CN108109124A - Indefinite position picture watermark restorative procedure based on deep learning - Google Patents
Indefinite position picture watermark restorative procedure based on deep learning Download PDFInfo
- Publication number
- CN108109124A CN108109124A CN201711443089.9A CN201711443089A CN108109124A CN 108109124 A CN108109124 A CN 108109124A CN 201711443089 A CN201711443089 A CN 201711443089A CN 108109124 A CN108109124 A CN 108109124A
- Authority
- CN
- China
- Prior art keywords
- watermark
- picture
- layer
- deep learning
- pixel
- 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
- 238000000034 method Methods 0.000 title claims abstract description 14
- 238000013135 deep learning Methods 0.000 title claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 238000003672 processing method Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000000750 progressive effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000005303 weighing Methods 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 2
- 238000013145 classification model Methods 0.000 claims 1
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000011084 recovery Methods 0.000 abstract description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G06T5/77—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/40—Filling a planar surface by adding surface attributes, e.g. colour or texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The present invention relates to a kind of indefinite position picture watermark restorative procedures based on deep learning, which is characterized in that comprises the following steps:(1)The acquisition and pretreatment of training data;(2)Watermark parted pattern is trained;(3)Watermark region is filled.It is an advantage of the invention that:Machine substitutes manpower, cost-effective, liberates manpower;It operates more standardized, avoids artificial very different;Compared to simple watermark recovery technique, the present invention can be with Automatic-searching watermark location.
Description
Technical field
The present invention relates to the indefinite position picture watermark restorative procedures based on deep learning.
Background technology
The prior art is mostly to use picture fix tool artificial treatment picture or the spy by algorithm process specific position
The watermark of setting shape, can not accurately position the watermark location in picture, restore watermark automatically.
The work efficiency of existing way is extremely low, and cost is high, it is impossible to the processing of mass data is adapted to, it is difficult to face big data
The challenge in epoch.
The content of the invention
The defects of to overcome the prior art, the present invention provide a kind of indefinite position picture watermark reparation based on deep learning
Method, the technical scheme is that:
Indefinite position picture watermark restorative procedure based on deep learning, comprises the following steps:
(1)The acquisition and pretreatment of training data;
(2)Watermark parted pattern is trained;
(3)Watermark region is filled.
The step(1)Specific processing method be:
(1-1)The picture number Zhang Zuowei samples of watermark are selected, the watermark location size and shape of each pictures is different;With
Photoshop is marked the watermarking section of each pictures, and the template picture of mark is png forms, and no-watermark position is
Transparent color;
(1-2)To step(1-1)The template picture of middle mark carries out binary conversion treatment, and processing step is as follows:
(a)Prototype drawing is converted into gray-scale map;
(b)Gray-level histogram equalization processing is carried out to gray-scale map;
(c)Binaryzation is carried out with 16~32 threshold value;
(1-3)Model segmentation is carried out to the Prototype drawing of binaryzation, it is specific as follows:
(a)Template picture is identical with picture shape to be marked, travels through the position of template picture and can find and corresponding is waiting to mark
Remember the position on picture, select to travel through each pixel in template picture;
(b)If traversing the value of current pixel point as 0,29 *, 29 neighborhoods using centered on current pixel are cut as bearing
Sample if the value of current pixel point is 1, cuts 29*29 neighborhoods using centered on current pixel as positive sample;
(3)When positive sample quantity is much smaller than negative sample quantity, stochastical sampling is carried out to negative sample, makes positive and negative sample proportion equal
Weighing apparatus;
The step(2)Specific processing method be:Picture pixels point is in step(1)In be divided into and have watermark and no-watermark two
Kind, the neighborhood for having the pixel of watermark is positive sample, and the neighborhood of the pixel of no-watermark is negative sample, uses CNN convolutional neural networks
As disaggregated model, model is established based on CNN, model is divided into eight layers, first layer:Convolutional layer, convolution kernel 3*3, step-length 1, edge
Filling;The second layer:Convolutional layer, convolution kernel 3*3, step-length 1, edge filling;Third layer:Pond layer, pond core 2*2;4th layer:
Convolutional layer, convolution kernel 1*1, step-length 1;Layer 5:Convolutional layer, convolution kernel 1*1, step-length 1;Layer 6:Pond layer, Chi Huahe
2*2;Layer 7:Full context layer;8th layer:Full context layer;Activation primitive uses reLu,, activation primitive is above-mentioned the
First, two, four, five layers of use, eventually by cross entropy loss function counting loss into backpropagation;Pass through training, automatic identification
Watermark location, size and shape in picture.
