WO2020238673A1

WO2020238673A1 - Watermark detection model generation and watermark detection method, apparatus and device

Info

Publication number: WO2020238673A1
Application number: PCT/CN2020/090797
Authority: WO
Inventors: 邓磊; 刘波; 刘绍辉; 刘永亮
Original assignee: 阿里巴巴集团控股有限公司
Priority date: 2019-05-30
Filing date: 2020-05-18
Publication date: 2020-12-03
Also published as: CN112017092A

Abstract

A watermark detection model generation and watermark detection method, apparatus and device. The model generation method comprises: obtaining a plurality of watermark data sets (S101), wherein carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the watermark template is a template used when watermark information is added to the carrier objects; generating a plurality of watermark detection branch modules according to the plurality of watermark data sets (S102), wherein each watermark detection branch module is used to obtain a preliminary watermark detection result for the carrier object on the basis of the specified watermark embedding strength information and the specified size relationship information; and generating a watermark detection model according to the plurality of watermark detection branch modules (S103), wherein the watermark detection model is used to obtain a final watermark detection result for the carrier object according to the preliminary watermark detection results of the plurality of watermark detection branch modules. By means of the above-mentioned method, the existing problem of great difficulty and high complexity of watermark detection is solved.

Description

Watermark detection model generation and watermark detection method, device and equipment

This application claims the priority of the Chinese patent application filed on May 30, 2019 with the application number 201910461993.5 and the invention title "Watermark detection model generation and watermark detection methods, devices and equipment", the entire content of which is incorporated herein by reference Applying.

Technical field

This application relates to the field of computer technology, and specifically relates to a method, device, electronic device, and storage device of a watermark detection model. This application also relates to a method, device, electronic device, and storage device of a watermark detection model.

Background technique

The images contain rich and intuitive information. Currently, in the fields of social networking, shopping and tourism on the Internet, a large number of images are needed to convey information to users. Due to the extremely fast and convenient dissemination of Internet information, more and more individuals and companies choose to embed watermarks on their images, such as watermarking trademarks or web addresses in certain areas of the images, in order to protect the intellectual property rights of the images. Therefore, before using the image, the image user needs to review the image to detect whether there is a watermark in the image, so as to avoid misuse or infringement. However, with the rapid development and application of the Internet, image users will obtain a large number of images through various channels every day. In this case, manual review cannot meet the demand.

In the prior art, traditional image processing and optimization methods are used to automatically detect the watermark embedded in the image by analyzing the consistency of a large number of image watermarks. This detection method inputs the collection of pictures into the algorithm, separates the watermark and the background image through the algorithm, and then detects the watermark. With this method, it is necessary to collect a large number of image sets that use the same watermark, and it is required that the watermark embedding size, relative embedding position, and embedding strength in these images with the same visible watermark are the same. These requirements are difficult to meet in practical applications.

Therefore, the prior art has the problems of greater difficulty and complexity in detecting watermarks.

Summary of the invention

This application provides a method, device, electronic device, and storage device for generating a watermark detection model to solve the existing problems of greater difficulty and complexity in detecting watermarks.

This application provides a method for generating a watermark detection model, including:

Obtain multiple watermark data sets; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the difference between the size information of the watermark template and the size information of the carrier object The watermark template is a template used when adding watermark information to the carrier object;

According to the multiple watermark data sets, multiple watermark detection branch modules are generated; each watermark detection branch module is used to obtain a preliminary watermark detection result for the carrier object based on the specified watermark embedding strength information and the specified size relationship information;

A watermark detection model is generated according to the multiple watermark detection branch modules; the watermark detection model is used to obtain a final watermark detection result for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules.

Optionally, the obtaining a watermark data set includes:

Obtain a carrier object set; wherein, each carrier object in the same carrier object set has the same watermark embedding strength information and the same size relationship information; each carrier object in a different carrier object set has different watermark embedding strength information and different size relationship information;

Obtain the label information of the carrier object set;

According to the carrier object set and the label information, multiple watermark data sets are generated.

Optional, also includes:

Save the label information in a text file;

Converting the text file into a non-text file in a predetermined format;

The generating multiple watermark data sets according to the carrier object set and the label information includes:

According to the carrier object set and the non-text file, multiple watermark data sets are generated.

Optionally, the label information includes at least one of the following:

Type information of the watermark in the carrier object;

The position information of the watermark embedded in the carrier object;

The name information of the watermark in the carrier object.

Optionally, the generating multiple watermark detection branch modules according to the multiple watermark data sets includes:

Input the carrier object in the watermark data set into the convolutional layer in the neural network model, and extract the feature map of the carrier object;

Input the feature map of the carrier object into the candidate area network in the neural network model to generate a recommended area containing a watermark;

Input the feature map of the carrier object and the recommended region into the region of interest pooling layer in the neural network model to obtain a feature map of the recommended region;

Input the recommended area feature map to the classifier to obtain the test value of the watermark category in the carrier object;

Training a neural network model according to the test value and the label information to generate a watermark detection branch module.

