A carrier-free ownership marking method for cartoon images based on blockchain authentication

Abstract

The present invention discloses a carrier-free ownership marking method for cartoon images based on blockchain authentication: when information is embedded, the secret information is first converted into a set of attribute tags of the cartoon image, and then driven by the tag set, a cartoon image expressing the secret information is directly generated using a generative adversarial network: when information is verified, the attribute information of the cartoon image is first extracted, and then the transmitted secret information is obtained according to the attribute information. The present invention can effectively prevent malicious thieves from discovering the ownership marking information, so that the thieves have no chance to delete or tamper with the ownership information; the ownership information is expressed using the attribute information of the cartoon image, which has better robustness than the underlying feature information; arbitrary ownership information can be expressed by first searching, and if no information is found, the regeneration method can be used; the ownership image is stored by blockchain, and the authentication is accurate and reliable based on the unchangeable characteristics of blockchain.

Description

Technical Field

[0001] The present invention relates to the technical field of blockchain and digital forensics, and in particular to a carrier-free ownership marking method for cartoon images based on blockchain authentication. Background Art

[0002] In the era of big data, there are many big data sets composed of the same type of metadata on the Internet, and these data may have huge commercial value. This patent mainly considers the protection of the ownership of animation image big data. As the main material for animation re-creation, it is of great significance to determine the ownership of animation image big data. The traditional data copyright authentication method mainly embeds information that can identify the ownership of the data in the data, namely digital watermark technology. According to the embedding method of digital watermarks, it can be divided into visible watermarks and invisible watermarks. Visible watermarks mainly refer to adding marking information that does not belong to the image itself to the digital image. The marking information is visible on the image, usually the stamp, email address, QR code, etc. of the image that can show the identity of the owner. Invisible watermarks refer to the embedding of ownership information by modifying the image to be protected. Generally speaking, the modified image is difficult to distinguish from the original image with the naked eye, and there is no obvious difference.

[0003] However, for big data images, existing methods are not very practical. First, both visible and invisible watermarks will affect the overall characteristics of the data. If each image in the data set contains watermark information, it is difficult for the image creation model trained using such a data set to get rid of the influence of the watermark information and generate anime images that meet expectations when re-creating. Secondly, data thieves can use existing technology to remove visible watermarks, making it difficult for the owner to provide evidence. In addition, for invisible watermarks, there is a certain possibility that it is difficult to correctly extract the ownership information after the image has been copied, compressed, and contaminated by noise for multiple times. The key is that both visible and invisible watermarks will leave traces of the watermark, which can easily attract the attention of thieves, and they will try to remove the information that can express ownership when stealing. Summary of the invention

[0004] In order to authenticate the ownership information of animation big data by using blockchain and carrier-free information hiding technology without making any modification to the animation image big data, the present invention provides a carrier-free ownership marking method for animation images based on blockchain authentication.

[0005] The technical solution adopted by the present invention is as follows: a method for marking the ownership of cartoon images without carriers based on blockchain authentication, including an ownership information embedding part and an ownership information verification part; the steps of the ownership information embedding part are as follows:

[0006] Step S1.1. Convert the ownership information of the cartoon image into binary ownership information;

[0007] Step S1.2. Convert the binary ownership information into a set of attribute labels of the animation image;

[0008] Step S1.3. Search for a matching image as the ownership information image in the animation image big data for ownership authentication according to the obtained attribute tag set. If found, go to step S1.5. If not found, go to step S1.4.

[0009] Step S1.4. Generate an ownership information image that conforms to the attribute tag set;

[0010] Step S1.5. Upload the ownership information image and ownership addition time information to the private blockchain of the ownership certification center.

[0011] Preferably, the steps of the ownership information verification part are as follows:

[0012] Step S2.1. Download the ownership information image from the authentication center blockchain;

[0013] Step S2.2. Use the approximate image to search whether there is any ownership information image in the image database to be verified. If so, go to step S2.3. If not, the verification ends and ownership verification cannot be performed.

[0014] Step S2.3. Extracting the attribute label set of the ownership information image;

[0015] Step S2.4. Obtain binary ownership information by converting the attribute tag set;

[0016] Step S2.5. Decode the binary ownership information to obtain the ownership information.

