Exploration of Computer Vision Technology and Legal Consideration of Online False Propaganda Behavior in the Perspective of Civil and Commercial Law

Under the background of digital economy, the rapid change of network information technology promotes the rapid development of e-commerce, with the help of the efficiency of the network communication media and the virtual nature of the characteristics of the network propaganda behavior of the manifestation of the forms of behavior gradually presents diversified characteristics, and along with it, the network of false propaganda behaviors are also endless [1-2]. As some e-commerce platforms have a low threshold of entry, easy to operate, high income characteristics, making the unscrupulous merchants take advantage of the situation, the industry competition is getting more and more intense. They are more in order to quickly seize the market, crowding out competitors, the pursuit of higher economic benefits, etc. thus ignoring the legal requirements, the implementation of the disguised exaggerated product performance and quality, brush speculation and other false propaganda behaviors, has become a breeding ground for network fraud, a serious infringement on the legitimate rights and interests of consumers, the rights and interests of property, and the order of the network business has been a great destructive force, some of which are even on the competitor’s goods to make Malicious evaluation, personal attacks, network violence and other false propaganda behavior [3-5]. Traditional offline false propaganda behavior through advertising or other ways other than advertising, its operation is simple and very public. Compared with the traditional false propaganda, the use of the network to carry out false propaganda requires strong professionalism but its covert nature, unscrupulous merchants use the network to update fast, difficult to forensic, timeliness and other hidden characteristics of the network false propaganda constantly implemented network false propaganda and other acts of unfair competition [6-7].

In network transactions, consumers can only understand the performance of goods or services through publicity, reviews, etc. and can not see with their own eyes, consumers in the process of online shopping in an unequal information disadvantaged position, seen and known and felt only through the network management body’s publicity and recommendations, often making consumers get a very poor shopping experience and quality of functional defects, and even damage to human health. The consumers often get very poor shopping experience and goods with defective quality and function, even damage to human health [8-9]. More due to the lack of systematic legal regulation of network platforms so that consumers do not know how to correctly claim their rights when their rights are violated, in the long run, the vicious circle, the platform’s regulatory incompetence condones the network merchants of this false propaganda of the crooked wind, so that consumers are losing confidence in the credibility of the platform more and more, affecting the healthy and prosperous development of the network market [10-11].

As an embodiment of unfair competition, online false propaganda behavior will infringe on the legitimate rights and interests of consumers, crowding out other network operators and destroying the order of fair competition in the market [12]. However, as a relatively new way of network publicity, due to the covert and diverse characteristics of network publicity behavior, the current Chinese legislation for the network false publicity behavior of civil liability provisions are still many deficiencies, unable to give full play to the civil liability for network false publicity behavior of the normative role of the legitimate rights and interests of consumers are difficult to effectively protect [13-14]. Therefore, civil liability as the current legislative level on the network false propaganda specification of the main way, based on the elements of civil liability, the network false propaganda civil liability systematic research and improvement will be very necessary [15-16].

Civil liability is a more effective way to regulate network false propaganda at the legislative level, for network false propaganda, the complexity of the type of subject and the behavior of the unspecified and other special characteristics of the current network false propaganda civil liability system still has certain deficiencies, the system of legislation revision speed is difficult to match the network technology development and media form, so that some of the use of the network as a means of false propaganda within the platform Operators can not be effectively sanctioned, the legitimate rights and interests of consumers as well as fair and standardized market competition order can not be strongly maintained [17-18].

In this paper, we have gained a deeper understanding of DirectShow technology, image binarization preprocessing algorithm, and template matching-based image recognition algorithm from the application and further research and algorithm practice. The article adopts the “first and last key extraction method” to extract key frames, and in the image binarization preprocessing, a highly adaptable and highly automated binarization preprocessing algorithm is used to preprocess the images of the video of the network propaganda behaviors, and based on the characteristics and requirements of the algorithm of this paper to identify the images, the image recognition algorithm is appropriately adjusted, and sequential similarity is used. Based on the characteristics and requirements of this paper’s algorithm to recognize images, the image recognition algorithm is properly adjusted, and the sequential similarity detection algorithm is adopted to achieve the purpose of excluding non-matching points in advance at a smaller threshold, saving the time of searching for false propaganda behaviors in the image, and improving the algorithm’s efficiency.