Optionally, the number of watermark detection branches is equal to the number of watermark data sets; one watermark data set is used to generate a watermark detection branch module.

This application also provides a watermark detection method, including:

Obtain the carrier object to be detected;

The carrier object to be detected is input into the watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to obtain the target according to the watermark embedding strength information and size relationship information of the carrier object to be detected The watermark detection result of the carrier object to be detected.

Optionally, the inputting the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected includes:

Input the carrier object to be detected into a watermark detection model, and obtain a plurality of preliminary watermark detection results for the carrier object to be detected;

According to the plurality of preliminary watermark detection results for the carrier object to be detected, a final watermark detection result for the carrier object to be detected is obtained.

Optionally, it also includes: obtaining the confidence of the preliminary detection result for each watermark;

The obtaining the final detection result of the watermark for the carrier object to be detected according to the preliminary detection results of the plurality of watermarks for the carrier object to be detected includes:

According to the preliminary detection results of the multiple watermarks and the confidence of the preliminary detection results for each watermark, a final detection result of the watermark for the carrier object to be detected is determined.

Optionally, the determining the final detection result of the watermark for the carrier object to be detected according to the preliminary detection results of the multiple watermarks and the confidence of the preliminary detection results for each watermark includes:

The preliminary detection result of the watermark with the greatest confidence is used as the final detection result of the watermark.

Optionally, inputting the carrier object to be detected into a watermark detection model to obtain multiple preliminary watermark detection results for the carrier object to be detected includes:

Inputting the carrier object to be detected into the convolutional layer of the watermark detection model, and extracting a feature map of the carrier object to be detected;

Input the feature map of the carrier object to be detected into the candidate area network of the watermark detection model to generate a recommended area containing a visible watermark;

Input the feature map of the carrier object to be detected and the recommended region into the interest region pooling layer of the watermark detection model to obtain a feature map of the recommended region;

The recommended area feature map is input into a classifier to obtain preliminary results of watermark detection in a plurality of carrier objects to be detected.

Optionally, the preliminary detection result of the watermark includes at least one of the following information:

The category information of the watermark in the carrier object to be detected;

Position information of the watermark in the carrier object to be detected.

This application also provides a watermark detection method, including:

Obtain the carrier object to be detected;

Obtain the watermark embedding strength information and size relationship information of the carrier object to be detected. The size relationship information is used to indicate the relationship between the size information of the watermark template and the size information of the carrier object. The template used when the carrier object adds watermark information;

According to the watermark embedding strength information and the size relationship information, a watermark detection result for the carrier object to be detected is obtained.

This application also provides a device for generating a watermark detection model, including:

The watermark data set obtaining unit is used to obtain multiple watermark data sets; wherein the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the size of the watermark template The relationship between information and size information of the carrier object, the watermark template is a template used when adding watermark information to the carrier object;

The watermark detection branch generating unit is used to generate multiple watermark detection branch modules according to the multiple watermark data sets; each watermark detection branch module is used to obtain the target carrier object based on the specified watermark embedding strength information and the specified size relationship information The preliminary detection result of the watermark;

The watermark detection model generation unit is configured to generate a watermark detection model according to the multiple watermark detection branch modules; the watermark detection model is used to obtain the final watermark for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules Test results.

This application also provides an electronic device, including:

processor;

The memory is used to store the program of the method for generating the watermark detection model. After the device is powered on and runs the program of the method for generating the watermark detection model through the processor, the following steps are executed:

This application also provides a storage device that stores a program of a method for generating a watermark detection model. The program is run by a processor to perform the following steps:

This application additionally provides a watermark detection device, including:

A carrier object to be detected obtaining unit for obtaining a carrier object to be detected;

The watermark detection result obtaining unit is configured to input the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to embed the watermark according to the carrier object to be detected The intensity information and the size relationship information obtain a watermark detection result for the carrier object to be detected.

This application additionally provides an electronic device, including:

processor;

The memory is used to store the program of the watermark detection method. After the device is powered on and runs the program of the watermark detection method through the processor, the following steps are executed:

Obtain the carrier object to be detected;

This application additionally provides a storage device that stores a program of the watermark detection method, and the program is run by a processor to perform the following steps:

Obtain the carrier object to be detected;

Compared with the prior art, this application has the following advantages:

This application provides a method for generating a watermark detection model. The first embodiment of this application trains multiple branches under different watermark embedding strength information and different size relationship information, and constructs watermark detection based on the multiple branches trained Model, using the above-mentioned watermark detection model can effectively complete the detection task of visible watermarks under different watermark embedding strengths and different size relationships, and solve the existing problems of greater difficulty and complexity in detecting watermarks.

