CN110991488B - Picture watermark identification method using deep learning model - Google Patents
Picture watermark identification method using deep learning model Download PDFInfo
- Publication number
- CN110991488B CN110991488B CN201911090218.XA CN201911090218A CN110991488B CN 110991488 B CN110991488 B CN 110991488B CN 201911090218 A CN201911090218 A CN 201911090218A CN 110991488 B CN110991488 B CN 110991488B
- Authority
- CN
- China
- Prior art keywords
- watermark
- pictures
- picture
- training
- classification
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/09—Recognition of logos
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a picture watermark identification method using a deep learning model, which comprises the following steps: s1: dividing the pictures into 80 lists according to different manual categories; s2: 1000 pictures are added to the 80 lists to serve as training corpus, and the total number of pictures is 8 ten thousand; s3: carrying out data enhancement on the picture library of 8 ten thousand pictures added in the step S2 to obtain a sample picture library; s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue. Compared with the traditional method for manually checking the picture watermark, the method has the advantages that the design is reasonable, the manual intensity can be greatly reduced, the error caused by manual work can not occur, and the accurate identification of the picture with the watermark is improved.
Description
Technical Field
The invention relates to the technical field of image detection, in particular to a picture watermark identification method using a deep learning model.
Background
The images contain rich information in multiple dimensions, and in the field of mobile internet news, there is a constant need for a large number of images and text to deliver useful information to users. Because the internet information is fast and convenient to spread, more and more copyright holders and organizations embed their logo watermarks into image information to form copyrighted pictures, so that the copyrighted pictures can be protected, or the copyrights can be used for business, publishing, exhibition and other purposes. Therefore, the news client needs to audit the image information of each news and identify whether the image contains watermark information, so that misuse and infringement are avoided.
On the one hand, the manuscripts produced by the cooperative media every day are more and more, the image information is more and more, the daily news produced reaches 50 tens of thousands, and the quantity of the news is far beyond the limit of manual auditing. On the other hand, watermark information is difficult to identify at a glance in an image, and the watermark information has the characteristics of small area, light color, translucency, transparency and the like, so that the difference between the watermarked image and the non-watermarked image is very small, and the degree of distinction is low. Therefore, accurate identification of the picture with the watermark is difficult to realize in a manual auditing mode, so we propose a picture watermark identification method using a deep learning model to solve the above-mentioned problems.
Disclosure of Invention
Based on the technical problems in the background technology, the invention provides a picture watermark identification method using a deep learning model.
The invention provides a picture watermark identification method using a deep learning model, which comprises the following steps:
s1: dividing the pictures into 80 lists according to different manual categories;
s2: 1000 pictures are added to the 80 lists to serve as training corpus, and the total number of pictures is 8 ten thousand;
s3: carrying out data enhancement on the picture library of 8 ten thousand pictures added in the step S2 to obtain a sample picture library;
s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue;
s5: carrying out watermark identification training on the sample picture library in the S3 and the watermark pictures in the S4 to obtain 81 basic bone trunk models, and editing the positions and the classifications of the watermark marked in the management background;
s6: placing the marked watermark pictures together with the enhanced data set of the watermark pictures into a trunk model for training, adding the classification of the watermark pictures, and obtaining an identification watermark model containing watermark types after training;
s7: the trained watermark model is deployed on a server of tensorsurface-gpu, and whether the picture contains the watermark can be identified through the picture address.
Preferably, in the step S1, the specific classifications of the 80 lists are: people, backbags, umbrellas, handbags, ties, suitcases, bicycles, automobiles, motorcycles, airplanes, buses, training, trucks, boats, traffic lights, fire hydrants, stop signs, parking meters, benches, birds, cats, dogs, horses, sheep, cattle, elephants, bears, zebra, giraffes, flying discs, snowboards, sport balls, kites, baseball bats, baseball gloves, skateboards, surfboards, tennis rackets, bottles, wine glass, cups, forks, knives, spoons, bowls, bananas, apples, sandwiches, oranges, broccoli, carrots, hotdogs, pizza, donuts, cakes, chairs, benches, pot, beds, dining tables, toilets, televisions, notebook computers, mice, remote, keyboards, cell phones, microwaves, ovens, toasters, sinks, refrigerators, books, clocks, vases, scissors, teddy bear, blowers, toothbrushes.
