Cross-platform infringement real-time monitoring and evidence obtaining method and system, product and medium

Abstract

The invention discloses a cross-platform infringement real-time monitoring and evidence obtaining method and system, a product and a medium, and relates to the field of mode comparison. The method comprises the following steps: receiving an original data chart image and a suspected infringement data chart image, and respectively extracting approximate numerical value information; separately dispersing the original data chart image and the suspected infringement data chart image into a preset number of equal-width intervals; obtaining an original vertical trend sequence and a suspected infringement vertical trend sequence according to the approximate value information; relative variations between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence are calculated respectively, and uniform relative trend feature signatures are obtained; comparing the original unified relative trend feature signature with the suspected infringement unified relative trend feature signature, and calculating to obtain a trend similarity score; and when the trend similarity score exceeds a preset infringement threshold, determining that the chart is an infringement chart. By implementing the technical scheme provided by the invention, the accuracy of real-time monitoring and evidence obtaining of cross-platform infringement is improved.

Description

Technical Field

[0001] This application relates to the field of pattern comparison, and in particular to a cross-platform real-time infringement monitoring and evidence collection method, system, product, and medium. Background Technology

[0002] Currently, with the development of big data analytics and business intelligence, the demand for data reports and visualization charts is growing across various industries. These charts not only present complex data insights intuitively, but also reflect a company's core competitiveness and professional analytical capabilities. They are widely used and crucial in fields such as finance, market research, and consulting.

[0003] In related technologies, deep learning-based image feature extraction and comparison methods are commonly used to monitor and identify copyright infringement in visual charts. Specifically, the original copyrighted chart and suspected infringing charts scraped from the internet are preprocessed, and then a pre-trained convolutional neural network (CNN) model is used to extract deep feature vectors from the images. These feature vectors are considered abstract representations of the chart content. Subsequently, the visual similarity between the two charts is determined by calculating the cosine similarity or Euclidean distance between the vectors, thereby identifying potential infringement.

[0004] However, when infringers redraw the original chart, adjust visual elements, or fine-tune the data to generate a new chart, although deep learning models can capture the visual patterns of the image, they mainly focus on the apparent features such as texture, color, and shape at the pixel level. When a chart is re-rendered, for example, by changing the color scheme, font style, chart type (such as changing a bar chart to a line chart), or even by slightly modifying the data points while keeping the core data trend unchanged, these visual changes can cause the extracted feature vectors to deviate significantly from the feature vectors of the original chart. This makes it difficult for the model to accurately determine the equivalence of the two charts at the deep semantic level of data trends and analytical conclusions, which may lead to missed detections and thus affect the accuracy of cross-platform real-time monitoring and evidence collection of infringement. Summary of the Invention

[0005] This application provides a method, system, product, and medium for cross-platform real-time infringement monitoring and evidence collection, which can improve the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0006] The first aspect of this application provides a cross-platform real-time infringement monitoring and evidence collection method, which includes: The system receives original data chart images and suspected infringing data chart images, and extracts approximate numerical information to obtain original approximate numerical information and suspected infringing approximate numerical information, respectively. The horizontal dimension of both images is discretized into a preset number of equal-width intervals, resulting in original equal-width intervals and suspected infringing equal-width intervals. Based on the original and suspected infringing approximate numerical information, the average position or average height in the vertical dimension is calculated within each original and suspected infringing equal-width interval, resulting in original and suspected infringing vertical trend sequences. The relative changes between adjacent elements in both sequences are calculated and encoded as discrete states, resulting in original and suspected infringing unified relative trend feature signatures. The original and suspected infringing unified relative trend feature signatures are compared to calculate a trend similarity score. When the trend similarity score exceeds a preset infringement threshold, the suspected infringing data chart image is determined to be an infringing chart.

[0007] In the above embodiments, by uniformly extracting different types of chart images into vertical trend sequences, and further calculating the relative changes of adjacent elements and encoding them into discrete states, a unified relative trend feature signature is generated, making charts from different sources and in different forms of expression highly comparable in terms of core data trends, thereby improving the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the relative change between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence are calculated and encoded as discrete states to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature, specifically including: Calculate the original relative change sequence and the suspected infringement relative change sequence between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence; sort the original relative change sequence and the suspected infringement relative change sequence internally, and encode each element as a discrete level state according to the sorting result to obtain the original discrete level state sequence and the suspected infringement discrete level state sequence; use the original discrete level state sequence and the suspected infringement discrete level state sequence as the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature, respectively.

[0009] In the above embodiments, by internally sorting and encoding these relative change sequences into discrete level states, the generated unified relative trend feature signature can effectively isolate the absolute numerical differences and overall scaling effects of the chart in the vertical dimension, making the feature signature highly robust to changes in the overall magnitude of the data. This allows for a more accurate capture of the inherent shape and dynamic pattern of the chart trend, improving the accuracy and stability of infringement identification when infringers circumvent the rules by adjusting the data magnitude or visual scale.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the original relative change sequence and the suspected infringement relative change sequence are internally sorted, and each element is encoded as a discrete level state according to the sorting result, resulting in the original discrete level state sequence and the suspected infringement discrete level state sequence, specifically including: For each element in the original relative change sequence and the suspected infringement relative change sequence, according to the preset quantization interval into which the value falls, it is encoded as the original quantization level state and the suspected infringement quantization level state; the original quantization level state sequence and the suspected infringement quantization level state sequence are respectively used as the original discrete level state sequence and the suspected infringement discrete level state sequence.

[0011] In the above embodiments, by encoding each element in the relative change sequence according to the preset quantization interval into which the value falls, the generated discrete level state sequence can directly reflect the absolute magnitude of the trend change, making the feature signature highly robust to the linear scaling of the overall data magnitude, and ensuring the semantic consistency of trend changes between different charts, thereby improving the accuracy and stability of infringement identification when infringers circumvent by adjusting the data magnitude or visual proportion.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the original unified relative trend feature signature is compared with the suspected infringing unified relative trend feature signature to calculate a trend similarity score, specifically including: Generate a reverse sequence of the suspected infringing unified relative trend feature signature; calculate the first trend similarity score between the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature using a dynamic time warping algorithm; calculate the second trend similarity score between the original unified relative trend feature signature and the reverse sequence using a dynamic time warping algorithm; select the maximum value between the first trend similarity score and the second trend similarity score as the trend similarity score.

