.000 title claims abstract description 21
- 238000000034 method Methods 0.000 claims abstract description 14
- 239000000203 mixture Substances 0.000 claims description 21
- 108010014172 Factor V Proteins 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 3
- 238000009472 formulation Methods 0.000 claims description 2
- 238000013136 deep learning model Methods 0.000 abstract description 10
- 239000000463 material Substances 0.000 abstract description 3
- 238000013480 data collection Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007115 recruitment Effects 0.000 description 1
- 238000004088 simulation 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of synthetic methods of Car license recognition training data, it is the following steps are included: S1. creates a background image;Each of traversal background image pixel, each pixel have random chance p1 to generate different basic factors patterns on background image;S2. a vehicle body image is selected at random from vehicle body image data base, random cropping goes out local vehicle body image from the vehicle body image of selection, merges background image and local vehicle body image to form combination picture;S3. region is randomly selected on combination picture draws license plate image.The present invention has the advantages that the synthetic method of Car license recognition training data of the invention has the advantages that economic, efficient and amount is big, it solves the problems, such as spending human and material resources in traditional manual data collection's method, inefficiency, quickly can provide unlimited training data for Car license recognition deep learning model.
Description
Technical field
The present invention relates to a kind of image composing technique, especially a kind of synthetic method of Car license recognition training data, the skill
Art is based on programming automatic generation and mark is capable of the composograph of the license plate image of analog subscriber actual photographed, is license plate text
Positioning and the training of identification deep learning model provide sufficient image data.
Background technique
Car license recognition program is based on series of computation machine algorithm, the license plate in automobile image for being shot to camera into
Row positioning and identification, are widely used in the scenes such as parking lot, highway, the Car license recognition program of mainstream is all based on depth at present
Practise algorithm.
Car license recognition program based on deep learning is a kind of deep learning model, at the beginning of program is just established, just as just going out
Raw baby, programming system inside is without any knowledge accumulation, for the license plate number in image also without any predictive ability.
It needs the later period to provide the largely license plate image data that have marked to deep learning model, makes it through continuous forward calculation and anti-
Learn the potential knowledge in image data to adjustment parameter, to obtain the ability of intelligent predicting license plate number, this process
Do training or study.Data in training process are the Knowledge Sources of deep learning model, and importance is self-evident, it may be said that
The quality of training data determines the quality of the deep learning model finally trained.
The training data of current Car license recognition deep learning model is mainly obtained by the method for manually acquiring and marking.
Substantially process is: disposing the shooting of the methods of camera ten hundreds of from complete by the certain manpower of recruitment or on road first
The different type in each province of state and the license plate image of scene;It then will be where license plate number in license plate image by data mark person
The coordinate information and license plate number of rectangle frame, which mark out, to be come.
The method of traditional artificial acquisition and mark license plate image data has many limitations: (1) needing to expend big
The human and material resources and financial resources of amount;(2) it is long to can be used for the trained license plate image data creating period;(3) the license plate figure of specific type
Picture is difficult to obtain, such as consulate's vehicle, police car, military vehicle etc..
Summary of the invention
The purpose of the present invention is to provide a kind of synthetic method of Car license recognition training data, the Technological Economy, efficiently, and
The license plate image data bulk of generation is unlimited, can provide enough data for Car license recognition deep learning model.
The purpose of the present invention is achieved through the following technical solutions: a kind of synthetic method of Car license recognition training data, it is wrapped
Include following steps:
S1. a background image img1 is created, width is w1 pixel, a height of h1 pixel;It traverses in background image img1
Each pixel, each pixel have random chance p1 to generate different basic factors patterns on background image img1;
S2. a vehicle body image is selected at random from vehicle body image data base, the random cropping from the vehicle body image of selection
Local vehicle body image img2 out, merges background image img1 and part vehicle body image img2 to form combination picture img_mix, on
The vehicle body image stated in vehicle body image data base does not contain license board information;
S3. region is randomly selected on combination picture img_mix draws license plate image.