Preferably, in the step S2, the data enhancement mode includes affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, clipping/expanding, blurring, and the like, so as to obtain a sample picture library after enhancing the picture with various samples.
Preferably, in the step S3, the data enhancement mode includes affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, clipping/expanding, blurring, and the like, so as to obtain a sample picture library after enhancing the picture with various samples.
Preferably, in S4, the management background is an independently developed management background, so that the unified operation of an operator is facilitated.
Preferably, in the step S5, after the watermark picture including the new classification is trained and stored, the classification of the main stem model is increased by 1 class, at this time, the classification of the main stem model is increased to 81, the picture recognition process of the server recognizes that the main stem model is changed, and then the first process is automatically restarted, and then the picture watermark request is processed.
Preferably, in the step S6, the process of training the trunk model is to first perform pre-training with the enhanced sample picture library, where the pre-training is to perform supervised classification training. Then a specific tuning (fine tuning) is performed on the samples, 25% of positive samples (candidate boxes 0.5-1 with real box IoU) and 75% of negative samples (candidate boxes 0.1-0.5 with real box IoU) in the tuned dataset.
Preferably, during tuning training, N complete pictures are added to each mini-batch first, then R candidate frames selected from the N pictures are added, the R candidate frames can multiplex network features of the first 5 stages of the N pictures, n=2 and r=128 in the article, reference data of the server is that a Ubuntu 18.04 operating system, a 64G memory, a CPU 2620v4 x 2, a display card 1080TI x 2, a hard disk SSD 500G x 2+sata, a 2t x 2 and cuda 10.0 are installed in the server.
Compared with the prior art, the method has the advantages that the main rod model and the watermark picture adding mode are adopted for training, the main rod model improves the classification accuracy through the set pre-training, and the watermark recognition efficiency is improved.
Compared with the traditional method for manually checking the picture watermark, the method has the advantages that the design is reasonable, the manual intensity can be greatly reduced, the error caused by manual work can not occur, and the accurate identification of the picture with the watermark is improved.
Detailed Description
The invention is further illustrated below in connection with specific embodiments.
Example 1
In this embodiment, a picture watermark identifying method using a deep learning model is provided, including the following steps:
s1: dividing the pictures into 80 lists according to different manual categories;
s2: 1000 pictures are added to the 80 lists to serve as training corpus, and the total number of pictures is 8 ten thousand;
s3: carrying out data enhancement on the picture library of 8 ten thousand pictures added in the step S2 to obtain a sample picture library;
s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue;
s5: carrying out watermark identification training on the sample picture library in the S3 and the watermark pictures in the S4 to obtain 81 basic bone trunk models, and editing the positions and the classifications of the watermark marked in the management background;
s6: placing the marked watermark pictures together with the enhanced data set of the watermark pictures into a trunk model for training, adding the classification of the watermark pictures, and obtaining an identification watermark model containing watermark types after training;
s7: the trained watermark model is deployed on a server of tensorsurface-gpu, and whether the picture contains the watermark can be identified through the picture address.