Description of the drawings

Figure 1 is a schematic diagram of a first application scenario embodiment provided by this application;

Figure 2 is a schematic diagram of a second application scenario embodiment provided by this application;

Fig. 3 is a flowchart of a method for generating a watermark detection model provided by the first embodiment of the present application;

Fig. 4 is a flowchart of obtaining a watermark data set provided by the first embodiment of the present application;

FIG. 5 is a schematic diagram of generating multiple watermark detection branch modules according to multiple watermark data sets according to the first embodiment of the present application;

Fig. 6 is a flow chart of generating multiple watermark detection branch modules according to multiple watermark data sets according to the first embodiment of the present application;

Fig. 7 is a schematic diagram of a watermark detection model provided by the first embodiment of the present application;

Fig. 8 is a flowchart of a watermark detection method provided by the second embodiment of the present application;

FIG. 9 is a flowchart of obtaining preliminary detection results of multiple watermarks for the carrier object to be detected according to the second embodiment of the present application;

FIG. 10 is a flowchart of a watermark detection method provided by the third embodiment of this application;

FIG. 11 is a schematic diagram of a device for generating a watermark detection model provided by the fourth embodiment of the present application;

FIG. 12 is a schematic diagram of an electronic device provided by a fifth embodiment of the present application;

FIG. 13 is a schematic diagram of a watermark detection device provided by a seventh embodiment of the present application;

FIG. 14 is a schematic diagram of an electronic device provided by an eighth embodiment of the present application.

Detailed ways

In the following description, many specific details are explained in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without violating the connotation of the present invention. Therefore, the present invention is not limited by the specific implementation disclosed below.

In order to show this application more clearly, first introduce the application scenarios of the watermark detection method provided in the embodiments of this application.

Some embodiments provided in this application can be applied to a scenario where a client interacts with a server, as shown in FIG. 1, which is a schematic diagram of an embodiment of the first application scenario provided in this application. The client first establishes a connection with the server. After the connection, the client sends the image to be detected to the server. The server receives the image to be detected and performs watermark detection on the image to be detected, and obtains the watermark detection result of the image to be detected, and the watermark detection result of the image to be detected Send to the client, and the client receives the watermark detection result of the image to be detected.

In addition, some embodiments provided in this application can be applied to the internal processing of the client, as shown in FIG. 2, which is a schematic diagram of the second application scenario embodiment provided in this application. After the client obtains the image to be detected, it detects the image to be detected, obtains the watermark detection result of the image to be detected, and then outputs the watermark detection result of the image to be detected. It should be noted that the above two application scenarios are only two examples of application scenarios. The purpose of providing these two application scenario embodiments is to facilitate the understanding of the watermark detection method of this application, and is not used to limit the watermark detection of this application. method.

The first embodiment of the present application provides a method for generating a watermark detection model, which is described below with reference to FIG. 3, FIG. 4, FIG. 5, and FIG. 6.

As shown in Figure 3, in step S101, multiple watermark data sets are obtained; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the watermark The relationship between the size information of the template and the size information of the carrier object. The watermark template is a template used when the watermark information is embedded in the carrier object.

The watermark embedding strength may refer to the watermark embedding strength factor. The stronger the watermark embedding strength, the better the robustness of the watermark.

The size relationship may refer to the scale ratio of the watermark template to the carrier object, that is, the ratio of the width and height of the watermark template rectangle to the width and height of the carrier object.

Please refer to Fig. 4, which is a flow chart for obtaining a watermark data set according to the first embodiment of this application, which specifically includes steps S101-1 to S101-3.

As shown in Figure 4, in step S101-1, a carrier object set is obtained; wherein, each carrier object in the same carrier object set has the same watermark embedding strength information and the same size relationship information; each carrier object in a different carrier object set has Different watermark embeds strength information and different size relationship information.

Each carrier object set can refer to a collection of multiple carrier objects embedded with watermarks.

Obtaining a carrier object set may refer to obtaining multiple carrier object sets, each carrier object in each carrier object set has the same watermark embedding strength information and the same size relationship information; each carrier object in different carrier object sets has different watermarks Embedded strength information and different size relationship information.

