Trademark infringement risk analysis method based on big data

Abstract

The invention discloses a trademark infringement risk analysis method based on big data, and relates to the technical field of trademark infringement, and the method comprises the steps: widening a data collection channel according to a to-be-evaluated trademark, and introducing a dynamic price elastic model; performing image data preprocessing on the target trademark identifier, and constructing an infringement risk assessment model; constructing a consumption behavior prediction model, and constructing a consumption cognitive map; and a real-time processing architecture is constructed, the early warning information is customized in a personalized manner, and targeted contents are pushed. According to the invention, based on a service trademark infringement evaluation model and a sales infringement commodity evaluation model of machine learning, the characteristics of a target trademark are tightly deduced, the data value is deeply mined, the potential infringement risk is pre-judged by using an ultra-precise algorithm system and an efficient processing flow and constructing a consumption cognition map, the accurate control of the trademark infringement risk is realized, and the accuracy of the trademark infringement risk is improved. The conditions of service trademark infringement and sales infringement commodities are monitored through higher timeliness, and a firm defense line is built for trademark protection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of trademark infringement, in particular to a trademark infringement risk analysis method based on big data. BACKGROUND

[0002] Trademark infringement refers to the use of a trademark that is identical or similar to a registered trademark on the same or similar goods (services) without the permission of the trademark owner, or other interference, obstruction of the use of the registered trademark by the trademark owner, and damage to the legitimate rights and interests of the trademark owner.

[0003] Among them, the typical forms of trademark infringement include but are not limited to product trademark infringement, service trademark infringement and sales of infringing goods. The existing trademark infringement risk analysis is basically a big data comparison analysis of the images and trademark texts of trademarks. For example, a trademark infringement analysis method based on big data disclosed in Chinese patent publication No. CN118628769B includes obtaining the trademark content of a target trademark; determining the position relationship between the trademark image and the trademark text corresponding to the target trademark according to the trademark content, and the font corresponding to the trademark text of the target trademark; determining a plurality of comparison trademarks corresponding to the trademark image; extracting the RGB channel of the target trademark and each comparison trademark to determine the RGB reference value of the target trademark and each comparison trademark; comparing the target trademark and each comparison trademark to determine the graphic similarity, combining the RGB reference value of each comparison trademark, the graphic similarity to determine the corresponding overall matching parameter, and determining the infringement analysis result according to the overall matching parameter. The above method can identify the infringement of the images and texts of trademarks by comparing a large amount of trademark data, but it cannot analyze and evaluate the risk of service trademark infringement and sales of infringing goods, affecting the comprehensiveness of trademark infringement risk analysis. SUMMARY

[0004] The present application aims to provide a trademark infringement risk analysis method based on big data to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solution: a trademark infringement risk analysis method based on big data, comprising: According to the trademarks to be evaluated, including registered and planned registered trademarks, the service categories and product groups involved in the trademarks are determined; According to the service trademark data collection channel, the data docking with the trademark bureau database is optimized, the linkage with the market supervision department is strengthened, the online and offline integrated sampling inspection actions are participated in, the data is shared in real time, the online sales of infringing goods data is monitored, and the dynamic price elasticity model is introduced; Image data preprocessing is performed on the target trademark logo, an infringement risk assessment model is constructed, similarity calculation is performed on the sales of infringing goods, and the calculated results are used for risk infringement judgment; A component consumption behavior prediction model is constructed, search data on e-commerce platforms and social media public opinion are integrated, a consumer cognition graph is constructed, and based on Internet of Things and blockchain information, the production chain of goods is tracked; A real-time processing architecture based on edge computing and cloud computing collaboration is constructed, real-time monitoring is performed, early warning information is customized, and targeted content is pushed.

[0006] Preferably, the data collection channel includes, but is not limited to, e-commerce, social media, enterprise website, online service platform, industry forum website, and small but accurate data sources of professional service review website. The data obtained in the online and offline integrated sampling action includes, but is not limited to, physical information, store experience details, and purchase source, and analysis data of the sales network topology structure is added to trace the circulation path of infringing goods in the multi-level sales chain and locate the key transmission node. For online monitoring, a sales data prediction model is used to estimate sales volume, and the sales prediction formula used is: , wherein represents the prediction of the data value at time , represents the sum of data values at n time points from time t, and n is the length of the selected time window. Compared with the actual sales volume , the dynamic threshold value of k is initially set to 5, and when and , it is marked as a suspicious infringing good.