[0013] In the above embodiments, by generating a reverse sequence of the suspected infringing signature and calculating its similarity to the original signature when comparing the signature of the chart trend features, and finally selecting the maximum value of the two as the trend similarity score, this approach can counter the behavior of infringers who try to evade detection by simply reversing the order of the chart's horizontal dimensions (such as reversing the timeline). Regardless of whether the trend is a forward copy or a reverse presentation, this solution can capture the inherent shape consistency, improving the accuracy and robustness of identifying such variant infringement behaviors and ensuring comprehensive detection of core trend patterns.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after calculating the average position or average height in the vertical dimension within each original equal-width interval and suspected infringement equal-width interval based on the original approximate numerical information and the suspected infringement approximate numerical information, to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence, the method further includes: The original vertical trend sequence and the suspected infringement vertical trend sequence are smoothed to obtain the smoothed original vertical trend sequence and the smoothed suspected infringement vertical trend sequence. The relative change between adjacent elements in the smoothed original vertical trend sequence and the smoothed suspected infringement vertical trend sequence are calculated and encoded as discrete states to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature.

[0015] In the above embodiments, by smoothing the original vertical trend sequence and the suspected infringing vertical trend sequence, the relative changes between adjacent elements in subsequent calculations better reflect the core trend of the chart, rather than random disturbances. Encoding these smoothed relative changes as discrete states results in a more stable and robust trend feature signature. This improves the accuracy of infringement identification of core data trends and analysis conclusions, even when infringers may evade detection by introducing subtle noise or making minor modifications to the data, thus ensuring the effectiveness of the comparison.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the original unified relative trend feature signature is compared with the suspected infringing unified relative trend feature signature to calculate a trend similarity score, specifically including: The original unified relative trend feature signature and the suspected infringing unified relative trend feature signature are divided into multiple original sub-signatures and multiple suspected infringing sub-signatures respectively; each original sub-signature is compared with each suspected infringing sub-signature to calculate multiple local trend similarity scores; the maximum value among the multiple local trend similarity scores is selected as the trend similarity score.

[0017] In the above embodiments, by dividing the complete trend feature signature into multiple sub-signatures with a preset minimum length, and performing a comprehensive pairwise comparison of all original sub-signatures with suspected infringing sub-signatures, the highest local similarity score is finally selected as the overall trend similarity. This identifies infringers who only copy part of the trend in the chart, combine multiple segments, or make significant changes to the overall trend but retain the core local pattern. This improves the detection accuracy and robustness when facing partial or fragmented infringement, ensuring that even hidden infringement can be effectively captured.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, based on the original approximate numerical information and the suspected infringement approximate numerical information, the average position or average height in the vertical dimension is calculated within each original equal-width interval and the suspected infringement equal-width interval to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence, specifically including: Multiple original data series and multiple suspected infringing data series are identified and separated from the original approximate numerical information and the suspected infringing approximate numerical information, respectively, to obtain sets of original data series and sets of suspected infringing data series. For each original data series and suspected infringing data series, the average position or average height in the vertical dimension is calculated within each original equal-width interval and suspected infringing equal-width interval, to obtain multiple original series vertical trend sequences and multiple suspected infringing series vertical trend sequences. The sets of multiple original series vertical trend sequences and the sets of multiple suspected infringing series vertical trend sequences are respectively used as the original vertical trend sequence and the suspected infringing vertical trend sequence.

[0019] In the above embodiments, multiple independent data series are identified and separated from the original and suspected infringing charts, and the vertical trend sequence of each data series is calculated separately. This can handle complex charts containing multiple curves or multiple bar chart groups, avoiding the problem of simply aggregating multiple series charts into a single trend and losing key information. This improves the accuracy and precision of infringement identification when the infringer only copies or modifies a specific data series in the chart, and ensures the comprehensive detection capability of structural infringement within complex charts.

[0020] Secondly, embodiments of this application provide a cross-platform real-time infringement monitoring and evidence collection system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the cross-platform real-time infringement monitoring and evidence collection system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a cross-platform real-time infringement monitoring and evidence collection system, cause the cross-platform real-time infringement monitoring and evidence collection system to execute the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a cross-platform real-time infringement monitoring and evidence collection system, cause the cross-platform real-time infringement monitoring and evidence collection system to perform the method described in the first aspect and any possible implementation thereof.

[0023] It is understood that the cross-platform real-time infringement monitoring and evidence collection system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the cross-platform real-time infringement monitoring and evidence collection method provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application extracts different types of chart images into a unified vertical trend sequence, and further calculates the relative change of adjacent elements and encodes them into discrete states to generate a unified relative trend feature signature. This makes charts from different sources and in different forms of expression highly comparable in terms of core data trends, thereby improving the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0025] 2. By internally sorting and encoding these relative change sequences into discrete level states, this application generates a unified relative trend feature signature that can effectively isolate the absolute numerical differences and overall scaling effects of the chart in the vertical dimension. This makes the feature signature highly robust to changes in the overall magnitude of the data, thereby more accurately capturing the inherent shape and dynamic pattern of the chart trend. This improves the accuracy and stability of infringement identification when infringers try to circumvent the rules by adjusting the data magnitude or visual scale.

[0026] 3. This application encodes each element in the relative change sequence according to the preset quantization interval into which the value falls. The resulting discrete level state sequence can directly reflect the absolute magnitude of trend change, making the feature signature highly robust to linear scaling of the overall data magnitude. It also ensures the semantic consistency of trend changes between different charts, improving the accuracy and stability of infringement identification when infringers circumvent the rules by adjusting the data magnitude or visual proportion. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a cross-platform real-time infringement monitoring and evidence collection method in an embodiment of this application; Figure 2 This is another flowchart illustrating the cross-platform infringement real-time monitoring and evidence collection method in this application embodiment; Figure 3 This is an exemplary hardware structure diagram of a cross-platform real-time infringement monitoring and evidence collection system in this application embodiment. Detailed Implementation