2

Application of Computer Vision Technology in Network Supervision

Computer vision is an important part of artificial intelligence, and its main task is to enable computers to “understand” images and videos. In radio and television monitoring, computer vision technology can be used to identify scenes, objects, characters and other information in the video, so as to realize the understanding and analysis of program content. For example, target detection algorithms can accurately and quickly identify specific content such as advertisement clips, violent scenes, etc., which can help supervisory authorities to detect violations in time.

Computer vision technology mainly uses deep learning algorithms, convolutional neural networks, and reinforcement learning.Deep learning is a machine learning method that imitates the neural network of the human brain and processes and analyzes large-scale data through the structure of multilevel neuron networks.Convolutional neural networks are a special type of deep learning network, mainly used for image recognition and image classification.Reinforcement learning is a method of learning through trial and error that enables computers to self-select actions and maximize rewards through interaction with the environment.Computer vision technology can enable real-time monitoring and analysis of IPTV program images through target recognition, behavior analysis, and other functions, and then detect illegal programs.

3

Video acquisition and key frame extraction technology

3.1

Key frame extraction method

Keyframes are key image frames used to describe a shot, which usually reflect the main content of the shot.The use of keyframes can lead to a significant reduction in the amount of matching computation. However, traditional keyframe extraction methods require relatively large and complex calculations. Considering the real-time requirements of the system, it is decided that the whole video is not segmented, using DirectShow technology, for the identification of false propaganda behavior in video advertisements, combined with the key frame extraction algorithms mentioned above, the key frame extraction algorithm will be used “first and last key frame extraction method”, the idea is as follows:

1) The first and last sections of the advertisement samples will be extracted with denser frames, in which the length of the first and last sections will be determined by the length of the advertisement samples.

2) Advertisements in the middle of the sample to take a longer interval, sparse key frame feature extraction, where the interval is also determined by the length of the advertisement sample.

3.2

Video Image Acquisition Based on DirectShow Technology

On the Windows system platform, DirectShow technology [19] is used to implement multimedia stream playback and frame extraction by shielding a series of issues such as media formats, data transfer, hardware differences, and synchronization, which greatly simplifies the development of multimedia applications.

DirectShow employs a system enumeration to utilize devices that are unknown in advance in the event of uncertain hardware configuration. Types of directories are represented by a CLSID. The application simply enumerates these types of directories to determine how many and what type of capture devices are installed on the system.

After specifying the capture device as the source filter, Direct Show’s Sample Grabber interface is used as the conversion filter, and the interface function is used to control the output data of the source filter, and image frames are captured at certain intervals according to the reference clock, the flow of which is shown in Figure 1.

After capturing the video frames, they are saved as bitmap files and the process is shown in Fig. 2.

So far, the framework of DirectShow technology has been introduced, and based on DirectShow technology, the acquisition of key frames of TV commercial videos and their storage in bitmap format have been realized.

4

Image binarization preprocessing algorithm

4.1

Basic concepts of digital images

Chromaticity theory suggests [20] that any color can be obtained by mixing the three basic colors of red, green, and blue in different proportions. Red, green, and blue are known as the three primary colors, or RGB primary colors for short. In a PC’s display system, the displayed image is made up of individual pixels whose colors are based on the RGB model. The combination of the three color values determines the color seen in the image.

For the purpose of this study, mainly planar images are considered. Each point on the plane consists of only two coordinate values, therefore, the planar image function is a continuous two-dimensional function, i.e., Equation (1): (1) $f (x, y) = {f_{r e d} (x, y), f_{g r e e n} (x, y), f_{b l u e} (x, y)}$

Where f denotes the color of the point with spatial coordinates (x,y), f_red, f_green, f_blue denotes the value of the color components of the three primary colors red, green and blue, respectively, at that point. They are all continuous functions of space, i.e., each point in continuous space has an exact value corresponding to it.