[0007] Preferably, in the image data preprocessing, the trademark text data of the target trademark is converted into a vector form, service trademark text feature extraction, service trademark image feature extraction, and sales infringing goods feature extraction are performed, the integrated service trademark text feature vector, image feature vector, and corresponding registered trademark information are used as training samples, and an infringement or non-infringement label is marked. The service trademark text feature extraction uses TF-IDF technology to convert trademark names and product names into vectors, and the formula used is:

[0008] , wherein

[0009] ; The service trademark image feature extraction uses CNN technology to extract the features of trademark images and product images, and the convolution operation used is: .

[0010] Preferably, the sales of infringing goods feature extraction is refined for the price of the sales goods, and a price dispersion coefficient is introduced , and the calculation formula is: , wherein is the price marked by different stores of the same brand, L is the number of stores, is the average marked price, and the price dispersion coefficient is used to measure the rationality of price fluctuations.

[0011] Preferably, the sales of infringing goods feature extraction also evaluates the sales of goods, and the sales feature adds the median value of sales in the last 14 days , and the skewness coefficient of sales in the last 30 days , wherein the skewness coefficient is calculated by the ratio of the third central moment to the cube of the standard deviation, and the sales data is , and the mean is:

[0012] The second central moment is:

[0013] The third central moment is:

[0014] The skewness coefficient is calculated as: , wherein the skewness coefficient reflects the symmetry of the sales distribution, and the existing indicators are combined to comprehensively evaluate the sales anomaly; The brand keyword feature extraction integrates semantic role labeling technology to identify keywords in the text, including but not limited to subject, object, and modifier roles, and construct a feature vector by weighting according to the importance of the roles, and strengthen the infringement direction.

[0015] Preferably, the service trademark construction infringing risk evaluation model vector machine uses an adaptive parameter adjustment method based on a radial basis function, dynamically finds the optimal kernel function parameters and penalty coefficients through cross-validation and gradient descent algorithm, so that the classification accuracy and generalization ability of the model on the training set reach the best balance, and the model evaluation uses a comprehensive index system, in addition to accuracy, recall rate, F1 value, introduces the Matthew correlation coefficient, evaluates the model performance from multiple dimensions, and is used for the model to discriminate infringement.

[0016] Preferably, the accuracy is: ; The recall rate is: ; , wherein (TP) is true positive, (TN) is true negative, (FP) is false positive, and (FN) is false negative; The Matthew correlation coefficient is: .

[0017] Preferably, the text similarity calculation of the sales of infringing goods calculates the trademark name and the product name by cosine similarity calculation, and the cosine similarity calculation is: , wherein is the dot product of vectors A and B, is the modulus of vector A.

[0018] Preferably, the tracking of the commodity production link comprises the following specific steps: Raw material procurement link: record the supplier information through the RFID tag; Production link: bind the unique ID of the trademark printing equipment with the blockchain evidence; Logistics link: the GPS data is linked with the smart contract, and if the transportation route deviates from the preset track, the risk check is automatically triggered.

[0019] Preferably, the real-time processing architecture comprises and is not limited to the data source end of the edge node of the e-commerce platform server, uses a lightweight model for preliminary screening, quickly filters low-risk data, uploads suspected infringing data to the cloud, combines a heavy model and large-scale data storage for deep analysis, shortens the monitoring time delay; the early warning information is personalized, estimates the impact of the infringing behavior, brand reputation and market share of the trademark holder, builds a brand value evaluation model, sets the current value of the brand , the value loss rate caused by infringement is r, and the estimated loss value is: .

[0020] Technical effects and advantages of the present application: (1) The service trademark infringement evaluation model and the sales of infringing goods evaluation model based on machine learning of the present application closely follow the characteristics of the target trademark, deeply mine the data value, and predict potential infringement risks by using an ultra-precise algorithm system, an efficient processing flow and a constructed consumer cognition map, so as to realize accurate control of trademark infringement risks, monitor the service trademark infringement and the sales of infringing goods with stronger timeliness, and build a solid defense line for trademark protection; The present application can trace the circulation path of infringing goods in the multi-level sales chain, locate the key propagation nodes, estimate the sales of target brand goods, extract the features of the service trademark infringement and the sales of infringing goods, build a risk evaluation model, accurately determine the risk of abnormal data and infringing goods, and monitor and warn the trademark infringement risks in real time. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The present application is a method step schematic diagram. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0023] The present application provides a trademark infringement risk analysis method based on big data as shown in Figure 1 The present application provides a trademark infringement risk analysis method based on big data as shown in According to the trademarks to be evaluated, including registered and planned registered trademarks, the service categories and product groups involved in the trademarks are determined, so as to facilitate the big data trademark infringement risk analysis according to the target trademark.