[0017] The present invention has the following beneficial effects:

[0018] 1. Compared with the existing digital watermarking methods, it can effectively prevent malicious thieves from discovering the ownership mark information, so that thieves have no chance to delete or tamper with the ownership information;

[0019] 2. Using the attribute information of the animation image (high-level semantic features of the image) to express ownership information has better robustness compared with the underlying feature information;

[0020] 3. By searching first and then regenerating if no information is found, any ownership information can be expressed, while the existing search-based carrier-free information hiding method is difficult to express any ownership information;

[0021] 4. The present invention stores ownership images through blockchain, and based on the unchangeable characteristics of blockchain, the authentication is accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is the ownership information embedding process in the embodiment of the present invention.

[0023] Figure 2 The ownership information verification process in the embodiment of the present invention.

[0024] Figure 3 An embodiment of the conversion between ownership information and cartoon image attribute labels in an embodiment of the present invention.

[0025] Figure 4 The ownership information in the embodiment of the present invention represents an embodiment.

[0026] Figure 5 An original image and a generated image embodiment in an embodiment of the present invention.

[0027] Figure 6 A global module diagram of the system according to an embodiment of the present invention.

[0028] Figure 7 The basic functional flow of the system of the embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be further described below in conjunction with the embodiments and drawings.

[0030] This embodiment is a carrier-free ownership marking method for cartoon images based on blockchain authentication, which is used for the ownership protection of a large database of cartoon images. In principle, the blockchain is a decentralized distributed database, which is connected by several data blocks and stores open, transparent and unchangeable data information. Therefore, the data on the blockchain has an inherent advantage in the protection of ownership. In addition, the blockchain forms are divided into public chains, alliance chains and private chains, and their degree of decentralization is different. Accordingly, a suitable scheme can be selected according to the ownership authentication requirements of the cartoon images. At the same time, in order not to leave traces of modification on the images in the original large database, a carrier-free information hiding method is used to express the ownership information of the cartoon images. It should be noted that the carrier-free information hiding method does not mean that a carrier is not required, but that no modification is made to the carrier image, and an image that can express information is directly retrieved or generated according to the information to be expressed. The method of this embodiment first searches the database for images that can represent the encoded ownership information. If there is an image, it is directly uploaded to the blockchain of the ownership authentication center for storage. If not, an image that cannot be visually distinguished from the database is generated according to the ownership information, and it is uploaded to the blockchain as an ownership image.

[0031] The method and system of this embodiment include two parts: the copyright holder and the verification party. The global module diagram and basic function flow of the system are as follows: Figure 6 , 7 As shown. According to the function system, it can be divided into the following modules: ownership information conversion module, animation image label extraction module, label conversion module, ownership image retrieval module, ownership image generation module, blockchain access module. The specific technical solutions of each module are introduced below.

[0032] Ownership information conversion module: The purpose of this module is to convert big data ownership information in the form of text or images into a binary data stream through encoding conversion.

[0033] Cartoon image label extraction module: This embodiment uses the attribute label of the cartoon image to express the ownership information of the cartoon image. Specifically, the attribute label is predicted using the cartoon image label recognition model. The model is improved so that the model can detect the specified attribute label.

[0034] Label conversion module: The purpose of this module is to convert binary ownership information into image attribute information through a deep learning prediction model. It mainly obtains a sorted probability space through the prediction model, and then selects the attribute labels corresponding to the sorting position according to the ownership information fragment to form a label set.

[0035] Ownership image retrieval module: This module mainly uses the label conversion module to convert the ownership information, and then retrieves the corresponding image in the animation image database based on the label information combination.

[0036] Ownership image generation module: This embodiment uses the animation image database that needs to be protected and its corresponding animation image attribute labels as the training set to train the animation image generation model, with the aim of making the generated animation images that can express ownership information indistinguishable from the animation image data that needs to be protected.

[0037] Blockchain access module: This embodiment uses blockchain storage to further improve the credibility of ownership authentication. This module mainly stores the retrieved or generated ownership information image and the ownership generation time and other information on the private blockchain of the ownership authentication center through hash coding. When ownership authentication is required, the ownership information image is downloaded from the private blockchain of the ownership authentication center, and the data set to be verified is compared to see whether it contains the ownership image.

[0038] like Figure 1 The figure shows the ownership information embedding process. The specific implementation steps are introduced as follows.

[0039] Step 1: Use the ownership information conversion module to convert the ownership information into binary code.

[0040] This embodiment uses BCH coding to convert ownership information and binary information. BCH coding is used to convert ownership certification information (OCI) into binary information during embedding. BCH code can not only convert ownership information and binary information, but also perform error correction when converting binary information, so as to improve the accuracy of ownership authentication in this embodiment.

[0041] Step 2: Convert binary information into attribute labels of animation images through label conversion model.