A digital image is an approximate representation of a continuous image f(x,y) and is usually represented by a matrix consisting of the values of the sampled points, as in equation (2).

(2)

[\begin{matrix} f (0, 0) & f (0, 1) & \dots & f (0, M - 1) \\ f (1, 0) & f (1, 1) & \dots & f (1, M - 1) \\ ⋮ & ⋮ & ⋮ \\ f (N - 1, 0) & f (N - 1, 1) & \dots & f (N - 1, M - 1) \end{matrix}]

Each sampling unit is called a pixel, and in the above equation, M and N are the total number of pixels in the horizontal and vertical directions of the digital image, respectively. Image files generally have different extensions depending on their digital image format.The most common image format is the bitmap format, which files have a.bmp extension.

The color depth of a digital image is determined by the number of binary bits used to represent the color value of each pixel. The greater the color depth, the greater the number of colors that can be represented. Different color depths produce different types of image files, and the following types of image files are commonly used on computers.

1)

Monochrome Image

Monochrome image of each pixel point occupies only one, the value of only “0” or “1”, “0” represents black, “1 “0” represents black and ‘1’ represents white.

2)

Grayscale Image

A grayscale image has the following characteristics:

Grayscale image storage file with an image color table, this color table has a total of 256 items, each table item in the image color table consists of red, green, blue color components, and red, green, blue color component values are equal, see formula (3).

(3)

f_{r e d} (x, y) = f_{g r e e n} (x, y) = f_{b i n e} (x, y)

Each pixel consists of 8 bits with values ranging from 0 to 255, representing 256 different gray levels. The pixel value f(x,y) of each pixel is the table entry address of the image color table.

3)

Pseudo-color image

Pseudo-color image is similar to grayscale image with image color table in its storage file, pseudo-color image has the following characteristics:

The red, green and blue color component values in the image color table are not all equal, see equation (4).

(4)

f_{r e d} (x, y) \neq f_{g r e e n} (x, y) \neq f_{b l u e} (x, y)

The whole image has only 256 colors, to represent 256 kinds of 8 different colors, the pixel must be composed of 8 bits, each pixel value is not directly determined by the value of each base color component, but rather, the image of the value of the cable as a table entry address of the image color table. An image with 256 colors is called an 8-bit color image. 256-color images have a photographic effect, more actually.

4)

24-bit true color image

24-bit true color image storage file does not have an image color table, which has the following characteristics:

Each pixel in the image out of R, G, B three components, each component accounts for 8 bits, each pixel needs 24 bits.

f_red, f_green, f_blue takes values in the range 0-255.

The 8-bit bitmap is one of the more extensive image representations in image technology.

4.2

Image binarization preprocessing

The fuzzy algorithm clusters the pixels into two classes [21]: the first class Ω1, white; the second class Ω2, black. The first step is to initialize each pixel XJ with an affiliation of μ_ij, (i=1, 2 and j=1, 2, 3……,n, n are the number of pixels), i.e., μ₁₁ represents the affiliation of pixel 1 to class Ω1 and μ₂₁ represents the affiliation of pixel 1 to class Ω2. The initialization must be satisfied: (5) $\sum_{i = 1}^{c} μ_{i j} = 1$

Here c denotes the number of classifications i.e. c = 2. The initialization process is as follows:

μ_ij = (gray level of pixel XJ)/255

j = 1,2⋯⋯n (n is the total number of pixels)

The next step is to calculate the clustering center Vi, one for each class.

(6)

V_{i} = \frac{\sum_{j = 1}^{n} {(μ_{i j})}^{n} X_{j}}{\sum_{j = 1}^{n} {(μ_{i j})}^{n}}

Each time a new clustering center is obtained, the affiliation of each pixel in the sample with respect to the new clustering center is updated.

(7)

μ_{i j} = \frac{1}{\sum_{j = 1}^{c} {(d_{i k} / d_{j k})}^{2 / m - 1}} \forall i, \forall k;

Here m is the weighting parameter and m≥1,d_ij=|X_j–V_i| is the distance of the pixel from the clustering center.