In this embodiment, in S1, the specific classifications of the 80 lists are: people, backbags, umbrellas, handbags, ties, suitcases, bicycles, automobiles, motorcycles, airplanes, buses, training, trucks, boats, traffic lights, fire hydrants, stop signs, parking meters, benches, birds, cats, dogs, horses, sheep, cattle, elephants, bears, zebra, giraffes, flying disks, snowboards, sport balls, kites, baseball bats, baseball gloves, skateboards, surfboards, tennis rackets, bottles, wineglass, cups, forks, knives, spoons, bowls, bananas, apples, sandwiches, oranges, broccoli, carrots, hotdogs, pizza, donuts, cakes, chairs, benches, pot, beds, dining tables, toilets, televisions, notebook computers, mice, remote, keyboards, cell phones, microwaves, ovens, toasters, sinks, refrigerators, books, clocks, vases, scissors, teddy bear, blowers, toothbrushes, in S2, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S3, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S4, a management background is an autonomously developed management background, which is convenient for an operator to uniformly operate, in S5, after training and saving watermark pictures containing new classifications, the recognition classification of a main model is increased by 1 class, at this time, the recognition classification of the main model is increased by 81, the picture recognition process of a server can recognize that the main model is changed, and the first process can be restarted automatically, then, the picture watermark request is processed, and in S6, the process of training the trunk model is to first perform pre-training by using the enhanced sample picture library, wherein the pre-training is to perform supervised classification training. Then, specific tuning (fine tuning) is performed on the samples, 25% of positive samples (candidate frames with real frames IoU at 0.5-1) and 75% of negative samples (candidate frames with real frames IoU at 0.1-0.5) in a tuning data set are performed, during tuning training, N complete pictures are added to each mini-batch first, then R candidate frames selected from the N pictures are added, the R candidate frames can multiplex network characteristics of the first 5 stages of the N pictures, n=2, r=128 in the article, reference data of the server is that the server is provided with Ubuntu 18.04 operating system, 64G memory, CPU 26200 v4 x 2, graphic card 1080TI x 2, hard disk SSD 500G x 2+sata, 2t x 2 and cuda 10.0.
Example two
In this embodiment, a picture watermark identifying method using a deep learning model is provided, including the following steps:
s1: dividing the pictures into 80 lists according to different manual categories;
s2: 2000 pictures are added to the 80 lists as training corpus, and the total number of the pictures is 16 ten thousand;
s3: carrying out data enhancement on the picture library of the 16 ten thousand pictures added in the S2 to obtain a sample picture library;
s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue;
s5: carrying out watermark identification training on the sample picture library in the S3 and the watermark pictures in the S4 to obtain 81 basic bone trunk models, and editing the positions and the classifications of the watermark marked in the management background;
s6: placing the marked watermark pictures together with the enhanced data set of the watermark pictures into a trunk model for training, adding the classification of the watermark pictures, and obtaining an identification watermark model containing watermark types after training;
s7: the trained watermark model is deployed on a server of tensorsurface-gpu, and whether the picture contains the watermark can be identified through the picture address.
In this embodiment, in S1, the specific classifications of the 80 lists are: people, backbags, umbrellas, handbags, ties, suitcases, bicycles, automobiles, motorcycles, airplanes, buses, training, trucks, boats, traffic lights, fire hydrants, stop signs, parking meters, benches, birds, cats, dogs, horses, sheep, cattle, elephants, bears, zebra, giraffes, flying disks, snowboards, sport balls, kites, baseball bats, baseball gloves, skateboards, surfboards, tennis rackets, bottles, wineglass, cups, forks, knives, spoons, bowls, bananas, apples, sandwiches, oranges, broccoli, carrots, hotdogs, pizza, donuts, cakes, chairs, benches, pot, beds, dining tables, toilets, televisions, notebook computers, mice, remote, keyboards, cell phones, microwaves, ovens, toasters, sinks, refrigerators, books, clocks, vases, scissors, teddy bear, blowers, toothbrushes, in S2, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S3, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S4, a management background is an autonomously developed management background, which is convenient for an operator to uniformly operate, in S5, after training and saving watermark pictures containing new classifications, the recognition classification of a main model is increased by 1 class, at this time, the recognition classification of the main model is increased by 81, the picture recognition process of a server can recognize that the main model is changed, and the first process can be restarted automatically, then, the picture watermark request is processed, and in S6, the process of training the trunk model is to first perform pre-training by using the enhanced sample picture library, wherein the pre-training is to perform supervised classification training. Then, specific tuning (fine tuning) is performed on the samples, 25% of positive samples (candidate frames with real frames IoU at 0.5-1) and 75% of negative samples (candidate frames with real frames IoU at 0.1-0.5) in a tuning data set are performed, during tuning training, N complete pictures are added to each mini-batch first, then R candidate frames selected from the N pictures are added, the R candidate frames can multiplex network characteristics of the first 5 stages of the N pictures, n=2, r=128 in the article, reference data of the server is that the server is provided with Ubuntu 18.04 operating system, 64G memory, CPU 26200 v4 x 2, graphic card 1080TI x 2, hard disk SSD 500G x 2+sata, 2t x 2 and cuda 10.0.