[0028] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0030] In related technologies, deep learning-based image feature extraction and comparison methods are commonly used to monitor and identify copyright infringement in visual charts. Specifically, the original copyrighted chart and suspected infringing charts scraped from the internet are preprocessed, and then a pre-trained convolutional neural network (CNN) model is used to extract deep feature vectors from the images. These feature vectors are considered abstract representations of the chart content, and the visual similarity between the two charts is then determined by calculating the cosine similarity or Euclidean distance between the vectors, thereby identifying potential infringement. However, when infringers redraw the original chart, adjust visual elements, or fine-tune data to generate new charts, although deep learning models can capture the visual patterns of the images, they primarily focus on pixel-level appearance features such as texture, color, and shape. When a chart is re-rendered, for example by changing the color scheme, font style, chart type (such as changing a bar chart to a line chart), or even by slightly modifying the data points while keeping the core data trend unchanged, these visual changes can cause the extracted feature vectors to deviate significantly from the feature vectors of the original chart. This makes it difficult for the model to accurately determine the equivalence of the two charts at the deep semantic level of data trends and analytical conclusions, which may lead to missed detections and thus affect the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0031] In this embodiment, different types of chart images are received and their approximate numerical information is extracted. Then, their horizontal dimensions are discretized and a vertical trend sequence is calculated. Finally, by calculating the relative changes of adjacent elements and encoding them as discrete states, a unified relative trend feature signature is generated. This method can isolate differences in surface visual attributes such as color, size, and axis range, as well as minor perturbations in absolute data values. This makes charts from different sources and with different presentation formats highly comparable in terms of core data trends, thereby improving the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0032] Figure 1 This is a flowchart illustrating the cross-platform real-time infringement monitoring and evidence collection method used in the embodiments of this application, including the following steps: S101. Receive the original data chart image and the suspected infringing data chart image, and extract the approximate numerical information to obtain the original approximate numerical information and the suspected infringing approximate numerical information respectively.

[0033] Among them, the original data chart image refers to the original and legal chart image used as the comparison benchmark; the suspected infringing data chart image refers to the chart image to be tested for infringement; the approximate numerical information refers to the set of numerical values ​​that can represent the data expressed by the chart image, which are identified and parsed from the chart image, and are (x, y) coordinate sequences on line charts, (x, height) value sequences on bar charts, or (x, y) coordinate sets on scatter plots; the (x, y) coordinate sequence is used to represent the horizontal and vertical positions of data points in line charts or scatter plots; the (x, height) value sequence is used to represent the horizontal position of each bar in a bar chart and its corresponding height value.

[0034] First, the input chart image is preprocessed, such as grayscale conversion, binarization, and noise reduction, to enhance the contrast of chart elements.

[0035] Subsequently, image analysis techniques are used to identify key visual elements in the charts, including axes, tick marks, grid lines, text labels, and actual data elements (such as data points, data lines, or bars). Optical character recognition (OCR) technology is then used to identify the numerical scales, units, and label text on the axes, and to establish a mapping relationship between image pixel coordinates and actual data coordinates.

[0036] Finally, based on the identified chart type (line chart, bar chart, or scatter plot), the corresponding approximate numerical information is extracted from the data elements: for line charts and scatter plots, the (x, y) coordinate sequence of the data points is extracted; for bar charts, the horizontal position (x) of each bar and its corresponding height value sequence are extracted.

[0037] S102. Discretize the horizontal dimension of the original data chart image and the suspected infringing data chart image into a preset number of equal-width intervals to obtain the original equal-width interval and the suspected infringing equal-width interval.

[0038] Among them, horizontal dimension discretization refers to dividing the continuous or discrete range of chart data on the horizontal axis (usually representing time, category, etc.) into sub-intervals of fixed width; preset quantity refers to the number of intervals predetermined based on experience, experimentation, or domain knowledge during system design or deployment. For example, this number can be determined through experimental verification or domain expert advice based on the balance between the requirements for chart trend granularity and computational efficiency; equal-width intervals refer to these sub-intervals having the same width in the horizontal dimension.

[0039] First, from the extracted approximate numerical information, determine the horizontal dimension range (i.e., the minimum and maximum values ​​of all x-coordinates) of the original chart and the suspected infringing chart.

[0040] Then, based on the preset number of intervals, the specific width of each equal-width interval is calculated, and the start and end boundaries of all equal-width intervals are determined. For example, if the horizontal range is 100 and the preset number is 1, then the width of each interval is 1. This process is applied independently to the original chart and the suspected infringement chart, thereby generating the original equal-width interval sequence and the suspected infringement equal-width interval sequence, respectively.

[0041] In some embodiments, the discretization of the horizontal dimension can be achieved in a variety of ways: Optionally, based on discretization with a fixed number of intervals, firstly, the maximum and minimum values ​​of all x-coordinates in the original approximate numerical information are obtained to determine the original horizontal dimension range; then, according to the preset fixed number of intervals N, the width of each equal-width interval is calculated as (maximum x - minimum x) / N; then, starting from the minimum x, the start and end boundaries of N equal-width intervals are determined sequentially to generate the original equal-width intervals; finally, the above steps are repeated for the suspected infringing approximate numerical information to generate the corresponding suspected infringing equal-width intervals.

[0042] It is understandable that other methods can be used to discretize the horizontal dimension, which are not limited here.

[0043] S103. Based on the original approximate numerical information and the suspected infringement approximate numerical information, calculate the average position or average height in the vertical dimension within each original equal-width interval and the suspected infringement equal-width interval to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence.

[0044] Among them, the average position or average height in the vertical dimension refers to the statistical average of the vertical coordinates (y-value or height value) of all data points falling into each equal-width interval, to represent the overall vertical trend of the data in that interval; the original vertical trend sequence refers to the sequence composed of the average vertical dimensions calculated from the original chart data in each equal-width interval; the suspected infringement vertical trend sequence refers to the sequence composed of the average vertical dimensions calculated from the suspected infringement chart data in each equal-width interval.

[0045] Iterate through each predefined equal-width interval and check which data points in the approximate numerical information extracted in S101 have horizontal coordinates that fall within the current interval. For all data points that fall within the interval, sum the vertical coordinates (y-values ​​for line and scatter plots, and height values ​​for bar charts) and count the number of data points.

[0046] Then, the accumulated sum is divided by the number of data points to obtain the average position or average height in the vertical dimension within that interval. If there are no data points in a certain interval, the average value can be set to a preset value, or it can be estimated from adjacent intervals using interpolation methods. This process is applied independently to the original approximate numerical information and the suspected infringement approximate numerical information, ultimately generating the original vertical trend sequence and the suspected infringement vertical trend sequence, respectively.

[0047] In some embodiments, the calculation of the average position or average height in the vertical dimension can be achieved in a variety of ways: Optionally, the calculation based on weighted average or interpolation is performed by first traversing each data point in the original approximate numerical information and assigning it to the corresponding original equal-width interval, accumulating the vertical coordinates and counting them; then, for intervals containing data points, the average vertical value is calculated; then, for intervals not containing data points, an interpolation method (such as linear interpolation or spline interpolation) is used to estimate the average vertical value based on the average vertical value of adjacent intervals containing data points, or a weighted average is used, where the weights can be based on the distance from the data point to the center of the interval to more accurately reflect the interval trend; finally, an original vertical trend sequence is formed, and the above steps are repeated for the suspected infringing approximate numerical information to obtain the suspected infringing vertical trend sequence.