For the prior art, the present invention has the advantages that
The synthetic method of Car license recognition training data of the invention has the advantages that economic, efficient and amount is big, solves biography
The problem of spending human and material resources in manual data collection's method of system, inefficiency, can quickly be Car license recognition deep learning
Model provides unlimited training data.In addition, program can synthesize the license plate image of the double-deck license plate, so that training the depth come
Degree learning model can identify the license plate number of bilateral license plate.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the synthetic method of Car license recognition training data of the present invention.
Fig. 2 is demonstration graph of the license plate image under two-dimensional coordinate system.
Fig. 3 is the present invention composograph generated for Car license recognition training.
Fig. 4 is the present invention composograph generated for Car license recognition training.
Fig. 5 is the present invention composograph generated for Car license recognition training.
Fig. 6 is the present invention composograph generated for Car license recognition training.
Fig. 7 is the present invention composograph generated for Car license recognition training.
Fig. 8 is the present invention composograph generated for Car license recognition training.
Fig. 9 is the present invention composograph generated for Car license recognition training.
Figure 10 is the present invention composograph generated for Car license recognition training.
Specific embodiment
The content of present invention is described in detail with embodiment with reference to the accompanying drawings of the specification:
As Fig. 1 to 10 show a kind of embodiment signal of the synthetic method of Car license recognition training data provided by the invention
Figure.
A kind of synthetic method of Car license recognition training data, it the following steps are included:
Step 1:
S1. a background image img1 is created, width is w1 pixel, a height of h1 pixel;It traverses in background image img1
Each pixel, each pixel have random chance p1 (random value of the random chance p1 value in section [0,0.002])
Different basic factors patterns is generated on background image img1;
Background image img1 is single channel image, and w1 is 512 pixels, h1 is 320 pixels, gray scale value in section [0,
255] random value in.
Basic factors pattern is drawn by the graphic plotting function in OpenCV shape library in step S1, graphic plotting
Function includes ellipse, polylines, rectangle, and when drafting randomly selects one from above three graphic plotting function
It is a.Basic factors pattern is drawn centered on above-mentioned pixel when drawing.
Ellipse, polylines, rectangle are respectively intended to draw the fundamental figures such as curve, broken line, rectangle, with mould
Curve, broken line and rectangle in quasi- reality scene.
In step S1 the size, thickness and gray scale of basic factors pattern pass through respectively factor1, factor2,
Factor3 is controlled, and factor1, factor2, factor3 are respectively the random value in different numerical intervals.
On background image img1 during drawing basics element patterns, random chance p1 plays base in control background image
In addition the effect of plinth element patterns quantity is distinguished using 3 values random controlling elements factor1, factor2, factor3
For controlling size, thickness and the gray scale of drawn basic factors pattern.
Value of 3 controlling elements when generating basic factors pattern every time is all different, and is all originated from a certain range of
Random value, the value range of 3 controlling elements are [5,600], [1,200], [0,255] respectively.
When drawing broken line, it is used to control the number on vertex in broken line using an additional controlling elements vertex_num
Amount, a random value of the vertex_num value in section [2,50].It in this way could when generating background image img1 every time
Arbitrarily complicated background image is generated, background image img1 generated can also have certain simulation to the randomness of reality scene
Ability.
Step 2:
S2. a vehicle body image is selected at random from vehicle body image data base, the random cropping from the vehicle body image of selection
Local vehicle body image img2 out, merges background image img1 and part vehicle body image img2 to form combination picture img_mix, on
The vehicle body image stated in vehicle body image data base does not contain license board information;
In step S2: the vehicle body image in vehicle body image data base is the gray level image of wide 320 pixel, high 515 pixel;From
The local vehicle body image img2 that one wide 320 pixel, high 320 pixel are cut out in the vehicle body image randomly selected, is cut
Coordinate of the upper left corner of local vehicle body image img2 on original image is (0, y), y value one in section [0,195] with
Machine value;
Background image img1 is the gray level image of wide 512 pixel, high 320 pixel, background image img1 and local vehicle body figure
When as img2 mixing, coordinate of the upper left corner img2 on img1 is (x, 0), and x value one in section [0,192] random
Value.