The step(3)Specific processing method be:Pass through step(3)Middle watermark parted pattern obtains a watermark
Masterplate indicates watermark location, the picture of size and shape in picture;By the way of the progressive filling in edge, by watermark region
The circle of most edge one start, the pixel that likeness in form outer ring is filled to inner ring gradually, until filling entire watermark region.
It is an advantage of the invention that:Machine substitutes manpower, cost-effective, liberates manpower;Operate it is more standardized, avoid manually
It is very different;Compared to simple watermark recovery technique, the present invention can be with Automatic-searching watermark location.
Specific embodiment
The invention will now be further described with reference to specific embodiments, the advantages and features of the present invention will be with description and
It is apparent.But these embodiments are only exemplary, do not form any restrictions to the scope of the present invention.People in the art
Member it should be understood that without departing from the spirit and scope of the invention can to the details of technical solution of the present invention and form into
Row modifications or substitutions, but these modifications and replacement are each fallen in protection scope of the present invention.
Indefinite position picture watermark restorative procedure the present invention relates to a kind of 1, based on deep learning, comprises the following steps:
(1)The acquisition and pretreatment of training data;
(2)Watermark parted pattern is trained;
(3)Watermark region is filled.
The step(1)Specific processing method be:
(1-1)The picture number Zhang Zuowei samples of watermark are selected, the watermark location size and shape of each pictures is different;With
Photoshop is marked the watermarking section of each pictures, and the template picture of mark is png forms, and no-watermark position is
Transparent color;
(1-2)To step(1-1)The template picture of middle mark carries out binary conversion treatment, and processing step is as follows:
(a)Prototype drawing is converted into gray-scale map;
(b)Gray-level histogram equalization processing is carried out to gray-scale map;
(c)Binaryzation is carried out with 16~32 threshold value;
(1-3)Model segmentation is carried out to the Prototype drawing of binaryzation, it is specific as follows:
(a)Template picture is identical with picture shape to be marked, travels through the position of template picture and can find and corresponding is waiting to mark
Remember the position on picture, select to travel through each pixel in template picture;
(b)If traversing the value of current pixel point as 0,29 *, 29 neighborhoods using centered on current pixel are cut as bearing
Sample if the value of current pixel point is 1, cuts 29*29 neighborhoods using centered on current pixel as positive sample;
(3)When positive sample quantity is much smaller than negative sample quantity, stochastical sampling is carried out to negative sample, makes positive and negative sample proportion equal
Weighing apparatus;
The step(2)Specific processing method be:Picture pixels point is in step(1)In be divided into and have watermark and no-watermark two
Kind, the neighborhood for having the pixel of watermark is positive sample, and the neighborhood of the pixel of no-watermark is negative sample, uses CNN convolutional neural networks
As disaggregated model, model is established based on CNN, model is divided into eight layers, first layer:Convolutional layer, convolution kernel 3*3, step-length 1, edge
Filling;The second layer:Convolutional layer, convolution kernel 3*3, step-length 1, edge filling;Third layer:Pond layer, pond core 2*2;4th layer:
Convolutional layer, convolution kernel 1*1, step-length 1;Layer 5:Convolutional layer, convolution kernel 1*1, step-length 1;Layer 6:Pond layer, Chi Huahe
2*2;Layer 7:Full context layer;8th layer:Full context layer;Activation primitive uses reLu,, activation primitive is above-mentioned the
First, two, four, five layers of use, eventually by cross entropy loss function counting loss into backpropagation;Pass through training, automatic identification
Watermark location, size and shape in picture.