Example III
In this embodiment, a picture watermark identifying method using a deep learning model is provided, including the following steps:
s1: dividing the pictures into 80 lists according to different manual categories;
s2: 3000 pictures are added to the 80 lists as training corpus, and the total number of pictures is 24 ten thousand;
s3: carrying out data enhancement on the picture library of the 24 ten thousand pictures added in the S2 to obtain a sample picture library;
s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue;
s5: carrying out watermark identification training on the sample picture library in the S3 and the watermark pictures in the S4 to obtain 81 basic bone trunk models, and editing the positions and the classifications of the watermark marked in the management background;
s6: placing the marked watermark pictures together with the enhanced data set of the watermark pictures into a trunk model for training, adding the classification of the watermark pictures, and obtaining an identification watermark model containing watermark types after training;
s7: the trained watermark model is deployed on a server of tensorsurface-gpu, and whether the picture contains the watermark can be identified through the picture address.
In this embodiment, in S1, the specific classifications of the 80 lists are: people, backbags, umbrellas, handbags, ties, suitcases, bicycles, automobiles, motorcycles, airplanes, buses, training, trucks, boats, traffic lights, fire hydrants, stop signs, parking meters, benches, birds, cats, dogs, horses, sheep, cattle, elephants, bears, zebra, giraffes, flying disks, snowboards, sport balls, kites, baseball bats, baseball gloves, skateboards, surfboards, tennis rackets, bottles, wineglass, cups, forks, knives, spoons, bowls, bananas, apples, sandwiches, oranges, broccoli, carrots, hotdogs, pizza, donuts, cakes, chairs, benches, pot, beds, dining tables, toilets, televisions, notebook computers, mice, remote, keyboards, cell phones, microwaves, ovens, toasters, sinks, refrigerators, books, clocks, vases, scissors, teddy bear, blowers, toothbrushes, in S2, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S3, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, shearing/expanding, blurring and the like, so as to obtain a sample picture library after enhancing the sample diversity, in S4, a management background is an autonomously developed management background, which is convenient for an operator to uniformly operate, in S5, after training and saving watermark pictures containing new classifications, the recognition classification of a main model is increased by 1 class, at this time, the recognition classification of the main model is increased by 81, the picture recognition process of a server can recognize that the main model is changed, and the first process can be restarted automatically, then, the picture watermark request is processed, and in S6, the process of training the trunk model is to first perform pre-training by using the enhanced sample picture library, wherein the pre-training is to perform supervised classification training. Then, specific tuning (fine tuning) is performed on the samples, 25% of positive samples (candidate frames with real frames IoU at 0.5-1) and 75% of negative samples (candidate frames with real frames IoU at 0.1-0.5) in a tuning data set are performed, during tuning training, N complete pictures are added to each mini-batch first, then R candidate frames selected from the N pictures are added, the R candidate frames can multiplex network characteristics of the first 5 stages of the N pictures, n=2, r=128 in the article, reference data of the server is that the server is provided with Ubuntu 18.04 operating system, 64G memory, CPU 26200 v4 x 2, graphic card 1080TI x 2, hard disk SSD 500G x 2+sata, 2t x 2 and cuda 10.0.
It can be derived that the second embodiment is the optimal recognition condition for the picture watermark.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should make equivalent substitutions or modifications according to the technical scheme of the present invention and the inventive concept thereof, and should be covered by the scope of the present invention.