Combination picture img_mix is obtained by following mixed formulation in step S2:
Img_mix=img1*rate+img2* (1-rate), wherein rate value one in section [0,1] it is random
Value.
Vehicle body image in vehicle body image data base is free of the vehicle body image of license plate using 2000 in ccpd data set
Data set, their original size are all the color images of wide 720 pixel, high 1160 pixel.Here they are all reduced and is turned
It is changed to the gray level image of wide 320 pixel, high 515 pixel.
Step 3:
S3. region is randomly selected on combination picture img_mix draws license plate image.
Step S3 includes following sub-step:
S3-1: license plate image is divided into three component parts, respectively license plate background, license plate frame and license plate number;
S3-2: the gray value of license plate background, license plate frame and license plate number three is determined;
Random value of the gray scale of license plate background, the license plate frame and license plate number all values in section [0,255], but phase
Between to guarantee license plate background and license plate number gray scale difference value absolute value be greater than 20, to guarantee license plate number and license plate background not
Can because gray value is too similar mix it is indistinguishable.
S3-3: simultaneously corresponding license plate number is randomly generated according to selected license plate type in selected license plate type;
Randomly choosing a license plate type first, (the license plate type supported at present has blue board, yellow card, green board, person who is not a member of any political party and Hei
Board), license plate number random corresponding to the license plate type is then generated, such as blue board, the license plate number of generation
It can be " Fujian A698KH ".
S3-4: license plate background, the size of license plate frame and license plate number and relative position are determined;
Detailed process is as follows by step S3-4:
First select the font size plate_size in license plate number;License plate is calculated according to font size plate_size
Rectangle surrounds the wide w2, high h2 of frame where number;And then the width for calculating license plate background and license plate frame is w2+2*delta,
A height of h2+2*delta;Random value of the line thickness value of last license plate frame in specific sections;Wherein delta=h2*
Factor5, factor5=0.2.
Since license plate number is located at the middle of license plate background and license plate frame, license plate background and license plate frame and license plate number
The relative position of code is constant, meets following formula through measurement discovery license plate:
H2/delta=factor5=0.2
So rectangle surrounds the wide w2 and high h2 of frame where determining font size plate_size you can get it license plate number,
And then obtain the width and height of license plate background and license plate frame.
Random value of the font size plate_size value in section [15,70], unit is pixel.
Random value of the line thickness value of license plate frame in section [1,6].
S3-5: the position of license plate background, license plate frame and license plate number on combination picture img_mix is determined;
Since the relative position of license plate background and license plate frame and license plate number is constant, so only being needed here for license plate number
Code one random position coordinates of distribution.But due to the needs of Car license recognition model, also there is limitation: license plate here
The distance on up and down four boundary of the number apart from combination picture img_mix cannot be less than padding (padding value
For 10 pixels).
So the x value random value of license plate number top left co-ordinate is in section [padding, w1-padding-w2], y value
Random value is in section [padding, h1-padding-h2].
Detailed process is as follows by step S3-5:
Assuming that the random value that license plate number top left co-ordinate is taken is (x1, y1), then its bottom right angular coordinate is (x1+w2, y1
+ h2), the top left co-ordinate of license plate background and license plate frame is (x1-delta, y1-delta), and bottom right angular coordinate is (x1+w2+
Delta, y1+h2+delta).
Wherein, x1+w2=x2, y1+h2=y2, x1-delta=x3, y1-delta=y4, x1+w2+delta=x4, y1
+ h2+delta=y4.
S3-6: license plate background, license plate frame and license plate number are plotted on combination picture img_mix.