[0042] This embodiment uses a long short-term memory network (LSTM) to convert binary information and attribute labels. LSTM can calculate the probability space of the next output given a partial input. The present invention uses an animation image label extraction module to extract the attribute information of the animation image library from the animation image big data that needs to be authenticated for ownership authentication for training. This embodiment uses the animation image label prediction model Illustration2Vec to extract the attribute labels of animation images in the data set for training LSTM. Among them, this embodiment improves Illustration2Vec so that it can predict several specific types of features, such as hairstyle, hair color, eye color, etc., according to actual needs. Then, as shown in the following example, Figure 3 As shown in the figure, the trained LSTM model is used to convert the ownership information and the attribute labels of the animation image. First, select the first attribute label in the first attribute mapping table according to the ownership information. As input, the probability space of the next attribute label is calculated through LSTM. In this embodiment, the attribute label depends on the order of the ownership information fragments in the probability space. For example, the corresponding attribute is hair color, which represents 3-bit ownership information. The probability of the next attribute generated by LSTM for the input can be expressed as: , because 3 bits of secret information need to be represented, 8 labels are selected as candidate sets according to certain rules. This paper selects the top 8 labels in probability ranking, and then selects the corresponding ranked labels according to the secret information as In this embodiment, "001" corresponds to "black hair" which is ranked second. And so on, all attribute tag sets are obtained.

[0043] Step 3: According to the obtained attribute tag set, search for matching images as ownership information images in the animation image big data that needs to be authenticated. If found, go to step 5; if not found, go to step 4.

[0044] Step 4: Generate an animated image that meets the attribute label set through the animated image generation network based on the attribute label set. This embodiment is based on ACGAN (Auxiliary Classifier Generative adversarial networks), and uses the animated image database that needs to be protected and its corresponding animated image attribute labels as the training set to train the animated image generation model. The purpose is to make the generated animated image that can express the ownership information indistinguishable from the animated image data that needs to be protected. At the same time, this embodiment adds a post-verification module to ACGAN, which will evaluate the image quality and select animated images with better quality to express the ownership information on the basis of ensuring that it can be correctly extracted. The generated image is similar to the original image in the database. Figure 5 After the ownership information image is generated, it is stored in the animation image big data, and any image is deleted immediately.

[0045] Step 5: Upload the ownership information image and ownership addition time and other information to the private blockchain of the ownership certification center.

[0046] like Figure 2 The following is the ownership information verification process. The specific steps are as follows:

[0047] Step 1: Download the ownership information image from the certification center blockchain.

[0048] Step 2: Use the approximate image retrieval solution to search whether there are any ownership information images in the image database to be verified. If there are, go to step 3. If not, the verification ends and ownership verification cannot be performed.

[0049] Step 3: Use Illustration2Vec that is consistent with the ownership information embedding method to extract the attribute label set of the ownership image.

[0050] Step 4: Use the LSTM that is consistent with the ownership information embedding method to transform the attribute label set to obtain binary ownership information.

[0051] Step 5: Use the BCH encoding method consistent with the ownership information embedding party to decode the binary information to obtain the ownership information.

[0052] Obviously, the above embodiments of the present invention are only examples for explaining the present invention, and are not intended to limit the implementation methods of the present invention. Obvious changes or modifications derived from the essence of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for marking the ownership of cartoon images without carriers based on blockchain authentication. It is characterized in that It includes the ownership information embedding part and the ownership information verification part; The steps of embedding the ownership information are as follows: Step S1.

1. Convert the ownership information of the cartoon image into binary ownership information; Step S1.

2. Convert the binary ownership information into a set of attribute labels of the animation image; Step S1.

3. Search for a matching image as the ownership information image in the animation image big data for ownership authentication according to the obtained attribute tag set. If found, go to step S1.

5. If not found, go to step S1.

4. Step S1.

4. Generate an ownership information image that conforms to the attribute tag set; Step S1.

5. Upload the ownership information image and ownership addition time information to the private blockchain of the ownership certification center; The steps of the ownership information verification part are as follows: Step S2.

1. Download the ownership information image from the authentication center blockchain; Step S2.

2. Use the approximate image to search whether there is any ownership information image in the image database to be verified. If so, go to step S2.

3. If not, the verification ends and ownership verification cannot be performed. Step S2.

3. Extracting the attribute label set of the ownership information image; Step S2.

4. Obtain binary ownership information by converting the attribute tag set; Step S2.

5. Decode the binary ownership information to obtain the ownership information.

Images