[0024] According to the service trademark, the data collection channel is widened, and the data collection channel includes but is not limited to e-commerce, social media, corporate website, online service platform, industry forum website, and small but accurate data sources of professional service review website. For example, in the industry forum, practitioners or users often discuss various service details, which may reveal clues of unauthorized use of trademarks; the business introduction and user feedback on the professional service review website are also important carriers of service trademark use information; even for example, in a high-end design service forum, designers share project experience and mention the software tools used, if the service trademark is unauthorized, it can be captured.

[0025] Optimize the data docking with the trademark office database, not only obtain the basic registration information, but also introduce the historical change records, dispute details and other deep data of the trademark, which can reflect the growth context and potential controversy of the trademark, and provide more stereoscopic reference for infringement judgment. Strengthen the linkage with the market supervision department, participate in the online and offline integrated sampling inspection action, share data in real time, the data obtained in the online and offline integrated sampling inspection action include but are not limited to physical information, store experience details, source of goods, and analysis data of sales network topology, trace the flow path of infringing goods in the multi-level sales chain, and locate the key transmission nodes; And monitor the data of infringing goods sold online, introduce a dynamic price elasticity model, and adjust the price range of the target brand goods in real time according to market supply and demand changes, promotion cycle, seasonal factors, etc. Among them, for online monitoring, the sales data prediction model is used to estimate the sales volume, and the sales prediction formula used is: Among them, represents the prediction of the data value at time ​The sum of data values from n time points back from time t, n is the length of the selected time window, compared with the actual sales , the dynamic threshold of k is initially set to 5, when and , it is marked as a suspected infringing commodity, combined with sales data, a time series analysis prediction model is used to estimate the normal sales trend line . The time series of sales data is , here the length of the time window n = 6 is selected, taking a fast-moving brand as an example, its sales fluctuate greatly, a small window is suitable, when predicting the next time sales using a simple moving average method, assuming the past 6 days sales of a certain commodity of the brand are (100, 120, 90, 110, 130, 140) pieces, then pieces. Compared with the actual sales, when , here the threshold k = 5 is initially set, and k can be adjusted according to the characteristics of the brand, and it is marked as a suspected commodity.

[0026] Image data preprocessing is performed on the target trademark logo, and the trademark text data of the target trademark is converted into vector form, and service trademark text feature extraction, service trademark image feature extraction and sales of infringing commodities feature extraction are performed. Service trademark text feature extraction uses TF-IDF technology to convert trademark names and product names into vectors, and the formula used is: , wherein , In the text classification task, TF-IDF as a feature extraction method can effectively represent the text content and help identify important keywords, improve the classification performance of the model, and the calculation is simple, the effect is obvious, especially in information retrieval and text mining, and has strong noise reduction ability, which can effectively reduce the weight of high-frequency common words and improve the weight of rare but important words; Service trademark image feature extraction uses CNN technology to extract features of trademark images and product images, and the convolution operation used is: Image feature extraction extracts local features of images through convolution operation, and reduces the dimension and calculation amount of feature maps through pooling operation, which can calculate the feature map area of the trademark image.

[0027] Sales of infringing commodities feature extraction refines the price of the sales commodity, introduces the price dispersion coefficient , the calculation formula is: , wherein is the price of the same brand in different stores, and L is the number of stores, For the average price, the price dispersion coefficient is used to measure the rationality of price fluctuations. Assuming that a certain brand of sports shoes is priced at (200, 180, 220, 190, 210) yuan in 5 different stores, the average price = 200 yuan, then the dispersion coefficient = 0.079; The feature extraction of sales of infringing goods also evaluates the sales of the goods. The sales feature adds the median value of the sales in the past 14 days , and the skewness coefficient of sales in the past 30 days The skewness coefficient is calculated by the ratio of the third central moment to the cube of the standard deviation. Let the sales data be The mean is : The second central moment is

[0028] The third central moment is

[0029] The skewness coefficient is calculated as The skewness coefficient reflects the symmetry of the sales distribution and combines with the existing indicators to comprehensively evaluate the sales anomaly. For example, the sales of a certain product in the past 30 days are (10, 12, 15, 13, 11, 14, 16, 18, 20, 22, 25, 28, 30, 32, 35, 38, 40, 42, 45, 48, 50, 52, 55, 58, 60, 62, 65, 68, 70, 72). The mean is = 35.7, the second central moment is = 250.9, and the third central moment is = 1720.2, then = 1.09. According to the coefficient calculation, the difference of the product determines the sales anomaly of the product. The brand keyword feature extraction integrates semantic role labeling technology to identify keywords in the text, including but not limited to subject, object, and modifier roles. The feature vector is constructed by weighting the roles according to their importance, and the infringement direction is strengthened. For example, if the keyword is "Nike sports shoes", "Nike" is the subject with a weight of (0.6), and "sports shoes" is the object with a weight of (0.4). The feature vector is constructed as ([0.6 x vector (Nike), 0.4 x vector (sports shoes)]).