After getting crucial parameter by above-mentioned several steps, so that it may use text () function of Pillow shape library
License plate number, license plate background and license plate frame are plotted to combination picture with rectangle () function of OpenCV shape library
On img_mix.
Step 4:
The Gaussian mode of Random-fuzzy degree is carried out to the combination picture img_mix for being painted with license plate image obtained in step S3
Paste and Motion Blur processing, to simulate the blur effect of captured image in reality scene.
Step 5:
The perspective transform that probability is p2 is carried out to composograph obtained by step S4, composograph is in x, y, z in perspective transform
The range that perspective view on three dimension directions changes at random is [- 30,30], and vehicle is shot in reality scene to simulate with this
The angle that may change when board.So far just complete an opening and closing at black and white license plate image.
What the present invention finally synthesized is black white image, and compared to color image, smaller black white image data volume (is colored
The 1/3 of rgb image data amount), and black white image will not lose license plate number information, and this helps speed up deep learning model
Processing speed.
Claims (10)
1. a kind of synthetic method of Car license recognition training data, which is characterized in that it the following steps are included:
S1. a background image img1 is created, width is w1 pixel, a height of h1 pixel;It traverses each in background image img1
A pixel, each pixel have random chance p1 to generate different basic factors patterns on background image img1;
S2. a vehicle body image is selected at random from vehicle body image data base, random cropping is out from the vehicle body image of selection
Portion vehicle body image img2 merges background image img1 and part vehicle body image img2 to form combination picture img_mix, above-mentioned vehicle
Vehicle body image in body image data base does not contain license board information;
S3. region is randomly selected on combination picture img_mix draws license plate image.
2. a kind of synthetic method of Car license recognition training data according to claim 1, it is characterised in that: base in step S1
Plinth element patterns are drawn by the graphic plotting function in OpenCV shape library, graphic plotting function include ellipse,
Polylines, rectangle, when drafting, randomly select one from above three graphic plotting function.
3. a kind of synthetic method of Car license recognition training data according to claim 2, it is characterised in that: base in step S1
Size, thickness and the gray scale of plinth element patterns pass through factor1, factor2, factor3 respectively and are controlled, factor1,
Factor2, factor3 are respectively the random value in different numerical intervals.
4. a kind of synthetic method of Car license recognition training data according to claim 1, which is characterized in that in step S2:
Vehicle body image in vehicle body image data base is the gray level image of wide 320 pixel, high 515 pixel;
The local vehicle body image img2 of one wide 320 pixel, high 320 pixel, institute are cut out from the vehicle body image randomly selected
Coordinate of the upper left corner of the local vehicle body image img2 of cutting on original image is (0, y), and y value is in section [0,195]
One random value;
Background image img1 is the gray level image of wide 512 pixel, high 320 pixel, background image img1 and local vehicle body image
When img2 is mixed, coordinate of the upper left corner img2 on img1 is (x, 0), a random value of the x value in section [0,192].
5. a kind of synthetic method of Car license recognition training data according to claim 1, which is characterized in that multiple in step S2
Image img_mix is closed to obtain by following mixed formulation:
Img_mix=img1*rate+img2* (1-rate), wherein a random value of the rate value in section [0,1].
6. a kind of synthetic method of Car license recognition training data described in -5 any one according to claim 1, which is characterized in that
Step S3 includes following sub-step:
S3-1: license plate image is divided into three component parts, respectively license plate background, license plate frame and license plate number;
S3-2: the gray value of license plate background, license plate frame and license plate number three is determined;
S3-3: simultaneously corresponding license plate number is randomly generated according to selected license plate type in selected license plate type;
S3-4: license plate background, the size of license plate frame and license plate number and relative position are determined;
S3-5: the position of license plate background, license plate frame and license plate number on combination picture img_mix is determined;
S3-6: license plate background, license plate frame and license plate number are plotted on combination picture img_mix.