The step(3)Specific processing method be:Pass through step(3)Middle watermark parted pattern obtains a watermark
Masterplate indicates watermark location, the picture of size and shape in picture;By the way of the progressive filling in edge, by watermark region
The circle of most edge one start, the pixel that likeness in form outer ring is filled to inner ring gradually, until filling entire watermark region.
Claims (4)
1. the indefinite position picture watermark restorative procedure based on deep learning, which is characterized in that comprise the following steps:
(1)The acquisition and pretreatment of training data;
(2)Watermark parted pattern is trained;
(3)Watermark region is filled.
2. the indefinite position picture watermark restorative procedure according to claim 1 based on deep learning, which is characterized in that institute
The step of stating(1)Specific processing method be:
(1-1)The picture number Zhang Zuowei samples of watermark are selected, the watermark location size and shape of each pictures is different;With
Photoshop is marked the watermarking section of each pictures, and the template picture of mark is png forms, and no-watermark position is
Transparent color;
(1-2)To step(1-1)The template picture of middle mark carries out binary conversion treatment, and processing step is as follows:
(a)Prototype drawing is converted into gray-scale map;
(b)Gray-level histogram equalization processing is carried out to gray-scale map;
(c)Binaryzation is carried out with 16~32 threshold value,
(1-3)Model segmentation is carried out to the Prototype drawing of binaryzation, it is specific as follows:
(a)Template picture is identical with picture shape to be marked, travels through the position of template picture and can find and corresponding is waiting to mark
Remember the position on picture, select to travel through each pixel in template picture;
(b)If traversing the value of current pixel point as 0,29 *, 29 neighborhoods using centered on current pixel are cut as bearing
Sample if the value of current pixel point is 1, cuts 29*29 neighborhoods using centered on current pixel as positive sample;
(3)When positive sample quantity is much smaller than negative sample quantity, stochastical sampling is carried out to negative sample, makes positive and negative sample proportion equal
Weighing apparatus.
3. the indefinite position picture watermark restorative procedure according to claim 1 based on deep learning, which is characterized in that institute
The step of stating(2)Specific processing method be:Picture pixels point is in step(1)In be divided into and have two kinds of watermark and no-watermark, have water
The neighborhood of the pixel of print is positive sample, and the neighborhood of the pixel of no-watermark is negative sample, using CNN convolutional neural networks as classification
Model establishes model based on CNN, and model is divided into eight layers, first layer:Convolutional layer, convolution kernel 3*3, step-length 1, edge filling;The
Two layers:Convolutional layer, convolution kernel 3*3, step-length 1, edge filling;Third layer:Pond layer, pond core 2*2;4th layer:Convolutional layer,
Convolution kernel 1*1, step-length 1;Layer 5:Convolutional layer, convolution kernel 1*1, step-length 1;Layer 6:Pond layer, pond core 2*2;The
Seven layers:Full context layer;8th layer:Full context layer;Activation primitive uses reLu,, activation primitive is above-mentioned first and second, four,
Five layers of use, eventually by cross entropy loss function counting loss into backpropagation;By training, the water in automatic identification picture
Print position, size and shape.