For example, n groups of visible watermark picture sets (ie, carrier object sets) are obtained, which are respectively numbered 1, 2, 3...n. The watermark embedding strength in the visible watermark picture set No. 1 is α=x1, the scale ratio of the watermark template to the added picture (carrier object) is s=y1; the watermark embedding strength in the visible watermark picture set No. 2 is α=x2, the watermark template The scale ratio to the added picture is s=y2; the watermark embedding strength of the visible watermark picture set of No. 3 is α=x3, and the scale ratio of the watermark template to the added picture is s=y3; and so on, n-1 is visible The watermark embedding strength of the watermark picture set is α=xn-1, the scale ratio of the watermark template and the added picture is s=yn-1; the watermark embedding strength of the visible watermark picture set n is α=xn, the watermark template and the added picture The scale ratio of the picture is s=yn (xi, yi in the above description are all decimals between 0 and 1, i∈[1,n]). The watermark embedding positions in the above n groups of visible watermark picture sets are random, and the position information of the watermark added is recorded during the process of embedding the watermark template.

As shown in Fig. 4, in step S101-2, the label information of the carrier object set is obtained.

The label information includes at least one of the following: category information of the watermark in the carrier object; position information of the watermark embedded in the carrier object; and name information of the watermark in the carrier object. The category of the watermark in the carrier object is determined according to the watermark embedding strength information and size relationship information of the carrier object. For example, the category information of the watermark in the carrier object in the visible watermark picture set No. 1 in the above example may be (α=x1, s=y1).

As shown in FIG. 4, in step S101-3, a plurality of watermark data sets are generated according to the carrier object set and the label information.

As an implementation manner, in order to facilitate automated processing, the label information may be saved in a text file; and then the text file is converted into a non-text file in a predetermined format. The non-text file in the predetermined format may be an XML file.

Still using the above example, you can first save the label information of each visible watermark picture set to a TXT file; then convert the TXT file to a VOC2007 format XML file, and convert the No. 1 visible watermark picture set to the No. 1 XML file Merge the No. 1 watermark data set, merge the No. 2 visible watermark picture set and the No. 2 XML file to create the No. 2 watermark data set, merge the No. 3 visible watermark picture set and the No. 3 XML file to create the No. 3 watermark data set, so By analogy, construct a watermark data set from 1 to n.

As shown in Figure 3, in step S102, based on the multiple watermark data sets, multiple watermark detection branch modules are generated; each watermark detection branch module is used to obtain information based on the specified watermark embedding strength information and the specified size relationship information The preliminary detection result of the watermark for the carrier object.

The specified watermark embedding strength refers to the watermark embedding strength of the watermark data set corresponding to the watermark detection branch module. For example, the watermark embedding strength corresponding to the watermark detection branch 1 is α=x1; the watermark embedding strength corresponding to the watermark detection branch n is α=xn.

The specified size relationship refers to the size relationship of the watermark data set corresponding to the watermark detection branch module. For example, the size relationship corresponding to the watermark detection branch 1 is the scale ratio s=y1.

The preliminary watermark detection result refers to a preliminary watermark detection result obtained by a watermark detection branch module. Each watermark detection branch module obtains a preliminary watermark detection result, and then obtains a final watermark detection result according to the preliminary watermark detection result.

The number of the watermark detection branches may be equal to the number of the watermark data sets; one watermark data set is used to generate a watermark detection branch module. For example, as shown in Fig. 5, according to n watermark data sets, n watermark detection branch modules are generated.

For example, input the n watermark data sets constructed in step 101 into the basic target detection network Faster R-CNN to obtain branches 1 to n. Among them, Faster R-CNN is a target detection network based on deep learning.

Please refer to FIG. 6, which provides a flow chart of generating multiple watermark detection branch modules according to multiple watermark data sets according to the first embodiment of this application, which specifically includes steps S102-1 to S102-4.

As shown in FIG. 6, in step S102-1, the carrier object in the watermark data set is input to the convolutional layer in the neural network model, and the feature map of the carrier object is extracted.

For example, as shown in Figure 5, the neural network model is Faster R-CNN, and the watermark data set includes watermark data sets from 1 to n. The pictures (carrier objects) with visible watermarks in the watermark data set are input to the convolutional layer Conv layers( For example, select VGG16 convolutional network) to extract feature maps (feature maps) of pictures with visible watermarks.

As shown in FIG. 6, in step S102-2, the feature map of the carrier object is input into the candidate area network in the neural network model to generate a recommended area containing a watermark.

For example, as shown in Figure 5, the above feature map with visible watermark pictures is input into the candidate regional network RPN (Region Proposals Networks) layer to generate anchors (the target detection algorithm Faster R-CNN based on deep learning is used in each pixel of the image The RPN layer uses softmax to determine whether the anchors belong to the foreground (anchor containing the watermark) or the background (anchor not containing the watermark), and then use the bounding box regression to correct the anchors to obtain an accurate recommended area containing the watermark.

As shown in FIG. 6, in step S102-3, the feature map of the carrier object and the recommended region are input into the region of interest pooling layer in the neural network model to obtain a feature map of the recommended region.

