A watermark recognition method and recognition system based on artificial intelligence

By using a TensorFlow-based dataflow neural network model, the problem of low discrimination in watermark recognition is solved, enabling accurate recognition and editing prompts for watermarks in images or documents, thus improving the accuracy and efficiency of watermark recognition.

CN113837914BActive Publication Date: 2024-11-29BEIJING KINGSOFT OFFICE SOFTWARE INC +1

Patent Information

Application Number
CN202010511115.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-06-08
Publication Date
2024-11-29
Estimated Expiration
2040-06-08

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively identify watermarks in images or documents, especially due to the low visual saliency of watermarks, resulting in low distinguishability and making accurate watermark recognition impossible.

Method used

A dataflow neural network model based on TensorFlow is used to identify watermarks in images or documents through feature extraction and classifier training. Multiple watermarked and watermark-free images are used as training data to build a dataflow neural network classifier, and images are labeled during the recognition process to distinguish between watermarked and watermark-free images.

Benefits of technology

It achieves accurate recognition of watermarked images or documents, and can detect them on each page of the document, returning results to prompt the user whether the watermark needs to be edited.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113837914B_ABST
    Figure CN113837914B_ABST
Patent Text Reader

Abstract

The present application proposes a watermark recognition method and system based on artificial intelligence, which relates to the field of document visual recognition. The method includes: obtaining a target image to be recognized; extracting features of the target image to be recognized to obtain multiple target feature information; based on a data stream neural network classifier, recognizing the multiple target feature information to determine whether the target image to be recognized is a watermarked image; wherein the data stream neural network classifier is trained and acquired using multiple watermarked images and multiple watermark-free images. The present invention uses a network neural model to detect documents or images, detects part or all pages in the document, converts each page into an image in units of pages (this step is not required if it is in image format), inputs it into a trained neural network classifier, identifies whether it contains a watermark, and after the identification is completed, prompts the user whether it needs to edit the watermark.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of document visual recognition, and in particular to a watermark recognition method and recognition system based on artificial intelligence. Background Art

[0002] When documents or images are transmitted on the Internet, due to the speed and portability, more and more individuals and organizations choose to embed watermark information in their images or documents, such as watermarking a trademark or URL in part of the image, and watermarking a trademark or "draft" in the background of the document, in order to protect the ownership of the image or document. Therefore, before using or disseminating an image or document, it is necessary to review the image or document to identify whether it contains watermark information to avoid misuse and infringement. With the rapid development of the Internet, image or document providers use many ways such as user uploads or crawler downloads to obtain a large amount of image or document information every day, and the amount has far exceeded the limit of manual review. Therefore, it has become an urgent need to use computers to automatically review images or documents and identify whether they contain watermarks.

[0003] The visual significance of watermark information in an image or document is very low, and it has the characteristics of small area, light color, and high transparency. The difference between a watermarked image or document and an unwatermarked image or document is often very small, and the distinction is low.

[0004] At present, there are few effective watermark recognition technologies that can achieve accurate recognition of watermarked images or documents. Summary of the invention

[0005] The present application provides an artificial intelligence-based watermark recognition method and recognition system to achieve accurate recognition of watermarked images or documents.

[0006] The technical solutions adopted are as follows:

[0007] The present invention provides a watermark recognition method based on artificial intelligence, comprising:

[0008] Get the target image to be identified;

[0009] Extracting features of the target image to be identified to obtain multiple target feature information;

[0010] Based on the data stream neural network classifier, the plurality of target feature information are identified to determine whether the target image to be identified is a watermarked image;

[0011] The data stream neural network classifier is obtained by training with multiple watermarked images and multiple non-watermarked images.

[0012] Preferably, the method further comprises:

[0013] The target to be identified is formatted and converted into a picture format to obtain the target picture to be identified.

[0014] Preferably, the method further comprises:

[0015] Build a data stream neural network classifier.

[0016] Preferably, establishing a data stream neural network classifier includes:

[0017] Generate image training data; the image training data includes multiple watermarked images and multiple non-watermarked images;

[0018] The image training data is used to establish a data stream neural network classifier through a data stream neural network model.

[0019] Preferably, generating image training data includes:

[0020] Get multiple watermarked pictures;

[0021] Perform feature extraction on each of the watermarked images to obtain watermarked image training data;

[0022] Get multiple pictures without watermark;

[0023] Perform feature extraction on each of the watermark-free images to obtain watermark-free image training data;

[0024] Each of the watermarked image training data and each of the non-watermarked image training data are used as image training data.

[0025] Preferably, the method further comprises:

[0026] When the target image to be identified is a watermarked image, marking the target image to be identified using a first mark;

[0027] When the target image to be identified is a watermark-free image, the target image to be identified is marked with a second mark.