7. a kind of synthetic method of Car license recognition training data according to claim 6, which is characterized in that step S3-4 tool
Body process is as follows:
First select the font size plate_size in license plate number;License plate number is calculated according to font size plate_size
The wide w2, high h2 of place rectangle encirclement frame;And then the width for calculating license plate background and license plate frame is w2+2*delta, it is a height of
h2+2*delta;Random value of the line thickness value of last license plate frame in specific sections;Wherein delta=h2*
Factor5, factor5=0.2.
8. a kind of synthetic method of Car license recognition training data according to claim 7, which is characterized in that step S3-5 tool
Body process is as follows:
Assuming that the random value that license plate number top left co-ordinate is taken is (x1, y1), then its bottom right angular coordinate is (x1+w2, y1+
H2), the top left co-ordinate of license plate background and license plate frame is (x1-delta, y1-delta), and bottom right angular coordinate is (x1+w2+
Delta, y1+h2+delta).
9. a kind of synthetic method of Car license recognition training data according to claim 6, which is characterized in that it further includes step
Rapid S4: the Gaussian Blur of Random-fuzzy degree is carried out to the combination picture img_mix for being painted with license plate image obtained in step S3
With Motion Blur processing.
10. a kind of synthetic method of Car license recognition training data according to claim 6, which is characterized in that it further includes
Step S5: the perspective transform that probability is p2 is carried out to composograph obtained by step S4, composograph is in x, y, z three in perspective transform
The range that perspective view on a dimension direction changes at random is [- 30,30].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910671343.3A CN110399874A (en) | 2019-07-24 | 2019-07-24 | A kind of synthetic method of Car license recognition training data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910671343.3A CN110399874A (en) | 2019-07-24 | 2019-07-24 | A kind of synthetic method of Car license recognition training data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110399874A true CN110399874A (en) | 2019-11-01 |
Family
ID=68325870
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910671343.3A Pending CN110399874A (en) | 2019-07-24 | 2019-07-24 | A kind of synthetic method of Car license recognition training data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110399874A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111275796A (en) * | 2020-01-17 | 2020-06-12 | 北京迈格威科技有限公司 | License plate synthesis method and device, computer equipment and storage medium |
CN111369518A (en) * | 2020-02-28 | 2020-07-03 | 创新奇智(合肥)科技有限公司 | Sample expansion method and device, electronic equipment and readable storage medium |
CN111508045A (en) * | 2020-03-12 | 2020-08-07 | 深兰人工智能芯片研究院(江苏)有限公司 | Picture synthesis method and device |
CN111651512A (en) * | 2020-05-27 | 2020-09-11 | 福建博思软件股份有限公司 | Multisource heterogeneous commodity feature weight solving method and device based on semi-supervised learning |
CN114999051A (en) * | 2022-06-16 | 2022-09-02 | 广州市懒人时代信息科技有限公司 | Smart community platform security monitoring system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504744A (en) * | 2014-12-31 | 2015-04-08 | 深圳市捷顺科技实业股份有限公司 | License plate image composition simulating method and device |
US20160203380A1 (en) * | 2015-01-13 | 2016-07-14 | Xerox Corporation | Annotation free license plate recognition method and system |
CN109147011A (en) * | 2018-08-27 | 2019-01-04 | 平安科技(深圳)有限公司 | License plate image generation method, device, computer equipment and storage medium |
CN109325989A (en) * | 2018-08-27 | 2019-02-12 | 平安科技(深圳)有限公司 | License plate image generation method, device, equipment and medium |
CN109377467A (en) * | 2018-09-28 | 2019-02-22 | 阿里巴巴集团控股有限公司 | Generation method, object detection method and the device of training sample |
-
2019