[0048] It is understandable that other methods can be used to calculate the average position or average height in the vertical dimension, and no limitation is made here.

[0049] In some embodiments, when the original data chart contains multiple independent data series, the deep semantics of the chart can be captured by identifying and separating these data series and calculating the vertical trend sequence for each series independently, thereby avoiding the loss of key information due to simple aggregation and improving the accuracy of infringement identification for multi-data-series charts.

[0050] First, the original approximate numerical information and suspected infringing approximate numerical information extracted from S101 are analyzed in depth. For line charts, multiple independent lines are identified, each representing a data series; for bar charts, multiple groups of bars, either parallel or stacked, are identified, each group representing a data series; for scatter plots, data points are categorized into different data series based on attributes such as color, shape, or label. This process is accomplished by analyzing the connections between data points, visual attributes, or metadata information, thereby decomposing the original approximate numerical information into multiple independent sets of original data series and sets of suspected infringing data series.

[0051] Next, each data series in the separated sets of original data series and suspected infringement data series is traversed. For each data series, the vertical trend is calculated independently. Specifically, for each original data series, the distribution of data points within each original equal-width interval generated in S102 is checked. Similar to the original logic in S103, the vertical coordinates (y-values ​​or height values) of all data points falling within the current equal-width interval are collected, and their average position or average height is calculated. If there are no data points for a certain data series within a certain interval, processing can be performed according to a preset strategy (e.g., an interpolation method determined through experimentation or domain knowledge, such as linear interpolation or keeping the previous valid value) to ensure that each data series has a corresponding vertical trend value across all equal-width intervals. This process is applied independently to each original data series, resulting in multiple original series vertical trend sequences. Similarly, the above operation is repeated for each suspected infringement data series to generate multiple suspected infringement series vertical trend sequences. By calculating the vertical trend sequence independently for each data series, even if an infringer selectively copies, reorders, or partially modifies the data series in the original chart, these changes can be captured at the series level, rather than being masked by the overall aggregated trend.

[0052] Finally, these independently generated sets of multiple original series vertical trend sequences are passed to subsequent steps as a whole set of original vertical trend sequences; similarly, multiple sets of suspected infringing series vertical trend sequences are passed to subsequent steps as a whole set of suspected infringing vertical trend sequences.

[0053] The above technical steps decompose multi-data series charts into independent series trends, avoiding information loss caused by simple aggregation. This enables the effective identification of infringers' operations such as copying and reordering of single or partial data series, improving the accuracy of infringement identification of core data trends and analysis conclusions in multi-data series scenarios.

[0054] In some embodiments, when there is sparsity, noise, or minor fluctuations in the original or suspected infringing chart data, the vertical trend sequence can be smoothed to eliminate noise and highlight the core trend, thereby improving the robustness of subsequent feature signatures to data perturbations and thus improving the accuracy of infringement identification.

[0055] First, both the original vertical trend sequence and the suspected infringement vertical trend sequence are smoothed. Smoothing involves applying specific mathematical filters or algorithms to adjust each data point in the sequence, making the value closer to the average or weighted average of neighboring points. This filters out high-frequency noise and highlights low-frequency, more stable trends. For example, a moving average method can be used, replacing the value of each data point with the average of several data points before and after it; or a Gaussian filter can be used, smoothing the data through a weighted average, where the weights are determined by a Gaussian function, making the neighbors closer to the data point have a greater influence on the smoothing result. The chosen smoothing method and its parameters (e.g., the window size of the moving average, the standard deviation of the Gaussian filter) are determined through experimental verification or domain knowledge based on the data noise characteristics, the desired smoothing degree, and the need to preserve trend details. Through smoothing, a smoothed original vertical trend sequence and a smoothed suspected infringement vertical trend sequence are obtained that are more stable and better represent the core trend of the data.

[0056] Subsequently, on these smoothed vertical trend sequences, the relative changes between adjacent elements are calculated and encoded as discrete states to generate the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature. This process is the same as the original logic of S104, but the input has become a smoothed sequence. Since the noise in the input sequence has been effectively suppressed, the calculated relative changes will be more stable and reliable, thus enabling the generated discrete state encoding to more accurately capture the true trend of the chart.

[0057] The above technical steps, by introducing smoothing processing on the trend sequence, suppress noise and minor fluctuations in the data, making the subsequently generated trend feature signature more reflective of the core trend of the chart. This improves the accuracy of infringement identification of core data trends and analysis conclusions when infringers may evade detection by introducing minor noise or slightly smoothing the data.

[0058] S104. Calculate the relative change between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence respectively, and encode them as discrete states to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature.

[0059] Among them, relative change refers to the numerical difference between two adjacent elements in a vertical trend sequence, usually expressed as (current value - previous value), reflecting the magnitude of the rise or fall of the trend; encoding as discrete state means mapping continuous relative change values ​​to a finite set of predefined discrete symbols or categories according to their magnitude or direction, such as encoding as "rising", "falling", "stable", etc.; discrete state refers to a finite set of symbols or categories used to represent relative change; the original unified relative trend feature signature refers to a discrete state sequence encoded by the relative change of the original vertical trend sequence, describing the trend of the original chart; the suspected infringement unified relative trend feature signature refers to a discrete state sequence encoded by the relative change of the suspected infringement vertical trend sequence, describing the trend of the suspected infringement chart.

[0060] The process iterates through adjacent element pairs in both the original and suspected infringing vertical trend sequences. For each pair of adjacent elements (V_i, V_{i+1}), the difference between them (V_{i+1} - V_i) is calculated; this difference represents the relative change. These continuous relative change values ​​are then mapped to predefined discrete states. For example, a small positive threshold T and a negative threshold -T can be set (these thresholds can be preset based on experience, experimentation, or a percentage of the data range). If the relative change is greater than T, it is encoded as "rising"; if it is less than -T, it is encoded as "falling"; otherwise, it is encoded as "stationary." This process transforms each numerical trend sequence into a sequence of discrete symbols, i.e., a unified relative trend feature signature.