For example, as shown in Figure 5, the feature map of the picture with visible watermark generated by the convolutional layer and the recommended area containing the watermark generated by the candidate area network layer are input to the RoI Pooling (region of interest pooling layer) layer, the region of interest The pooling layer extracts recommended feature maps (proposal feature maps).

As shown in Fig. 6, in step S102-4, the recommended area feature map is input to the classifier to obtain the test value of the watermark category in the carrier object.

As shown in FIG. 6, in step S102-5, the neural network model is trained according to the test value and the label information to generate a watermark detection branch module.

As shown in FIG. 3, in step S103, a watermark detection model is generated according to the multiple watermark detection branch modules; the watermark detection model is used to obtain the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules The final result of watermark detection.

As shown in Figure 7, it is a schematic diagram of a watermark detection model.

So far, the introduction of the first embodiment of the present application is completed. The first embodiment of this application trains multiple branches under different watermark embedding strength information and different size relationship information, and constructs a watermark detection model based on the multiple branches trained. The above watermark detection model can effectively detect different The detection task of visible watermarks under the relationship between the watermark embedding strength and different sizes solves the existing problems of difficulty and complexity in detecting watermarks.

The second embodiment of the present application provides a watermark detection method, which is described below with reference to FIG. 8 and FIG. 9.

As shown in Fig. 8, in step S601, a carrier object to be detected is obtained.

The carrier object to be detected may refer to any image. For example, the carrier object to be detected is an image embedded with a visible watermark.

As shown in Figure 8, in step S602, the carrier object to be detected is input into a watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to The watermark embedding strength information and the size relationship information of to obtain the watermark detection result for the carrier object to be detected.

The size relationship may refer to the scale ratio between the watermark template and the carrier object, that is, the ratio of the width and height of the watermark template rectangle to the width and height of the carrier object to be detected. Among them, the watermark template is the template used when adding watermark information to the carrier object to be detected.

The watermark detection result includes: the category information of the watermark in the carrier object to be detected; the position information of the watermark in the carrier object to be detected, the picture name of the watermark in the carrier object to be detected, etc.

The inputting the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected includes:

As shown in Figure 7, the watermark detection model consists of multiple watermark detection branch modules. Each watermark detection branch module can obtain a preliminary watermark detection result for the carrier object to be detected. According to the preliminary detection results of multiple watermarks, the The final detection result of the watermark of the carrier object.

Please refer to FIG. 9, which is a flow chart for inputting a carrier object to be detected into a watermark detection model to obtain a plurality of preliminary watermark detection results for the carrier object to be detected, which specifically includes steps S502-1 to S502-4.

As shown in FIG. 9, in step S502-1, the carrier object to be detected is input to the convolutional layer of the watermark detection model, and the feature map of the carrier object to be detected is extracted.

As shown in FIG. 9, in step S502-2, the feature map of the carrier object to be detected is input into the candidate area network of the watermark detection model to generate a recommended area containing a visible watermark.

As shown in FIG. 9, in step S502-3, the feature map of the carrier object to be detected and the recommended region are input into the interest region pooling layer of the watermark detection model to obtain a feature map of the recommended region.

As shown in FIG. 9, in step S502-4, the recommended area feature map is input to the classifier to obtain preliminary detection results of watermarks in a plurality of carrier objects to be detected.

As a preferred manner, the second embodiment of the present application may further include: obtaining the confidence of the preliminary detection result for each watermark.

The determining the final detection result of the watermark for the carrier object to be detected according to the preliminary detection results of the multiple watermarks and the confidence of the preliminary detection results for each watermark includes:

For example, as shown in Figure 7, the input picture (that is, the carrier object to be detected) is input into the multi-branch watermark detection model, and each branch will get their respective preliminary detection results: the position coordinates of the detection frame of the visible watermark and the detection The confidence score of the result (n branches get their respective confidence score1, score2, score3,...scoren-1 and scoren). If a branch does not detect a watermark, the watermark detection frame information output by this branch will be Is set to [0,0,0,0], and the detection confidence of this branch will also be set to 0, the multi-branch self-watermark detection model will be based on the confidence (score) of the detection results of these n branches as The decision criterion is to select the branch with the maximum score as the final detection result, and output the position coordinates, category, picture name, and confidence of the visible watermark detection frame. If no visible watermark is detected in all branches, the multi-branch watermark detection model will set the position of the visible watermark detection frame of the detected picture to [0,0,0,0], the confidence level to 0 and the picture name.

The third embodiment of the present application provides a watermark detection method, which is described below with reference to FIG. 10.

As shown in FIG. 10, in step 801, a carrier object to be detected is obtained.

As shown in Figure 10, in step 802, the watermark embedding strength information and size relationship information of the carrier object to be detected are obtained. The size relationship information is used to indicate the difference between the size information of the watermark template and the size information of the carrier object. Relationship, the watermark template is a template used when adding watermark information to the carrier object.