The newly generated μ_ij and V_i are computed continuously until the objective function: (8) $J_{m} = \sum_{i = 1}^{c} \sum_{j = 1}^{n} {(μ_{i j})}^{m} d^{2} (X_{j}, V_{i})$

Until there is no longer a significant change.

The next step is the inverse blurring of the affiliation degree of each pixel, i.e., in binarization, each pixel is to be classified as white or black. When the affiliation degree of a pixel satisfies μ_1j>μ_2j, the pixel can be considered as white, and vice versa, it is classified as black, thus completing the binarization of the image.

5

False propaganda image recognition based on template matching

Image matching technique refers to the process of searching for the same or similar image pattern from another image based on a known image pattern, i.e., a known image containing a pattern of images of false propaganda behavior, and searching for an image that also contains false propaganda behavior or involves false propaganda behavior from another image. The known image pattern is also known as the baseline template, and the image being searched is called the live image.The so-called template matching is a process in which the input image is overlaid with the template image, compared, and matched, and a determination is made as to whether the two are consistent. The process of detecting the position of template t(x,y) from input image f(x,y) based on template matching is as follows:

For the template t(x,y) capable of representing the object to be detected so that it coincides with the position of the point (u,v) in the image f(x,y), a degree of difference between t(x,y) and the pattern of the portion of the image with which it coincides can be calculated. This value indicates the reliability of the presence of the object at the point (u,v), and a smaller value indicates a higher likelihood that it is the object. In order to find the location of the object, it is sufficient to perform this operation for all points in the image and find the locations where the degree of difference is less than a certain threshold value. The degree of discrepancy can be utilized in the following equation: (9) $\iint_{s} | t (x, y) - f (x + u, y + v) | d x d y$

For the case of digital images, the degree of difference of the image from the template at the midpoint of the image can be calculated by the following formula: (10) $m (u, v) = \sum_{k = 1}^{n} \sum_{l = 1}^{m} | f (k + u - 1, l + v - 1) - t (k, l) |$

In the most basic template matching, each point has to be examined, and the amount of computation required to determine a frame is also very large with the minimum resolution of 320*288 that can be provided by a video capture card and a template size of 40*40. In order to save computational time, the sequential similarity detection algorithm is to be used. Instead of using a fixed threshold, the algorithm uses a monotonically growing sequence of net values, and if there is a pattern at m(u,v) that is consistent with the template, the value of m(u,v) is small. However, by taking a larger value at the inconsistency, the rate of increase of the sum of the absolute values of the differences between the pixels within the template and the pixels of the image will increase substantially, whereas at the matching point, this sum increases much more slowly. This allows the non-matching points to be eliminated in advance at a smaller threshold, thus greatly saving time in finding false advertising behaviors in the image.

Matching by correlation is computationally intensive because the template has to make correlation calculations at (N–M+1)² reference positions, which are useless except at the matching points. Therefore, the main points of the algorithm for sequential similarity detection are:

1) Define the absolute error value, see equation (11). (11) $ε (i, j, m_{k}, n_{k}) = | S^{i, j} (m_{k}, n_{k}) - S (i, j) - T (m, n) + T |$

where S(i, j) is the pixel average of the search subgraph under the template overlay and T is the pixel average of the template image, see Eq. (12) and Eq. (13).

(12)

S (i, j) = \frac{1}{M^{2}} \sum_{m = 1}^{M} \sum_{n = 1}^{M} S^{i, j} (m, n)

(13)

T = \frac{1}{M^{2}} \sum_{m = 1}^{M} \sum_{n = 1}^{M} T (m, n)

2) Take a fixed threshold value T_k

3) Randomly select a pixel point in sub-figure S^ij(m,n), calculate its error value ε with the corresponding point in T, and then add up the difference of this point with the difference of other points, and when the number of times of adding up exceeds R and the error exceeds T_k, then stop adding up and note down the number of times R, and define the SSDA detection surface as in Eq. (14). (14) $I (i, j) = {R | \min_{1 \leq R \leq m^{2}} [\sum_{k = 1}^{r} ε (i, j, m_{k}, n_{k}) \geq T_{k}]}$

4) The point with the largest value of I(i, j) is designated as the matching point, because it takes many accumulations at this point to make the total error ∑ε exceed T_k.