[0061] In some embodiments, the calculation of relative changes and discrete state encoding can be achieved in a variety of ways: Optionally, based on discrete encoding with a fixed threshold, the original vertical trend sequence is first traversed, and the difference between adjacent elements is calculated to obtain the original relative change sequence. Next, three preset thresholds are defined: a positive threshold T_up and a negative threshold T_down (for example, T_up can be set to 0.5% of the vertical range, and T_down to -0.5%). Then, for each element in the original relative change sequence, if the value is greater than T_up, it is encoded as "increasing"; if the value is less than T_down, it is encoded as "decreasing"; otherwise, it is encoded as "stable". Finally, the encoding results are combined to form the original unified relative trend feature signature, and the above steps are repeated for the suspected infringing vertical trend sequence to obtain the suspected infringing unified relative trend feature signature.

[0062] Optionally, based on discrete coding of relative levels, the original vertical trend sequence is first traversed, and the difference between adjacent elements is calculated to obtain the original relative change sequence. Then, the original relative change sequence is internally sorted, and each element is encoded as a discrete level state according to the sorting result (for example, all changes are divided into N levels, each level representing a relative size range, such as "significant increase", "slight increase", "stable", "slight decrease", "significant decrease"). Then, the coding results are combined to form the original unified relative trend feature signature. Finally, the above steps are repeated for the suspected infringement vertical trend sequence to obtain the suspected infringement unified relative trend feature signature.

[0063] It is understandable that other methods can be used to calculate relative changes and encode discrete states, and no limitations are made here.

[0064] S105. Compare the original unified relative trend feature signature with the suspected infringing unified relative trend feature signature, and calculate the trend similarity score.

[0065] Among them, comparison refers to quantifying the degree of similarity between two discrete state sequences through a specific algorithm; trend similarity score is a numerical value used to represent the degree of matching between two trend feature signatures in terms of shape and direction, and generally the higher the score, the more similar they are.

[0066] These two sequences, each consisting of discrete state symbols, are taken as input, and sequence alignment algorithms are applied to measure their degree of similarity. These algorithms typically assess similarity by calculating the "cost" required to transform one sequence into the other (e.g., the number of operations required to insert, delete, or replace a symbol) or by finding the optimal alignment path between the two sequences. Lower transformation costs or better alignment quality generally correspond to higher similarity scores. The chosen alignment algorithm (e.g., Dynamic Time Warping (DTW) or edit distance algorithms) is pre-defined based on anticipated sequence changes that infringement might cause (e.g., changes in sequence length, local distortions, or symbol substitutions).

[0067] Finally, a trend similarity score is output, which quantifies the degree of similarity between the trends of the two charts.

[0068] In some embodiments, the comparison of trend feature signatures and the calculation of similarity scores can be achieved in multiple ways: Optionally, based on the Dynamic Time Warping (DTW) algorithm, the comparison first uses the original unified relative trend signature and the suspected infringing unified relative trend signature as input sequences. Next, a cost matrix is ​​constructed, where each cell (i, j) represents the cumulative matching cost between the first i elements of the original signature and the first j elements of the suspected infringing signature. This cost is based on a preset distance between discrete state symbols (e.g., 0 for identical symbols and 1 for different symbols). Then, a dynamic programming algorithm is used to fill the cost matrix, calculating the optimal path from (0, 0) to (M, N) (where M and N are the lengths of the two signatures, respectively). The cumulative cost of this path is the DTW distance. Finally, the DTW distance is converted into a similarity score, for example, by (1 / (1+DTW distance)) or (maximum possible distance - DTW distance) / maximum possible distance, where the maximum possible distance is the maximum DTW distance estimated based on the sequence length.

[0069] It is understandable that other methods can be used to compare trend feature signatures and calculate similarity scores, which are not limited here.

[0070] In some embodiments, where infringers may reverse the data order of the chart in the horizontal dimension, such infringement can be identified by simultaneously comparing the forward and reverse sequences of the original trend signature and the suspected infringing trend signature, and selecting the best matching result, thereby improving the accuracy of infringement identification.

[0071] In some scenarios, infringers may change the timeline of the original chart from forward (e.g., from morning to night) to reverse (e.g., from night to morning), or reverse the order of the category axes, but the core data trends and relationships expressed by the chart remain unchanged. If only a single-direction comparison is performed, this reverse arrangement will cause the order of the trend feature signature sequence to be completely reversed, resulting in a significant decrease in the similarity score and causing false negatives.

[0072] First, a reverse sequence of the suspected infringement unified relative trend feature signature is generated. This means reversing the discrete state symbols in the suspected infringement trend signature according to their positions in the sequence, thus obtaining a new sequence that is opposite in direction to the original suspected infringement signature.

[0073] Subsequently, the Dynamic Time Warping (DTW) algorithm is used to calculate the first trend similarity score between the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature. The DTW algorithm is a sequence alignment method capable of handling inconsistent sequence lengths and local temporal distortions, calculating the distance or similarity between two sequences by finding the optimal alignment path. Here, the DTW algorithm can capture the direct similarity in trend shape between the original chart and the suspected infringing chart. The cost function used in the DTW algorithm to measure the difference between discrete state symbols (e.g., 0 cost for identical symbols, 1 cost for different symbols) is pre-defined based on the understanding of trend semantics and experimental verification.

[0074] Next, the Dynamic Time Warping (DTW) algorithm is used again to calculate the second trend similarity score between the original uniform relative trend signature and the reverse sequence of the suspected infringing uniform relative trend signature. This comparison is specifically used to detect whether the infringer has copied the trend of the original chart by reversing the horizontal dimension. By comparing with the reverse sequence, infringing charts that are identical to the original chart in trend shape but have the reverse horizontal order can be identified.

[0075] Finally, the maximum value between the first trend similarity score and the second trend similarity score is selected as the final trend similarity score. The advantage of this approach is that it captures the highest trend similarity regardless of whether the infringer directly copies the trend of the original chart or replicates it by reversing the horizontal dimension, thus ensuring effective detection of both forms of infringement.

[0076] The above technical steps, by introducing a comparison mechanism for the reverse sequence of trend signatures of suspected infringing charts, can effectively identify the behavior of infringers who evade detection by reversing the order of the horizontal dimensions of the chart. This improves the accuracy of identifying infringing behavior when infringers try to cover up their infringement by changing the order of data presentation (e.g., reversing the time axis or the category axis), but the core data trends and analysis conclusions remain unchanged.

[0077] S106. When the trend similarity score exceeds the preset infringement threshold, the suspected infringing data chart image is determined to be an infringing chart.

[0078] The infringement threshold is a pre-set value that serves as a critical point for determining whether two chart trends constitute infringement. This threshold is preset through a calibration process based on extensive experimental data, expert experience, or legal regulations, aiming to balance the false positive rate and the false negative rate.