As shown in FIG. 10, in step 803, a watermark detection result for the carrier object to be detected is obtained according to the watermark embedding strength information and size relationship information.

Corresponding to the method for generating a watermark detection model provided in the first embodiment of this application, the fourth embodiment of this application also provides a device for generating a watermark detection model.

As shown in Figure 11, the device for generating the watermark detection model includes:

The watermark data set obtaining unit 901 is used to obtain multiple watermark data sets; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the size of the watermark template The relationship between the size information and the size information of the carrier object, where the watermark template is a template used when adding watermark information to the carrier object;

The watermark detection branch generating unit 902 is configured to generate multiple watermark detection branch modules according to the multiple watermark data sets; each watermark detection branch module is used to obtain the target carrier based on the specified watermark embedding strength information and the specified size relationship information The preliminary detection result of the watermark of the object;

The watermark detection model generating unit 903 is configured to generate a watermark detection model according to the multiple watermark detection branch modules; the watermark detection model is configured to obtain a watermark for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules The final test result.

Optionally, the watermark data set obtaining unit is specifically configured to:

Obtain the label information of the carrier object set;

Optionally, the device further includes:

The label information saving unit is used to save the label information in a text file;

A text file conversion unit for converting the text file into a non-text file in a predetermined format;

Optionally, the label information includes at least one of the following:

Type information of the watermark in the carrier object;

The position information of the watermark embedded in the carrier object;

The name information of the watermark in the carrier object.

Optionally, the watermark detection branch generating unit is specifically configured to:

Optionally, the number of the watermark detection branch is equal to the number of the watermark data set; one watermark data set is used to generate a watermark detection branch module.

It should be noted that for the detailed description of the device for generating the watermark detection model provided by the fourth embodiment of the present application, reference may be made to the related description of the first embodiment of the present application, and details are not repeated here.

Corresponding to the method for generating a watermark detection model provided in the first embodiment of this application, the fifth embodiment of this application also provides an electronic device.

As shown in FIG. 12, the electronic device includes:

Processor 1001;

The memory 1002 is used to store the program of the method for generating the watermark detection model. After the device is powered on and runs the program of the method for generating the watermark detection model through the processor, the following steps are executed:

Optionally, the obtaining a watermark data set includes:

Obtain the label information of the carrier object set;

Optionally, the electronic device further performs the following steps:

Save the label information in a text file;

Converting the text file into a non-text file in a predetermined format;

Optionally, the label information includes at least one of the following:

Type information of the watermark in the carrier object;

The position information of the watermark embedded in the carrier object;

The name information of the watermark in the carrier object.

It should be noted that, for the detailed description of the electronic device provided in the fifth embodiment of the present application, reference may be made to the related description of the first embodiment of the present application, which is not repeated here.

Corresponding to the method for generating a watermark detection model provided in the first embodiment of this application, the sixth embodiment of this application also provides a storage device that stores a program of the method for generating a watermark detection model, and the program is run by the processor , Perform the following steps:

It should be noted that, for a detailed description of the storage device provided in the sixth embodiment of the present application, reference may be made to the related description of the first embodiment of the present application, which will not be repeated here.

Corresponding to the watermark detection method provided in the second embodiment of this application, the seventh embodiment of this application also provides a watermark detection device.

As shown in Figure 13, the watermark detection device includes:

The carrier object to be detected obtaining unit 1101 is configured to obtain the carrier object to be detected;

The watermark detection result obtaining unit 1102 is configured to input the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to determine the watermark of the carrier object to be detected Embedding intensity information and size relationship information to obtain a watermark detection result for the carrier object to be detected.

Optionally, the watermark detection result obtaining unit includes:

The preliminary detection result obtaining subunit is used to input the carrier object to be detected into a watermark detection model to obtain a plurality of preliminary watermark detection results for the carrier object to be detected;

The final detection result obtaining subunit is configured to obtain a final detection result of the watermark for the carrier object to be detected according to the plurality of preliminary watermark detection results for the carrier object to be detected.

Optionally, the device further includes: a confidence degree obtaining unit, configured to obtain the confidence degree of the preliminary detection result for each watermark;

The final detection result obtains the subunit, which is specifically used for:

Optionally, the final detection result obtains the subunit, which is specifically used for:

Optionally, the preliminary detection result obtains the subunit, which is specifically used for:

The category information of the watermark in the carrier object to be detected;

Position information of the watermark in the carrier object to be detected.

It should be noted that for the detailed description of the watermark detection device provided in the seventh embodiment of the present application, reference may be made to the related description of the second embodiment of the present application, which is not repeated here.

Corresponding to the watermark detection method provided in the second embodiment of this application, the eighth embodiment of this application also provides an electronic device.