Claims (1)
1. A picture watermark recognition method using a deep learning model, comprising the steps of:
s1: dividing the pictures into 80 lists according to different manual categories;
s2: 1000 pictures are added to the 80 lists to serve as training corpus, and the total number of pictures is 8 ten thousand;
s3: carrying out data enhancement on the picture library of 8 ten thousand pictures added in the step S2 to obtain a sample picture library;
s4: the operator uploads related watermark pictures in the management background, a plurality of watermark positions and classification of the watermark can be uploaded, the operator selects an automatic training function in the management background, and the management background can transmit the position information of the watermark pictures and the classification of the watermark to a processing queue;
s5: carrying out watermark identification training on the sample picture library in the S3 and the watermark pictures in the S4 to obtain 81 basic bone trunk models, and editing the positions and the classifications of the watermark marked in the management background;
s6: placing the marked watermark pictures together with the enhanced data set of the watermark pictures into a trunk model for training, adding the classification of the watermark pictures, and obtaining an identification watermark model containing watermark types after training;
s7: the trained watermark model is deployed on a server of tensorsurface-gpu, and whether the picture contains the watermark or not can be identified through a picture address;
in the step S1, the specific classifications of the 80 lists are: people, backbags, umbrellas, handbags, ties, suitcases, bicycles, automobiles, motorcycles, airplanes, buses, training, trucks, boats, traffic lights, fire hydrants, stop signs, parking meters, benches, birds, cats, dogs, horses, sheep, cattle, elephants, bears, zebra, giraffes, flying disks, snowboards, sport balls, kites, baseball bats, baseball gloves, skateboards, surfboards, tennis rackets, bottles, wineglass, cups, forks, knives, spoons, bowls, bananas, apples, sandwiches, oranges, broccoli, carrots, hotdogs, pizza, donuts, cakes, chairs, benches, pot, beds, dining tables, toilets, televisions, notebook computers, mice, remote, keyboards, cell phones, microwaves, ovens, toasters, sinks, refrigerators, books, clocks, vases, scissors, teddy bear, blowers, toothbrushes, in the step S3, the data enhancement mode comprises affine transformation, perspective transformation, contrast adjustment, gaussian noise, region discarding, cutting/expanding, blurring and the like, a sample picture library after sample diversity picture enhancement is obtained, in the step S4, a management background is an independently developed management background which is convenient for an operator to operate uniformly, in the step S5, after a watermark picture containing new classification is trained and stored, the identification classification of a trunk model is increased by 1 class, at the moment, the identification classification of the trunk model is increased by 81, the picture identification process of a server can identify that the trunk model is changed, the first process can be restarted automatically, then a picture watermark request is processed, in the step S6, the process of training the trunk model is to pretrain with the enhanced sample picture library, pretraining is to conduct supervised classified training, then specific tuning is conducted on the sample, during tuning training, firstly adding N complete pictures into each mini-batch, then adding R candidate frames selected from the N pictures, wherein the R candidate frames can multiplex network characteristics of the first 5 stages of the N pictures, N=2 and R=128, reference data of a server are that the server is provided with a Ubuntu 18.04 operating system, a 64G memory, a CPU 260V4x2, a display card 1080TI x 2, a hard disk 500G x 2+SATA, 2T x 2 and cuda 10.0.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911090218.XA CN110991488B (en) | 2019-11-08 | 2019-11-08 | Picture watermark identification method using deep learning model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911090218.XA CN110991488B (en) | 2019-11-08 | 2019-11-08 | Picture watermark identification method using deep learning model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110991488A CN110991488A (en) | 2020-04-10 |
CN110991488B true CN110991488B (en) | 2023-10-20 |
Family
ID=70083870
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911090218.XA Active CN110991488B (en) | 2019-11-08 | 2019-11-08 | Picture watermark identification method using deep learning model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110991488B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113837914A (en) * | 2020-06-08 | 2021-12-24 | 北京金山办公软件股份有限公司 | Watermark identification method and system based on artificial intelligence |