[0079] The system receives the trend similarity score calculated by S105 and compares it with a pre-set infringement threshold. If the trend similarity score is higher than the infringement threshold (generally, a higher score indicates higher similarity), the suspected infringing data chart image is determined to have substantial similarity to the original chart in core data trends, thus identifying it as an infringing chart. Conversely, if the similarity score does not reach or falls below the infringement threshold, it is considered that no infringement has been detected. Setting the infringement threshold is crucial; it is typically determined through experimentation and optimization on a test dataset containing known infringing and non-infringing chart pairs to achieve the desired detection precision and recall.

[0080] In the above embodiments, by uniformly extracting different types of chart images into vertical trend sequences, and further calculating the relative changes of adjacent elements and encoding them into discrete states, a unified relative trend feature signature is generated, making charts from different sources and in different forms of expression highly comparable in terms of core data trends, thereby improving the accuracy of cross-platform real-time infringement monitoring and evidence collection.

[0081] In other embodiments of this application, when infringers only copy a portion of the trend in a chart or scale the data range, global trend comparison may fail to effectively identify such partial or distorted infringement. The cross-platform real-time infringement monitoring and evidence collection method provided in this application can capture fragmented trends through discrete level coding and local sub-signature comparison, thereby improving the robustness of infringement identification.

[0082] like Figure 2 The diagram shown is another flowchart illustrating the cross-platform real-time infringement monitoring and evidence collection method provided in this application embodiment, including the following steps: S201. Receive the original data chart image and the suspected infringing data chart image, and extract the approximate numerical information to obtain the original approximate numerical information and the suspected infringing approximate numerical information respectively.

[0083] S202. Discretize the horizontal dimension of the original data chart image and the suspected infringing data chart image into a preset number of equal-width intervals to obtain the original equal-width interval and the suspected infringing equal-width interval.

[0084] S203. Based on the original approximate numerical information and the suspected infringement approximate numerical information, calculate the average position or average height in the vertical dimension within each original equal-width interval and the suspected infringement equal-width interval to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence.

[0085] Steps S201-S203 and Figure 1 Steps S101-S103 in the illustrated embodiment are similar and can be found in the descriptions of steps S101-S103, which will not be repeated here.

[0086] S204. Calculate the original relative change sequence and the suspected relative change sequence between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence.

[0087] Among them, the original vertical trend sequence refers to the sequence composed of the average vertical dimensions calculated from the original chart data within each equal-width interval; the suspected infringement vertical trend sequence refers to the sequence composed of the average vertical dimensions calculated from the suspected infringement chart data within each equal-width interval; adjacent elements refer to two data points that are consecutive in the sequence; the original relative change sequence refers to the sequence obtained by calculating the difference between each pair of adjacent elements in the original vertical trend sequence, reflecting the local change direction and magnitude of the original chart trend; the suspected infringement relative change sequence refers to the sequence obtained by calculating the difference between each pair of adjacent elements in the suspected infringement vertical trend sequence, reflecting the local change direction and magnitude of the suspected infringement chart trend.

[0088] Traverse the original vertical trend sequence, starting from the second element, subtract the value of the previous element from the value of the current element, and record the resulting differences in sequence to form the original relative change sequence. Similarly, perform the same calculation on the suspected infringement vertical trend sequence to obtain the suspected infringement relative change sequence.

[0089] In some embodiments, when it is necessary to group and quantize the absolute magnitude of relative changes to achieve more semantically consistent trend encoding, the relative change sequence can also be encoded by a preset quantization interval, thereby generating a more stable discrete-level state sequence that is sensitive to the absolute magnitude of changes, so as to improve the accuracy and interpretability of infringement identification.

[0090] Specifically, unlike the relative ranking based on internal sorting in S205, this step focuses on grouping the absolute magnitude of relative changes according to a preset quantization interval, so that the generated discrete level states have consistent semantics across charts. That is, a "significant increase" state represents a similar magnitude of growth in any chart, regardless of the relative magnitude of other changes within that chart.

[0091] First, a series of preset quantization intervals are defined. These intervals are a set of non-overlapping numerical ranges that collectively cover the entire possible range of values ​​for the relative change and are used to group the magnitude of the relative change. For example, intervals such as "(-∞, -X]", "(-X, -Y]", "(-Y, Y)", "(Y, X]", and "(X, +∞)" can be defined, where X and Y are preset thresholds optimized based on domain knowledge, historical data analysis, or experiments. These thresholds are typically determined based on the practical significance of changes in data trends (e.g., in the financial field, a 2% increase might be considered "small growth," while a 10% increase might be considered "large growth"). Each element in the original relative change sequence is iterated over to determine which preset quantization interval the value falls into. Then, the element is encoded as the original quantization level state corresponding to that interval (e.g., represented by a different integer or symbol). For example, if a relative change is +5% and the preset interval (Y, X] corresponds to "moderate increase," then the element is encoded as a "moderate increase" state. This process is applied independently to the original relative change sequence and the suspected infringement relative change sequence, generating original quantization level state sequences and suspected infringement quantization level state sequences, respectively.

[0092] Subsequently, these generated original quantized level state sequences and suspected infringement quantized level state sequences are used as the original discrete level state sequences and suspected infringement discrete level state sequences, respectively. This step clarifies that the quantized encoded sequences will serve as the basis for subsequent trend feature signatures, emphasizing their core role in trend comparison. Through this encoding method based on a preset quantization interval, even if the overall data range or volatility of the original chart and the suspected infringement chart differs, as long as the absolute change amplitude of their core trend falls within the same quantization interval, they can still be encoded as the same discrete level state, thereby improving the robustness of the trend feature signature to these changes.

[0093] The above technical steps, by encoding the relative change amount using a preset quantization interval, enable the generated discrete-level state sequence to more directly reflect the absolute magnitude of trend changes, enhance the semantic consistency of trend feature signatures when comparing different charts, and improve the ability to identify infringers who circumvent detection by fine-tuning data magnitudes, thereby improving the accuracy and interpretability of infringement detection.

[0094] S205. Sort the original relative change sequence and the suspected infringement relative change sequence internally, and encode each element as a discrete level state according to the sorting result to obtain the original discrete level state sequence and the suspected infringement discrete level state sequence.

[0095] Internal sorting refers to arranging all elements in a sequence in ascending or descending order of their numerical values. Discrete rank states refer to mapping continuous numerical relative changes to finite, predefined symbols or categories based on their relative position or numerical range in the sorted sequence. For example, these can be represented as "significant decrease," "moderate decrease," or "stable," indicating the relative size rank of an element in the original relative change sequence. Relative size rank refers to the relative size ranking or range of an element in the entire relative change sequence, reflecting whether the change is larger or smaller than other changes.