[0028] Preferably, the method further comprises:

[0029] The user is prompted whether to edit the watermark of the target image to be identified.

[0030] Preferably, editing the watermark of the target image to be identified includes:

[0031] Remove watermark from watermarked pictures or add watermark to non-watermarked pictures.

[0032] The present invention also provides a watermark recognition system based on artificial intelligence, comprising:

[0033] Interception module, obtains the target image to be identified;

[0034] A feature extraction module is configured to extract features from the target image to be identified to obtain multiple target feature information;

[0035] A detection module is configured to identify the plurality of target feature information based on a data stream neural network classifier, and determine whether the target image to be identified is a watermarked image;

[0036] The data stream neural network classifier is obtained by training using a plurality of watermarked images and a plurality of non-watermarked images.

[0037] Preferably, the system further comprises:

[0038] The training module is configured to generate image training data; the image training data includes a plurality of watermarked images and a plurality of non-watermarked images; and the image training data is used to establish a data stream neural network classifier through a data stream neural network model.

[0039] Preferably, the system further comprises:

[0040] The marking module is configured to mark the target image to be identified with a first mark when the target image to be identified is a watermarked image; and mark the target image to be identified with a second mark when the target image to be identified is a watermark-free image.

[0041] Compared with the prior art, this application has the following beneficial effects:

[0042] Compared with the prior art, the present invention uses TensorFlow, a network neural model, to detect documents or pictures, detect part or all pages in the document, convert each page into a picture (if it is in picture format, this step is not required), and input it into the trained neural network classifier. If the result returned is 1, it means that there is a watermark, and if the result returned is 0, it means that there is no watermark. When the recognition is completed, the user can be prompted whether to edit the watermark. Other features and advantages of the present invention will be described in the subsequent description, and part of them will become obvious from the description, or can be understood by practicing the present invention. The objects and other advantages of the present invention can be achieved and obtained through the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a watermark recognition method based on artificial intelligence according to an embodiment of the present invention;

[0044] Figure 2It is a schematic diagram of the structure of the watermark recognition system based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present application will be described in more detail below with reference to the accompanying drawings and embodiments.

[0046] It should be noted that, if there is no conflict, the embodiments of the present application and the various features in the embodiments can be combined with each other, all within the protection scope of the present application. In addition, although the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order than here.

[0047] like Figure 1 As shown, the embodiment of the present invention provides a watermark recognition method based on artificial intelligence, comprising:

[0048] S101, obtaining a target image to be identified;

[0049] S102, the image training data includes a plurality of watermarked images and a plurality of non-watermarked images;

[0050] S103, based on the data stream neural network classifier, identifying the multiple target feature information, and determining whether the target image to be identified is a watermarked image;

[0051] The data stream neural network classifier is obtained by training with multiple watermarked images and multiple non-watermarked images.

[0052] In the embodiment of the present invention, the method further comprises:

[0053] The target to be identified is formatted and converted into a picture format to obtain the target picture to be identified.

[0054] In the embodiment of the present invention, the method further comprises:

[0055] Build a data stream neural network classifier.

[0056] In an embodiment of the present invention, a data stream neural network classifier is established, including:

[0057] Generate image training data; the image training data includes multiple watermarked images and multiple non-watermarked images;

[0058] The image training data is used to establish a data stream neural network classifier through a data stream neural network model.

[0059] In an embodiment of the present invention, generating image training data includes:

[0060] Get multiple watermarked pictures;

[0061] Perform feature extraction on each of the watermarked images to obtain watermarked image training data;

[0062] Get multiple pictures without watermark;

[0063] Perform feature extraction on each of the watermark-free images to obtain watermark-free image training data;

[0064] Each of the watermarked image training data and each of the non-watermarked image training data are used as image training data.

[0065] The embodiment of the present invention adopts the neural network model TensorFlow to detect documents. When the network model is trained, the documents are exported into pictures, those without watermarks are marked as 0, and those with watermarks are marked as 1. Each picture and its mark are input into the neural network model for training.

[0066] In the embodiment of the present invention, the method further comprises:

[0067] When the target image to be identified is a watermarked image, marking the target image to be identified using a first mark;

[0068] When the target image to be identified is a watermark-free image, the target image to be identified is marked with a second mark.

[0069] In the embodiment of the present invention, the method further comprises:

[0070] The user is prompted whether to edit the watermark of the target image to be identified.

[0071] In the embodiment of the present invention, watermark editing of the target image to be identified includes:

[0072] Remove watermark from watermarked pictures or add watermark to non-watermarked pictures.