4. the indefinite position picture watermark restorative procedure according to claim 1 based on deep learning, which is characterized in that institute
The step of stating(3)Specific processing method be:Pass through step(3)Middle watermark parted pattern obtains a watermark masterplate, that is, indicates
Watermark location, the picture of size and shape in picture;By the way of the progressive filling in edge, enclosed by the most edge one of watermark region
Start, the pixel that likeness in form outer ring is filled to inner ring gradually, until filling entire watermark region.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711443089.9A CN108109124A (en) | 2017-12-27 | 2017-12-27 | Indefinite position picture watermark restorative procedure based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711443089.9A CN108109124A (en) | 2017-12-27 | 2017-12-27 | Indefinite position picture watermark restorative procedure based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108109124A true CN108109124A (en) | 2018-06-01 |
Family
ID=62211849
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711443089.9A Pending CN108109124A (en) | 2017-12-27 | 2017-12-27 | Indefinite position picture watermark restorative procedure based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108109124A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109065001A (en) * | 2018-06-20 | 2018-12-21 | 腾讯科技(深圳)有限公司 | A kind of down-sampled method, apparatus, terminal device and the medium of image |
CN111199175A (en) * | 2018-11-20 | 2020-05-26 | 株式会社日立制作所 | Training method and device for target detection network model |
CN111435544A (en) * | 2019-01-14 | 2020-07-21 | 珠海格力电器股份有限公司 | Picture processing method and device |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050123194A1 (en) * | 2002-02-21 | 2005-06-09 | Xerox Corporation | Method of embedding color information in printed documents using watermarking |
CN101118592A (en) * | 2007-08-22 | 2008-02-06 | 大连理工大学 | Printers evidence obtaining method based on character printing feature |
JP2010079507A (en) * | 2008-09-25 | 2010-04-08 | Fuji Xerox Co Ltd | Image processor and image processing program |
CN102194204A (en) * | 2010-03-15 | 2011-09-21 | 北京大学 | Method and device for embedding and extracting reversible watermarking as well as method and device for recovering image |
CN103533343A (en) * | 2013-09-29 | 2014-01-22 | 宁波大学 | Stereo image quality objective evaluation method based on digital watermarking |
CN104077577A (en) * | 2014-07-03 | 2014-10-01 | 浙江大学 | Trademark detection method based on convolutional neural network |
CN104217387A (en) * | 2014-01-22 | 2014-12-17 | 河南师范大学 | Image watermark embedding and extracting method and device based on quantization embedding |
CN104899821A (en) * | 2015-05-27 | 2015-09-09 | 合肥高维数据技术有限公司 | Method for erasing visible watermark of document image |
CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
CN106096605A (en) * | 2016-06-02 | 2016-11-09 | 史方 | A kind of image obscuring area detection method based on degree of depth study and device |
CN106204424A (en) * | 2016-06-30 | 2016-12-07 | 厦门美图之家科技有限公司 | Image removes water mark method, applies and calculating equipment |
CN106296593A (en) * | 2015-05-29 | 2017-01-04 | 腾讯科技(深圳)有限公司 | Image recovery method and device |
CN106934780A (en) * | 2017-03-15 | 2017-07-07 | 中山大学 | A kind of automatic watermark minimizing technology based on image repair |
CN107220655A (en) * | 2016-03-22 | 2017-09-29 | 华南理工大学 | A kind of hand-written, printed text sorting technique based on deep learning |
-
2017
- 2017-12-27 CN CN201711443089.9A patent/CN108109124A/en active Pending
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050123194A1 (en) * | 2002-02-21 | 2005-06-09 | Xerox Corporation | Method of embedding color information in printed documents using watermarking |
CN101118592A (en) * | 2007-08-22 | 2008-02-06 | 大连理工大学 | Printers evidence obtaining method based on character printing feature |
JP2010079507A (en) * | 2008-09-25 | 2010-04-08 | Fuji Xerox Co Ltd | Image processor and image processing program |
CN102194204A (en) * | 2010-03-15 | 2011-09-21 | 北京大学 | Method and device for embedding and extracting reversible watermarking as well as method and device for recovering image |
CN103533343A (en) * | 2013-09-29 | 2014-01-22 | 宁波大学 | Stereo image quality objective evaluation method based on digital watermarking |
CN104217387A (en) * | 2014-01-22 | 2014-12-17 | 河南师范大学 | Image watermark embedding and extracting method and device based on quantization embedding |