The error values of curves A and B quickly exceed the threshold, reflecting the fact that the template T is not at the matching point, while curve C grows more slowly with the number of accumulations and may be at the matching point.

6

Computer vision technology performance analysis

All the experimental data in this paper come from 220 publicity advertisements from ten online media collected by DirectShow-based video image acquisition technology and converted to 8000Hz, 16bit, mono wav data experiments every 256ms for a frame, 8ms for frame shift. Then use the image recognition algorithm proposed in this paper to recognize these advertisements. The algorithm’s experimental platform for desktop computers is configured as a dual-core memory data read locally. As mentioned earlier we need to use the training data to train the filter and noise parameters, in order to achieve adequate training, we intercepted these promotional advertisements broadcast by online media, each advertisement is truncated into 50 small segments, whether these segments contain false propaganda behavior to identify the judgment, the duration of these advertisements is distributed as shown in Table 1.

Table 1.

The length of the AD

Duration/S	Number	Proportion/%
0-5	10	5
5-10	44	22
10-15	45	22.5
15-30	41	20.5
30-45	54	27
45-60	4	2
More than60	2	1

Using the computer vision technique proposed in this paper to retrieve and recognize these advertisements respectively, the recall, precision, and F-value of the algorithm are shown in Fig. 3, Fig. 4, and Fig. 5, respectively, and it can be seen from the figures that the recall, precision, and F-value of the algorithm as a whole are all over 90%, and the average F-value of this algorithm reaches 95%, which fully illustrates the effectiveness and robustness of the algorithm. It shows that the method proposed in this paper is applicable to the supervision of online false propaganda behavior.

7

Computer vision technology application example analysis

The article uses the video monitoring system of a live selling platform as a case study object. The platform has 10 channels, with a total broadcasting time of more than 80h per day, and the video content covers the promotion and sale of many categories of commodities such as videos, household products, clothes, and so on. Due to the complicated content and large broadcast volume, it is difficult to detect the problem of false propaganda content in a comprehensive and timely manner using the traditional manual sampling method.The platform’s technical background is limited, and there is an urgent need for an efficient and intelligent video monitoring system. Statistics show that the platform generates an average of 200 GB of video data per day, with a peak bit rate of 50 Mb·s^-1, which puts high demands on false propaganda monitoring of videos. In addition, because the program’s content involves multiple product areas, the false propaganda is expressed in various forms.Therefore, the image recognition technology under computer vision is utilized to carry out intelligent video monitoring by combining the actual scenes of the platform. In terms of technical implementation, frame extraction and image acquisition of the video stream based on DirectShow and other technologies are adopted, on this basis, by setting up a multi-dimensional content feature library of false propaganda behaviors, and utilizing the template-matching image recognition algorithm based on the template matching, to achieve the accurate identification of false propaganda Based on this, the image recognition algorithm based on template matching is utilized to achieve accurate recognition of false propaganda content.

For the verification of advertisement broadcasting, the template matching-based image recognition algorithm is used to extract the key frame features of the advertisement material, conduct real-time comparison with the video stream, and combine with the timeline information to conduct segment-level localization and identification, which can realize the accurate verification of the advertisement content. The method has been practically applied in the platform, the scheme has achieved good results, and the recognition accuracy of different false propaganda behaviors on different platforms is shown in Figure 6. As can be seen from Figure 6, the recognition accuracy rate of the scheme in different types of false propaganda content is more than 90%, especially in the false propaganda behavior recognition accuracy rate of the user evaluation category is extremely good, the highest reached 100%.

Figure 7 shows the processing speed of this paper’s algorithm for false propaganda behaviors on live broadcast platforms, and the processing speed of this method for different false propaganda behaviors on different channels of live broadcast platforms can reach more than 200 f·s^-1, which basically meets the demand for real-time monitoring. The workload of manual verification is greatly reduced. The above scheme makes full use of the advantages of image recognition technology in intelligent video monitoring, and builds a set of efficient, accurate and comprehensive video monitoring system for this webcasting selling platform, which effectively improves the monitoring efficiency and content security, and has good value of popularization and application.