As shown in FIG. 14, the electronic device includes:

Processor 1201;

The memory 1202 is used to store the program of the watermark detection method. After the device is powered on and runs the program of the watermark detection method through the processor, the following steps are executed:

Obtain the carrier object to be detected;

Optionally, the electronic device further performs the following steps:

Obtain the confidence of the preliminary detection results for each watermark;

The category information of the watermark in the carrier object to be detected;

Position information of the watermark in the carrier object to be detected.

It should be noted that, for a detailed description of the electronic device provided in the eighth embodiment of the present application, reference may be made to the related description of the second embodiment of the present application, which will not be repeated here.

Corresponding to the watermark detection method provided in the second embodiment of this application, the ninth embodiment of this application also provides a storage device that stores a program of the watermark detection method, and the program is run by a processor to perform the following steps:

Obtain the carrier object to be detected;

It should be noted that, for the detailed description of the storage device provided in the ninth embodiment of the present application, reference may be made to the related description of the second embodiment of the present application, which will not be repeated here.

Although this application is disclosed as above in preferred embodiments, it is not intended to limit the application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the application. Therefore, this application The scope of protection shall be subject to the scope defined by the claims of this application.

In a typical configuration, the computing device includes one or more processors (CPU), input/output interfaces, network interfaces, and memory.

The memory may include non-permanent memory in a computer readable medium, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer readable media.

Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include non-transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

Claims

A method for generating a watermark detection model, which is characterized in that it includes:

Obtain multiple watermark data sets; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the difference between the size information of the watermark template and the size information of the carrier object The watermark template is a template used when adding watermark information to the carrier object;

According to the multiple watermark data sets, multiple watermark detection branch modules are generated; each watermark detection branch module is used to obtain a preliminary watermark detection result for the carrier object based on the specified watermark embedding strength information and the specified size relationship information;

A watermark detection model is generated according to the multiple watermark detection branch modules; the watermark detection model is used to obtain a final watermark detection result for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules.
The method according to claim 1, wherein said obtaining a watermark data set comprises:

Obtain a carrier object set; wherein, each carrier object in the same carrier object set has the same watermark embedding strength information and the same size relationship information; each carrier object in a different carrier object set has different watermark embedding strength information and different size relationship information;

Obtain the label information of the carrier object set;

According to the carrier object set and the label information, multiple watermark data sets are generated.
The method according to claim 2, further comprising:

Save the label information in a text file;

Converting the text file into a non-text file in a predetermined format;

The generating multiple watermark data sets according to the carrier object set and the label information includes:

According to the carrier object set and the non-text file, multiple watermark data sets are generated.
The method according to claim 3, wherein the label information includes at least one of the following:

Type information of the watermark in the carrier object;

The position information of the watermark embedded in the carrier object;

The name information of the watermark in the carrier object.
The method according to claim 4, wherein the generating multiple watermark detection branch modules according to the multiple watermark data sets comprises:

Input the carrier object in the watermark data set into the convolutional layer in the neural network model, and extract the feature map of the carrier object;

Input the feature map of the carrier object into the candidate area network in the neural network model to generate a recommended area containing a watermark;

Input the feature map of the carrier object and the recommended region into the region of interest pooling layer in the neural network model to obtain a feature map of the recommended region;

Input the recommended area feature map to the classifier to obtain the test value of the watermark category in the carrier object;

Training a neural network model according to the test value and the label information to generate a watermark detection branch module.
The method according to claim 1, wherein the number of the watermark detection branch is equal to the number of the watermark data set; one watermark data set is used to generate a watermark detection branch module.
A watermark detection method, characterized in that it comprises:

Obtain the carrier object to be detected;

The carrier object to be detected is input into the watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to obtain the target according to the watermark embedding strength information and size relationship information of the carrier object to be detected The watermark detection result of the carrier object to be detected.
The method according to claim 7, wherein the inputting the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected comprises:

Inputting the carrier object to be detected into a watermark detection model to obtain a plurality of preliminary watermark detection results for the carrier object to be detected;

According to the plurality of preliminary watermark detection results for the carrier object to be detected, a final watermark detection result for the carrier object to be detected is obtained.
The method according to claim 8, further comprising: obtaining the confidence of the preliminary detection result for each watermark;

The obtaining the final detection result of the watermark for the carrier object to be detected according to the preliminary detection results of the plurality of watermarks for the carrier object to be detected includes:

According to the preliminary detection results of the multiple watermarks and the confidence of the preliminary detection results for each watermark, a final detection result of the watermark for the carrier object to be detected is determined.
The method according to claim 9, wherein the determining the final detection result of the watermark for the carrier object to be detected according to the preliminary detection results of the multiple watermarks and the confidence of the preliminary detection results for each watermark, include:

The preliminary detection result of the watermark with the greatest confidence is used as the final detection result of the watermark.
The method according to claim 8, wherein inputting the carrier object to be detected into a watermark detection model to obtain multiple preliminary watermark detection results for the carrier object to be detected comprises:

Inputting the carrier object to be detected into the convolutional layer of the watermark detection model, and extracting a feature map of the carrier object to be detected;

Input the feature map of the carrier object to be detected into the candidate area network of the watermark detection model to generate a recommended area containing a visible watermark;

Input the feature map of the carrier object to be detected and the recommended region into the interest region pooling layer of the watermark detection model to obtain a feature map of the recommended region;

The recommended area feature map is input into a classifier to obtain preliminary results of watermark detection in a plurality of carrier objects to be detected.
The method according to claim 7, wherein the preliminary detection result of the watermark includes at least one of the following information:

The category information of the watermark in the carrier object to be detected;

Position information of the watermark in the carrier object to be detected.
A watermark detection method, characterized in that it comprises:

Obtain the carrier object to be detected;

Obtain the watermark embedding strength information and size relationship information of the carrier object to be detected. The size relationship information is used to indicate the relationship between the size information of the watermark template and the size information of the carrier object. The template used when the carrier object adds watermark information;

According to the watermark embedding strength information and the size relationship information, a watermark detection result for the carrier object to be detected is obtained.
A device for generating a watermark detection model is characterized in that it comprises:

The watermark data set obtaining unit is used to obtain multiple watermark data sets; wherein the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the size of the watermark template The relationship between information and size information of the carrier object, where the watermark template is a template used when adding watermark information to the carrier object;

The watermark detection branch generating unit is configured to generate multiple watermark detection branch modules according to the multiple watermark data sets; each watermark detection branch module is used to obtain the target carrier object based on the specified watermark embedding strength information and the specified size relationship information Preliminary detection results of the watermark;

The watermark detection model generation unit is configured to generate a watermark detection model according to the multiple watermark detection branch modules; the watermark detection model is used to obtain the final watermark for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules Test results.
An electronic device, characterized in that it comprises:

processor;

The memory is used to store the program of the method for generating the watermark detection model. After the device is powered on and runs the program of the method for generating the watermark detection model through the processor, the following steps are executed:

Obtain multiple watermark data sets; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the difference between the size information of the watermark template and the size information of the carrier object The watermark template is a template used when adding watermark information to the carrier object;

According to the multiple watermark data sets, multiple watermark detection branch modules are generated; each watermark detection branch module is used to obtain a preliminary watermark detection result for the carrier object based on the specified watermark embedding strength information and the specified size relationship information;

A watermark detection model is generated according to the multiple watermark detection branch modules; the watermark detection model is used to obtain a final watermark detection result for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules.
A storage device, characterized in that:

A program that stores a method for generating a watermark detection model, and the program is run by the processor to perform the following steps:

Obtain multiple watermark data sets; wherein, the carrier objects in the same watermark data set have the same watermark embedding strength information and the same size relationship information, and the size relationship information is used to indicate the difference between the size information of the watermark template and the size information of the carrier object The watermark template is a template used when adding watermark information to the carrier object;

According to the multiple watermark data sets, multiple watermark detection branch modules are generated; each watermark detection branch module is used to obtain a preliminary watermark detection result for the carrier object based on the specified watermark embedding strength information and the specified size relationship information;

A watermark detection model is generated according to the multiple watermark detection branch modules; the watermark detection model is used to obtain a final watermark detection result for the carrier object according to the preliminary watermark detection results of the multiple watermark detection branch modules.
A watermark detection device, characterized in that it comprises:

A carrier object to be detected obtaining unit for obtaining a carrier object to be detected;

The watermark detection result obtaining unit is configured to input the carrier object to be detected into a watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to embed the watermark according to the carrier object to be detected The intensity information and the size relationship information obtain a watermark detection result for the carrier object to be detected.
An electronic device, characterized in that it comprises:

processor;

The memory is used to store the program of the watermark detection method. After the device is powered on and runs the program of the watermark detection method through the processor, the following steps are executed:

Obtain the carrier object to be detected;

The carrier object to be detected is input into the watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to obtain the target according to the watermark embedding strength information and size relationship information of the carrier object to be detected The watermark detection result of the carrier object to be detected.
A storage device, characterized in that:

The program of the watermark detection method is stored, and the program is run by the processor to perform the following steps:

Obtain the carrier object to be detected;

The carrier object to be detected is input into the watermark detection model to obtain a watermark detection result for the carrier object to be detected; the watermark detection model is used to obtain the target according to the watermark embedding strength information and size relationship information of the carrier object to be detected The watermark detection result of the carrier object to be detected.