CN113222804B (en) * | 2021-06-02 | 2022-03-15 | 景德镇陶瓷大学 | Ceramic process-oriented up-sampling ceramic watermark model training method and embedding method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术(上海)有限公司 | The recognition methods of watermarked image and the system of identification |
CN107808358A (en) * | 2017-11-13 | 2018-03-16 | 携程计算机技术(上海)有限公司 | Image watermark automatic testing method |
CN107818321A (en) * | 2017-10-13 | 2018-03-20 | 上海眼控科技股份有限公司 | A kind of watermark date recognition method for vehicle annual test |
WO2019067730A1 (en) * | 2017-09-29 | 2019-04-04 | Digimarc Corporation | Watermark sensing methods and arrangements |
CN109598231A (en) * | 2018-12-03 | 2019-04-09 | 广州市百果园信息技术有限公司 | A kind of recognition methods of video watermark, device, equipment and storage medium |
CN109784181A (en) * | 2018-12-14 | 2019-05-21 | 平安科技(深圳)有限公司 | Picture watermark recognition methods, device, equipment and computer readable storage medium |
CN109815653A (en) * | 2019-01-31 | 2019-05-28 | 上海计算机软件技术开发中心 | A kind of extraction of pdf Text Watermarking and comparison method based on deep learning |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190213705A1 (en) * | 2017-12-08 | 2019-07-11 | Digimarc Corporation | Artwork generated to convey digital messages, and methods/apparatuses for generating such artwork |
-
2019
- 2019-11-08 CN CN201911090218.XA patent/CN110991488B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术(上海)有限公司 | The recognition methods of watermarked image and the system of identification |
WO2019067730A1 (en) * | 2017-09-29 | 2019-04-04 | Digimarc Corporation | Watermark sensing methods and arrangements |
CN107818321A (en) * | 2017-10-13 | 2018-03-20 | 上海眼控科技股份有限公司 | A kind of watermark date recognition method for vehicle annual test |
CN107808358A (en) * | 2017-11-13 | 2018-03-16 | 携程计算机技术(上海)有限公司 | Image watermark automatic testing method |
CN109598231A (en) * | 2018-12-03 | 2019-04-09 | 广州市百果园信息技术有限公司 | A kind of recognition methods of video watermark, device, equipment and storage medium |
CN109784181A (en) * | 2018-12-14 | 2019-05-21 | 平安科技(深圳)有限公司 | Picture watermark recognition methods, device, equipment and computer readable storage medium |
CN109815653A (en) * | 2019-01-31 | 2019-05-28 | 上海计算机软件技术开发中心 | A kind of extraction of pdf Text Watermarking and comparison method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN110991488A (en) | 2020-04-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Arefieva et al. | A machine learning approach to cluster destination image on Instagram | |
US20220103505A1 (en) | Social media influence of geographic locations | |
Kirilenko et al. | Sochi 2014 Olympics on Twitter: Perspectives of hosts and guests | |
US10356027B2 (en) | Location resolution of social media posts | |
US20170109615A1 (en) | Systems and Methods for Automatically Classifying Businesses from Images | |
US20150161529A1 (en) | Identifying Related Events for Event Ticket Network Systems | |
CN104598602B (en) | Pass through computer implemented information recommendation method and device based on scene | |
CN110555107B (en) | Method and device for determining service object theme and service object recommendation | |
Nie et al. | Multimodal learning toward micro-video understanding | |
US20170255625A1 (en) | Computer-implemented method for providing multimedia content and device | |
CN111259192A (en) | Audio recommendation method and device | |
CN102436483A (en) | Video advertisement detection method based on explicit sharing subspace | |
Serra-Cantallops et al. | Host community resignation to nightclub tourism | |
CN110991488B (en) | Picture watermark identification method using deep learning model | |
CN113329261B (en) | Video processing method and device | |
CN110489649B (en) | Method and device for associating content with tag | |
CN113810742A (en) | Virtual gift processing method and device, electronic equipment and storage medium | |
Oliveira et al. | Strategic communication for non-profit organisations: challenges and alternative approaches | |
Amengual et al. | Review of methods to predict social image interestingness and memorability | |
Lonn et al. | Smartphone picture organization: A hierarchical approach | |
Carah et al. | Tuning machines: an approach to exploring how Instagram’s machine vision operates on and through digital media’s participatory visual cultures | |
Hagen et al. | Image memorability prediction with vision transformers | |
CN107341461A (en) | Intelligent Recognition differentiates the method and its system of the art work true and false with analytical technology | |
CN108153785A (en) | The method and apparatus of generation displaying information | |
TWI804090B (en) | Learning system, learning method and program product |
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 |