8

Online false propaganda civil and commercial law legal considerations

8.1

Clarify the Criteria for Determining the Subjects of Civil Liability for Online False Propaganda

The social and economic impact of the platform operator’s behavior in the network false propaganda is more obvious, therefore, the civil legal responsibility of the platform operator in the network false propaganda should be clearly stipulated. In the process of network false propaganda, there is a close relationship between the platform operator and the platform operators, and the neglect of the platform’s duty of prudence will directly lead to the untruthfulness of the basic information of the products or services, which indirectly affects the consumers’ basic right to know.

Therefore, it is necessary to refine the platform operator’s prudential obligations, further compaction of its civil liability norms, through the self-regulatory review mechanism, can further compaction of the main body of the platform operators honest management consciousness, clear its main responsibility, through the strengthening of the network platform operators self-regulatory management, correct network platform operators standardized management consciousness, to spontaneous, active review management to promote the Network operators honest operation.

8.2

Clarify the standards for determining liability for civil acts of online false propaganda

By the computer vision technology in the network false behavior supervision in the practical application of the current stage of network false propaganda behavior is found to be rampant, so in addition to the subject of the civil liability status of the clarity of the civil liability for the behavior should also be the key to the refinement of the civil liability for network false propaganda.

First of all, civil liability as a network false propaganda behavior legal responsibility of the main way, the subject of civil liability should be civil liability for behavior as the key to the determination of civil liability, the author believes that in the clarification of the responsibility of the network false propaganda civil behavior, in the network false propaganda behavior of the civil liability of the competing, for the application of the different laws should have the order of the boundaries of the provisions should be reflected for the protection of consumer rights and interests as well as other fair competition lawful competition, should also be the key to the refinement of the civil liability. Consumer rights and interests protection and the concept of fair competition are related to the protection of legal rights and interests. Under the premise of regulating unfair competition in network publicity, it should be based on the Protection of Consumer Rights and Interests Law, and the protection of the legitimate rights and interests of consumers should be clarified.

Second, through special legislation, standardize and standardize the standards for online publicity conduct of platform operators and operators within the platform, forming a sound and reasonable basis for adjudication of civil legal liability, and making more uniform provisions on the standards of conduct for online false publicity in relevant legislation such as the Anti-Unfair Competition Law, the Advertising Law, and the E-Commerce Law, for example, in adhering to the core definition standard of "misleading" behavior, using "whether there is exaggerated publicity" and "false publicity" The auxiliary standards defined as the standards of conduct require judges to consider the facts of the case in an all-round way in the actual judicial adjudication, so as to protect the legitimate rights and interests of consumers and other regulated operators, as well as a fair and effective market competition order.

Through the platform operator and the platform operator behavior obligation broadening, can further regulate the platform operator and the platform operator network propaganda behavior, this legislation mandatory obligation specification can also provide strong support for the definition of its behavior standard. Therefore, for the platform operator’s civil liability compaction can be through for reminding the obligation to regulate the start, this remind and regulate the obligation is established in the network platform operator to form a normalized, long-term business management evaluation based on the reminder, is a kind of bulletin-type reminder. On the one hand, this kind of reminder can make the platform operator and the platform operator to produce the standardized integrity management consciousness, the platform operator and the platform operator should be in the network publicity to fulfill certain real business and publicity obligations.

As a direct trading platform, the platform is a bridge connecting online goods or service providers and consumers, and this bridge role means that both platform operators and platform operators should take up the obligation of prudence to maintain a clear and healthy online publicity environment. The platform operator and the platform operator civil obligations and responsibilities, to further ensure that the online e-commerce environment is clear, and further optimize the platform business environment, the platform operator should play a certain reminder in the network publicity and the duty of prudence.