- 2019-07-24 CN CN201910671343.3A patent/CN110399874A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504744A (en) * | 2014-12-31 | 2015-04-08 | 深圳市捷顺科技实业股份有限公司 | License plate image composition simulating method and device |
US20160203380A1 (en) * | 2015-01-13 | 2016-07-14 | Xerox Corporation | Annotation free license plate recognition method and system |
CN109147011A (en) * | 2018-08-27 | 2019-01-04 | 平安科技(深圳)有限公司 | License plate image generation method, device, computer equipment and storage medium |
CN109325989A (en) * | 2018-08-27 | 2019-02-12 | 平安科技(深圳)有限公司 | License plate image generation method, device, equipment and medium |
CN109377467A (en) * | 2018-09-28 | 2019-02-22 | 阿里巴巴集团控股有限公司 | Generation method, object detection method and the device of training sample |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111275796A (en) * | 2020-01-17 | 2020-06-12 | 北京迈格威科技有限公司 | License plate synthesis method and device, computer equipment and storage medium |
CN111275796B (en) * | 2020-01-17 | 2023-09-19 | 北京迈格威科技有限公司 | License plate synthesis method, license plate synthesis device, computer equipment and storage medium |
CN111369518A (en) * | 2020-02-28 | 2020-07-03 | 创新奇智(合肥)科技有限公司 | Sample expansion method and device, electronic equipment and readable storage medium |
CN111369518B (en) * | 2020-02-28 | 2022-11-01 | 创新奇智(合肥)科技有限公司 | Sample expansion method and device, electronic equipment and readable storage medium |
CN111508045A (en) * | 2020-03-12 | 2020-08-07 | 深兰人工智能芯片研究院(江苏)有限公司 | Picture synthesis method and device |
CN111651512A (en) * | 2020-05-27 | 2020-09-11 | 福建博思软件股份有限公司 | Multisource heterogeneous commodity feature weight solving method and device based on semi-supervised learning |
CN111651512B (en) * | 2020-05-27 | 2022-06-10 | 福建博思软件股份有限公司 | Multisource heterogeneous commodity feature weight solving method and device based on semi-supervised learning |
CN114999051A (en) * | 2022-06-16 | 2022-09-02 | 广州市懒人时代信息科技有限公司 | Smart community platform security monitoring system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110399874A (en) | A kind of synthetic method of Car license recognition training data | |
Benson et al. | Synthesising continuous-tone caricatures | |
CN111539273B (en) | Traffic video background modeling method and system | |
CN101061518B (en) | Flexible antialiasing in embedded devices | |
DE102018121019A1 (en) | EXPANDING REAL SENSOR RECORDS WITH SIMULATED SENSOR DATA | |
CN111047506B (en) | Environmental map generation and hole filling | |
CN109711246B (en) | Dynamic object recognition method, computer device and readable storage medium | |
CN109325437A (en) | Image processing method, device and system | |
CN110163640A (en) | A kind of method and computer equipment of product placement in video | |
CN110291358A (en) | Shape estimation device | |
US20200302579A1 (en) | Environment map generation and hole filling | |
CN110310315A (en) | Network model training method, device and object pose determine method, apparatus | |
US9956717B2 (en) | Mapping for three dimensional surfaces | |
US9558406B2 (en) | Image processing apparatus including an object setting section, image processing method, and program using the same | |
CN104751407A (en) | Method and device used for blurring image | |
Paulin et al. | Review and analysis of synthetic dataset generation methods and techniques for application in computer vision | |
CN113763569A (en) | Image annotation method and device used in three-dimensional simulation and electronic equipment | |
CN103593639A (en) | Lip detection and tracking method and device | |
CN113144613A (en) | Model-based volume cloud generation method | |
CN114692034A (en) | Image display method, image display device, electronic equipment and computer storage medium | |
DE10250602B4 (en) | System for generating a synthetic scene | |
CN113126944B (en) | Depth map display method, display device, electronic device, and storage medium | |
CN113240573B (en) | High-resolution image style transformation method and system for local and global parallel learning | |
CN113436279B (en) | Image processing method, device and equipment | |
CN112697159B (en) | Map editing method and system |
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 |