[0073] In an embodiment of the present invention, the network neural model TensorFlow is used to detect documents or pictures, and some or all pages in the document are detected. Each page is converted into a picture (if it is in picture format, this step is not required), and input into the trained neural network classifier. The returned result is 1, which means that there is a watermark, and the returned result is 0, which means that there is no watermark. After the recognition is completed, the user can be prompted whether to edit the watermark.

[0074] like Figure 2 As shown, the embodiment of the present invention also provides a watermark recognition system based on artificial intelligence, including:

[0075] The interception module 100 obtains the target image to be identified;

[0076] The feature extraction module 200 is configured to extract features from the target image to be identified to obtain a plurality of target feature information;

[0077] The detection module 300 is configured to identify the plurality of target feature information based on a data stream neural network classifier, and determine whether the target image to be identified is a watermarked image;

[0078] The data stream neural network classifier is obtained by training using a plurality of watermarked images and a plurality of non-watermarked images.

[0079] The system according to the embodiment of the present invention further includes:

[0080] The training module is configured to generate image training data; the image training data includes a plurality of watermarked images and a plurality of non-watermarked images; and the image training data is used to establish a data stream neural network classifier through a data stream neural network model.

[0081] The system according to the embodiment of the present invention is characterized in that it also includes:

[0082] The marking module is configured to mark the target image to be identified with a first mark when the target image to be identified is a watermarked image; and mark the target image to be identified with a second mark when the target image to be identified is a watermark-free image.

[0083] Although the embodiments disclosed in the present invention are as above, the contents are only embodiments adopted for facilitating understanding of the technical solution of the present invention, and are not used to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the core technical solution disclosed in the present invention, but the protection scope defined by the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A watermark recognition method based on artificial intelligence, characterized in that: include: Get the target image to be identified; Extracting features of the target image to be identified to obtain multiple target feature information; Based on the data stream neural network classifier, the plurality of target feature information are identified to determine whether the target image to be identified is a watermarked image; The establishment of the data stream neural network classifier includes: exporting the document into pictures, performing feature extraction on each of a plurality of watermarked pictures to obtain watermarked picture training data, performing feature extraction on each of a plurality of non-watermarked pictures to obtain non-watermarked picture training data, and using each of the watermarked picture training data and each of the non-watermarked picture training data as picture training data; setting a second mark for a picture without a watermark, setting a first mark for a picture with a watermark, and inputting each picture and the corresponding first mark or second mark into the data stream neural network model to train it; When the target image to be identified is a watermarked image, outputting the first mark; when the target image to be identified is a watermark-free image, outputting the second mark, comprising: when the target image to be identified is a watermarked image, marking the target image to be identified with the first mark, and outputting the first mark; when the target image to be identified is a watermark-free image, marking the target image to be identified with the second mark, and outputting the second mark; The method then further includes: prompting the user whether to perform watermark editing on the target image to be identified, including: removing the watermark from the watermarked image or adding a watermark to the non-watermarked image.

2. The method according to claim 1, characterized in that: The method also includes: The target to be identified is formatted and converted into a picture format to obtain the target picture to be identified.

3. A watermark recognition system based on artificial intelligence, characterized in that: include: The training module is configured to establish a data stream neural network classifier in the following manner: exporting the document into images, performing feature extraction on each of a plurality of watermarked images to obtain watermarked image training data, performing feature extraction on each of a plurality of non-watermarked images to obtain non-watermarked image training data, and using each of the watermarked image training data and each of the non-watermarked image training data as image training data; Setting a second mark for a picture without a watermark, setting a first mark for a picture with a watermark, inputting each picture and the corresponding first mark or second mark into a data stream neural network model to train it; Interception module, obtains the target image to be identified; A feature extraction module is configured to extract features from the target image to be identified to obtain multiple target feature information; A detection module, configured to identify the plurality of target feature information based on a data stream neural network classifier, determine whether the target image to be identified is a watermarked image, and output the first mark when the target image to be identified is a watermarked image; When the target image to be identified is a watermark-free image, outputting the second mark; a marking module, configured to mark the target image to be identified with a first mark when the target image to be identified is a watermarked image; When the target image to be identified is a watermark-free image, marking the target image to be identified using a second mark; The system is further configured to prompt the user whether to edit the watermark of the target image to be identified in the following manner: removing the watermark from the watermarked image or adding a watermark to the non-watermarked image.

Citation Information

Patent Citations

  • Method for identifying image with watermark and identifying system

    CN106096668A

  • Automatic image watermark detection method

    CN107808358A

  • Method and device for automatically removing watermarks of picture

    CN110599387A

Cited By

  • Method of verification for machine learning models

    US20240388444A1