8.3

Improvement of the subjective fault rules of civil subjects of online false propaganda

For the time being, the judicial practice due to the network false propaganda itself subject standards and behavioral standards of the complexity of the subjective fault of the network false propaganda civil liability is more difficult to identify, so by computer vision technology identified false propaganda behavior is more and more, the offenders ignore the laws and regulations. The complexity of the identity of the platform means that for the general operator, it is in the process of entering into a civil legal contract directly with consumers as equal subjects, it should have a certain degree of knowledge of the consequences of the behavior as well as the legal responsibility, so the subjective fault is the necessary constituent elements of civil liability for online false propaganda. Therefore, it is necessary to refine the subjective fault determination by the platform operator.

Therefore, in order to further improve the subjective fault rules of the civil subject of network false propaganda, the scope of civil liability regulation of the subject of network false propaganda should be clarified to a certain extent. Reference to the U.S. “Lanham Act” on the subjective fault rules of the civil subject of network false propaganda, appropriately reduce the “intentional” standard, in the network false propaganda led to the legitimate rights and interests of consumers or other operators of the norms of the business rights and interests of the infringement of the platform as long as the platform operator’s network false propaganda behavior led to the legitimate rights and interests of consumers and other norms of the rights and interests of the operators by the strictest standards. The rights and interests of other standardized operators have been severely damaged, then it should be presumed that the operator has this intentional standard, and for general damage, this false propaganda behavior for the implementation of the perpetrator’s intentionality should be combined with the content of the general obligations of the perpetrator to the subjective fault for appropriate consideration.

Civil liability of subjective fault in the determination of “intentional” is relatively abstract, this paper that the platform operator and the platform operator formulated the format contract can become the platform operator and the platform operator subjective fault determination of the reference standard. In the determination of the subjective fault of the civil subject of the network false propaganda, the platform operator formulated the terms of the contract can be the platform operator and the platform operator important subjective fault auxiliary embodiment, in the subjective fault determination of its civil liability should be on the provision of services and behavior of the content of a comprehensive determination of the network platform operator, as a transaction rules should be borne by the developer No-fault civil liability.

Therefore, if the operator in the format contract of a large number of format terms to avoid their obligations, then it can be presumed that the platform operator and the platform operator has a certain subjective fault, for the platform operator or the platform operator civil liability to improve the causal relationship of certain aids.

9

Conclusion

This paper combines DirectShow-based video image acquisition technology, key frame extraction technology and image recognition technology in computer vision processing technology to carry out video content recognition based on image features, realizing the precise identification of false propaganda behaviors in the network and proposing appropriate legal considerations.

The image recognition technology based on template matching proposed in this paper retrieves and identifies different advertisements, and the algorithm’s recall, precision, and F-value values are above 90%, and the average F-value value of the template-matching-based image recognition algorithm in the identification of false propaganda behaviors in video content reaches 95%, which is an excellent performance. The effectiveness and robustness of the algorithm are fully demonstrated.

The recognition accuracy of template matching-based image recognition technology in different types of false propaganda behavior is 90% and above, and its average recognition accuracy in the false propaganda behavior of user evaluation category is as high as 99%. The speed of template matching-based image recognition technology in false propaganda behavior recognition reaches 200 f·s^-1 and above, which meets the speed requirements of real-time monitoring. It improves verification efficiency and can be popularized and applied in the monitoring of online false propaganda behaviors.

Sprache:: Englisch

Zeitrahmen der Veröffentlichung:: 1 Hefte pro Jahr
Fachgebiete der Zeitschrift:: Biologie, Biologie, andere, Mathematik, Angewandte Mathematik, Mathematik, Allgemeines, Physik, Physik, andere

Zeitschrift RSS Feed

Exploration of Computer Vision Technology and Legal Consideration of Online False Propaganda Behavior in the Perspective of Civil and Commercial Law

Mei Chang

Online veröffentlicht: 17. März 2025

Eingereicht: 14. Okt. 2024

Akzeptiert: 30. Jan. 2025

DOI: https://doi.org/10.2478/amns-2025-0277

SchlüsselwörterImage recognition, Key frame extraction, Fuzzy C-mean clustering algorithm, Image binarization algorithm, False propaganda behavior monitoring

© 2025 Mei Chang, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Schlüsselwörter
Image recognition, Key frame extraction, Fuzzy C-mean clustering algorithm, Image binarization algorithm, False propaganda behavior monitoring