[0096] The original relative change sequence and the suspected infringement relative change sequence are processed separately. For each sequence, all elements are first internally sorted to obtain an ordered list of relative changes.

[0097] Then, based on the preset number of discrete levels (e.g., 5 levels), this ordered list is divided into several intervals, each interval corresponding to a discrete level state. For example, intervals can be divided according to quantiles (e.g., 20th quantile, 40th quantile, etc.) or according to a fixed numerical range. The method and number of these discrete level states are preset based on the requirements for trend granularity, tolerance for noise, and experimental results.

[0098] Finally, the original relative change sequence is traversed, and each element is encoded into a corresponding discrete level state based on the interval in which its value falls, thus forming the original discrete level state sequence. The same process is repeated for the suspected infringement relative change sequence to obtain the suspected infringement discrete level state sequence.

[0099] S206. The original discrete-level state sequence and the suspected infringing discrete-level state sequence are respectively used as the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature.

[0100] Among them, the original discrete level state sequence refers to the symbol sequence obtained by calculating the relative change and discrete level encoding of the original chart data; the suspected infringement discrete level state sequence refers to the symbol sequence obtained by calculating the relative change and discrete level encoding of the suspected infringement chart data; the original unified relative trend feature signature refers to the standardized and symbolic feature sequence that is ultimately used to represent the core trend of the original chart; the suspected infringement unified relative trend feature signature refers to the standardized and symbolic feature sequence that is ultimately used to represent the core trend of the suspected infringement chart.

[0101] This step is performed after S205 completes the generation of the original discrete-level state sequence and the suspected infringement discrete-level state sequence. This is a defining step, aiming to explicitly use the output of S205 as the formal input for subsequent comparison stages. This symbol sequence is the fingerprint of the core trend of the chart, removing surface information such as chart appearance, specific values, colors, and fonts, retaining only the relative direction and relative magnitude of data changes.

[0102] S207. The original unified relative trend feature signature and the suspected infringing unified relative trend feature signature are respectively divided into multiple original sub-signatures and multiple suspected infringing sub-signatures.

[0103] Among them, the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature refer to the symbol sequence that represents the core trend of the chart established in S206; the original sub-signature refers to a continuous sub-sequence with a preset minimum length extracted from the original unified relative trend feature signature; the suspected infringing sub-signature refers to a continuous sub-sequence with a preset minimum length extracted from the suspected infringing unified relative trend feature signature; the preset minimum length refers to the shortest sub-signature length that is predetermined during system design or deployment based on experience, experiments or the definition of meaningful trend segments. For example, this length can be determined through experimental verification or domain expert advice based on the balance between the judgment of the importance of local trends and computational efficiency.

[0104] In real-world infringement scenarios, infringers may not copy the entire chart, but rather only one or more key trend segments, or they may stitch together segments from multiple original charts to create a new chart. Such partial infringement is difficult to detect if only the complete signature is compared. By dividing the complete signature into multiple sub-signatures, a finer-grained comparison can be performed, thereby identifying any similar trend segments that reach a preset minimum length.

[0105] Traverse the original unified relative trend feature signature, starting from the beginning of the sequence, and generate all possible continuous subsequences with a length not less than a preset minimum length. These subsequences are then used as the original sub-signatures. For example, if the minimum length is 3, a signature of length 5 [A, B, C, D, E] will generate sub-signatures [A, B, C], [B, C, D], [C, D, E], [A, B, C, D], [B, C, D, E], [A, B, C, D, E]. Similarly, perform the same partitioning operation on the suspected infringing unified relative trend feature signature, generating multiple suspected infringing sub-signatures.

[0106] S208. Compare each original sub-signature with each suspected infringing sub-signature to calculate multiple local trend similarity scores.

[0107] Among them, the original sub-signature and the suspected infringing sub-signature refer to the symbol sequences that represent local trends obtained from the division in S207; the comparison refers to quantifying the similarity between two sub-signatures through a specific algorithm; the multiple local trend similarity scores refer to a series of values ​​obtained by comparing all original sub-signatures with all suspected infringing sub-signatures pairwise, with each value representing the similarity between a pair of sub-signatures.

[0108] A nested loop is executed: the outer loop iterates through all original sub-signatures, and the inner loop iterates through all suspected infringing sub-signatures. For each pair of original and suspected infringing sub-signatures, a sequence alignment algorithm is applied to calculate their similarity. Commonly used sequence alignment algorithms include Dynamic Time Warping (DTW) or Levenshtein Distance (LED). These algorithms can handle cases where sub-signature lengths are inconsistent and calculate the "cost" or "distance" required to transform one sub-signature into another, which is then converted into a similarity score. For example, DTW measures the shape similarity of two sequences by finding the optimal alignment path, while LED calculates the minimum number of operations required to transform one string (in this case, a sequence of symbols) into another. The choice of alignment algorithm and its parameters (e.g., the cost of matching / mismatching symbols) are preset based on the definition of trend similarity and experimental results.

[0109] Ultimately, this step generates a set containing all pairwise comparison results, i.e., multiple local trend similarity scores.

[0110] S209. Select the maximum value from multiple local trend similarity scores as the trend similarity score.

[0111] Among them, multiple local trend similarity scores refer to the set of similarity values ​​calculated in S208 after pairwise comparison of all original sub-signatures and suspected infringing sub-signatures; the maximum value refers to the score with the largest value in the set of values; the trend similarity score refers to the single value used to quantify the overall trend similarity between the original chart and the suspected infringing chart.

[0112] Specifically, this step is performed after S208 calculates all local trend similarity scores, making the system highly sensitive to local infringement. Even if a suspected infringing chart differs significantly from the original chart overall, as long as one or more trend sub-signatures are highly similar to a segment of the original chart, the strongest evidence of similarity can be captured by selecting the maximum value. This is crucial for identifying infringements that modify most of the content but retain the core trend segments.

[0113] Iterate through all the local trend similarity scores generated by S208 and find the score with the highest value. This highest score is considered to represent the strongest trend similarity between the original chart and the suspected infringing chart, and is therefore selected as the final trend similarity score.

[0114] S210. When the trend similarity score exceeds the preset infringement threshold, the suspected infringing data chart image is determined to be an infringing chart.

[0115] Step S210 and Figure 1 Step S106 in the illustrated embodiment is similar and can be found in the description of step S106, which will not be repeated here.