CN104077577A (en) * | 2014-07-03 | 2014-10-01 | 浙江大学 | Trademark detection method based on convolutional neural network |
CN104899821A (en) * | 2015-05-27 | 2015-09-09 | 合肥高维数据技术有限公司 | Method for erasing visible watermark of document image |
CN106296593A (en) * | 2015-05-29 | 2017-01-04 | 腾讯科技(深圳)有限公司 | Image recovery method and device |
CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
CN107220655A (en) * | 2016-03-22 | 2017-09-29 | 华南理工大学 | A kind of hand-written, printed text sorting technique based on deep learning |
CN106096605A (en) * | 2016-06-02 | 2016-11-09 | 史方 | A kind of image obscuring area detection method based on degree of depth study and device |
CN106204424A (en) * | 2016-06-30 | 2016-12-07 | 厦门美图之家科技有限公司 | Image removes water mark method, applies and calculating equipment |
CN106934780A (en) * | 2017-03-15 | 2017-07-07 | 中山大学 | A kind of automatic watermark minimizing technology based on image repair |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109065001A (en) * | 2018-06-20 | 2018-12-21 | 腾讯科技(深圳)有限公司 | A kind of down-sampled method, apparatus, terminal device and the medium of image |
CN109065001B (en) * | 2018-06-20 | 2021-06-25 | 腾讯科技(深圳)有限公司 | Image down-sampling method and device, terminal equipment and medium |
CN111199175A (en) * | 2018-11-20 | 2020-05-26 | 株式会社日立制作所 | Training method and device for target detection network model |
CN111435544A (en) * | 2019-01-14 | 2020-07-21 | 珠海格力电器股份有限公司 | Picture processing method and device |
CN111435544B (en) * | 2019-01-14 | 2021-11-05 | 珠海格力电器股份有限公司 | Picture processing method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108109124A (en) | Indefinite position picture watermark restorative procedure based on deep learning | |
CN106295653B (en) | Water quality image classification method | |
CN111179229A (en) | Industrial CT defect detection method based on deep learning | |
CN110222628A (en) | A kind of face restorative procedure based on production confrontation network | |
CN106709909A (en) | Flexible robot vision recognition and positioning system based on depth learning | |
CN107084991A (en) | The detection of quartz pushrod bubble and quality grading method based on machine vision | |
CN108197642A (en) | A kind of seal discrimination method and device | |
CN104156951B (en) | A kind of white blood cell detection method for BAL fluid smear | |
CN108154102A (en) | A kind of traffic sign recognition method | |
CN114627383B (en) | Small sample defect detection method based on metric learning | |
CN104680509B (en) | A kind of real-time circular printing image defect detection method | |
CN112037219A (en) | Metal surface defect detection method based on two-stage convolution neural network | |
CN107451588A (en) | A kind of pop can smooth surface coding ONLINE RECOGNITION method based on machine vision | |
CN114120317B (en) | Optical element surface damage identification method based on deep learning and image processing | |
CN104881665A (en) | Chip character identification and verification method and apparatus | |
CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
CN110189351A (en) | A kind of scratch image data amplification method based on production confrontation network | |
CN109800619A (en) | Maturity period citrusfruit image-recognizing method | |
CN113191222B (en) | Underwater fish target detection method and device | |
CN109086772A (en) | A kind of recognition methods and system distorting adhesion character picture validation code | |
CN110866915A (en) | Circular inkstone quality detection method based on metric learning | |
CN111091122B (en) | Training and detecting method and device for multi-scale characteristic convolutional neural network | |
CN116110036A (en) | Electric power nameplate information defect level judging method and device based on machine vision | |
CN112686265A (en) | Hierarchic contour extraction-based pictograph segmentation method | |
CN110414506B (en) | Bank card number automatic identification method based on data augmentation and convolution neural network |
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 | ||
CB03 | Change of inventor or designer information |
Inventor after: Li Zuochao Inventor after: Bai Junfeng Inventor after: Zhang Wenzhan Inventor after: Liu Ziyao Inventor after: Su Weijie Inventor before: Bai Junfeng Inventor before: Zhang Wenzhan Inventor before: Liu Ziyao Inventor before: Su Weijie |
|
CB03 | Change of inventor or designer information | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180601 |
|
WD01 | Invention patent application deemed withdrawn after publication |