[0116] In the above embodiments, the vertical trend of the chart is transformed into a relative change reflecting dynamic changes and encoded as discrete-level states to form a unified relative trend feature signature. By dividing these feature signatures into multiple sub-signatures with a preset minimum length, and comparing each original sub-signature with the suspected infringing sub-signature pairwise, the highest local similarity score is selected as the overall trend similarity. This can identify infringers who only copy part of the chart trend or combine multiple segments, improving the detection accuracy and robustness when facing partial or fragmented infringement.

[0117] The following describes an exemplary cross-platform infringement real-time monitoring and evidence collection system 300 provided in the embodiments of this application. Figure 3 This is an exemplary hardware structure diagram of the cross-platform infringement real-time monitoring and evidence collection system 300 provided in this application embodiment.

[0118] In some embodiments, the cross-platform real-time infringement monitoring and evidence collection system 300 is a computer device or includes a computer device. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements the methods in the embodiments of this application.

[0119] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0120] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0121] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0122] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0123] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A cross-platform real-time infringement monitoring and evidence collection method, characterized in that, include: The system receives original data chart images and suspected infringing data chart images, and extracts approximate numerical information to obtain original approximate numerical information and suspected infringing approximate numerical information respectively; the approximate numerical information is a (x, y) coordinate sequence on a line graph, a (x, height) value sequence on a bar chart, or a (x, y) coordinate set on a scatter plot. The horizontal dimensions of the original data chart image and the suspected infringing data chart image are discretized into a preset number of equal-width intervals to obtain the original equal-width interval and the suspected infringing equal-width interval; Based on the original approximate numerical information and the suspected infringement approximate numerical information, the average position or average height in the vertical dimension is calculated in each original equal-width interval and the suspected infringement equal-width interval to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence. The relative changes between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence are calculated and encoded as discrete states to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature. By comparing the original unified relative trend feature signature with the suspected infringing unified relative trend feature signature, a trend similarity score is calculated. When the trend similarity score exceeds a preset infringement threshold, the suspected infringing data chart image is determined to be an infringing chart.

2. The method according to claim 1, characterized in that, The process of calculating the relative changes between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence, and encoding them as discrete states, to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature, specifically includes: Calculate the original relative change sequence and the suspected relative change sequence between adjacent elements in the original vertical trend sequence and the suspected infringement vertical trend sequence; The original relative change sequence and the suspected infringement relative change sequence are internally sorted respectively, and each element is encoded as a discrete level state according to the sorting result to obtain the original discrete level state sequence and the suspected infringement discrete level state sequence; the discrete level state represents the relative size level of the element in the original relative change sequence. The original discrete-level state sequence and the suspected infringing discrete-level state sequence are respectively used as the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature.

3. The method according to claim 2, characterized in that, The process of internally sorting the original relative change sequence and the suspected infringement relative change sequence, and encoding each element as a discrete level state based on the sorting result, to obtain the original discrete level state sequence and the suspected infringement discrete level state sequence, specifically includes: Each element in the original relative change sequence and the suspected infringement relative change sequence is encoded as an original quantization level state and a suspected infringement quantization level state according to the preset quantization interval into which the value falls; the preset quantization interval is used to group the amplitude of the relative change sequence. The original quantized level state sequence and the suspected infringement quantized level state sequence are respectively used as the original discrete level state sequence and the suspected infringement discrete level state sequence.

4. The method according to claim 1, characterized in that, The comparison of the original unified relative trend feature signature with the suspected infringing unified relative trend feature signature, and the calculation of the trend similarity score, specifically includes: Generate the reverse sequence of the suspected infringing unified relative trend feature signature; The first trend similarity score between the original unified relative trend feature signature and the suspected infringing unified relative trend feature signature is calculated using a dynamic time warping algorithm. The second trend similarity score between the original unified relative trend feature signature and the reverse sequence is calculated using the dynamic time warping algorithm. The maximum value between the first trend similarity score and the second trend similarity score is selected as the trend similarity score.

5. The method according to claim 1, characterized in that, After calculating the average position or average height in the vertical dimension within each of the original equal-width intervals and the suspected infringement equal-width intervals based on the original approximate numerical information and the suspected infringement approximate numerical information, to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence, the method further includes: The original vertical trend sequence and the suspected infringement vertical trend sequence are smoothed respectively to obtain a smoothed original vertical trend sequence and a smoothed suspected infringement vertical trend sequence. The relative changes between adjacent elements in the smoothed original vertical trend sequence and the smoothed suspected infringement vertical trend sequence are calculated and encoded as discrete states to obtain the original unified relative trend feature signature and the suspected infringement unified relative trend feature signature.

6. The method according to claim 1, characterized in that, The comparison of the original unified relative trend feature signature with the suspected infringing unified relative trend feature signature, and the calculation of the trend similarity score, specifically includes: The original unified relative trend feature signature and the suspected infringing unified relative trend feature signature are respectively divided into multiple original sub-signatures and multiple suspected infringing sub-signatures; each of the original sub-signatures and the suspected infringing sub-signatures has a preset minimum length; Each original sub-signature is compared with each suspected infringing sub-signature, and multiple local trend similarity scores are calculated. The maximum value among the multiple local trend similarity scores is selected as the trend similarity score.

7. The method according to claim 1, characterized in that, The step of calculating the average position or average height in the vertical dimension within each of the original equal-width intervals and the suspected infringement equal-width intervals, based on the original approximate numerical information and the suspected infringement approximate numerical information, to obtain the original vertical trend sequence and the suspected infringement vertical trend sequence, specifically includes: Multiple original data series and multiple suspected infringing data series are identified and separated from the original approximate numerical information and the suspected infringing approximate numerical information, respectively, to obtain a set of original data series and a set of suspected infringing data series. For each of the original data series and the suspected infringement data series, calculate the average position or average height in the vertical dimension within each of the original equal-width interval and the suspected infringement equal-width interval to obtain multiple original series vertical trend sequences and multiple suspected infringement series vertical trend sequences. The original series vertical trend sequence set and the suspected infringement series vertical trend sequence set are respectively used as the original vertical trend sequence and the suspected infringement vertical trend sequence.

8. A cross-platform real-time infringement monitoring and evidence collection system, characterized in that, The cross-platform infringement real-time monitoring and evidence collection system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the cross-platform infringement real-time monitoring and evidence collection system to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on the cross-platform real-time infringement monitoring and evidence collection system, the cross-platform real-time infringement monitoring and evidence collection system performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is run on the cross-platform real-time infringement monitoring and evidence collection system, the cross-platform real-time infringement monitoring and evidence collection system performs the method as described in any one of claims 1-7.

Images