[0030] The service trademark infringement risk assessment model is constructed based on machine learning. The integrated service trademark text feature vector, image feature vector and corresponding registered trademark information are used as training samples, and the infringement or non-infringement label is marked. The vector machine of the service trademark infringement risk assessment model uses an adaptive parameter adjustment method based on a radial basis function, and through cross-validation and gradient descent algorithm, the optimal kernel function parameter and penalty coefficient are dynamically found, so that the classification accuracy and generalization ability of the model on the training set reach the best balance. The model evaluation adopts a comprehensive index system. In addition to accuracy, recall and F1 value, the Matthews correlation coefficient is introduced to evaluate the model performance from multiple dimensions, which is used for the discrimination of the model to infringement behavior.

[0031] The model evaluation accuracy calculation formula is: ; The recall rate is: ; , wherein (TP) is true positive, (TN) is true negative, (FP) is false positive, and (FN) is false negative. The Matthews correlation coefficient is: The model evaluation introduces the receiver operating characteristic curve (ROC), the area under the curve (AUC), the average precision mean (mAP) and other indicators to evaluate the discrimination ability of the model to infringement samples from different angles. When calculating the area under the curve, the (TPR) and (FPR) under different thresholds are plotted and integrated to obtain the AUC value. When calculating mAP, the prediction results are sorted by confidence first, and the precision (P) and recall rate of each class are calculated in turn. The (P-R) curve is plotted, and the average of the area under the curve is integrated to obtain the mAP value.

[0032] Text similarity calculation and image similarity calculation are performed on the rule-based sales infringement goods, and the calculation results are used for risk infringement judgment. The rule system is refined. In addition to price, sales volume and brand keywords, factors such as shop credit score , product praise rate , after-sales complaint rate , etc. are included. For example, rule 1: If , , the cosine similarity of brand keywords is greater than 0.8, >0.5, <60 (full score 100), <70%, it is determined as a high-risk infringement good. Rule 2: If >15%, the brand keyword matching degree comprehensive score M is greater than 60, >0.8, combined with suspicious degree of purchase channel score >0.5, determined as moderate risk infringement goods, and so on, to build multi-dimensional fine risk stratification.

[0033] The text similarity calculation of the sales of infringing goods calculates the trademark name and the product name by cosine similarity, and the cosine similarity calculation is: , wherein is the dot product of vectors A and B, is the modulus of vector A, and the brand keyword matching degree comprehensive score (M) can be obtained by weighted summation according to the keyword frequency, position weight, semantic correlation and other factors.

[0034] The component consumption behavior prediction model integrates e-commerce platform search data and social media public opinion to build a consumption cognition map. For example, when the search volume of a certain brand on the APP increases sharply, the system automatically analyzes the competitor names mentioned in the comment area. If keywords such as "shanzhai" and "high imitation" are found, combined with e-commerce platform sales data, potential infringement risks can be predicted. The essence of trademark infringement is that shanzhai products use the market heat of genuine products to make profits, and consumer behavior data can expose this "heat following" trend in advance. This model integrates cross-platform data from the user end to build a correlation chain of "brand cognition-demand change-infringement risk", solving the lag problem of traditional monitoring "only focusing on trademarks, not looking at the market".

[0035] Based on the Internet of Things and blockchain information, the production chain of goods is tracked, which includes the following specific steps: Raw material procurement link: record supplier information through RFID tags; Production link: bind the unique ID of the trademark printing equipment with the blockchain storage; Logistics link: GPS data and smart contract linkage, if the transportation route deviates from the preset track, automatically trigger risk check.

[0036] Use a graph database (such as Neo4j) to build a commodity traceability graph, and associate raw material ID, production equipment ID, logistics order number, and retail terminal ID into a data chain. When an infringing good is found, it is traced back through the RFID tag or batch number on the good, forming an infringing data chain of "infringing goods → logistics order number → warehouse → production equipment ID → raw material supplier → lock infringing factory".

[0037] A real-time processing architecture based on edge computing and cloud computing collaboration is constructed, the real-time processing architecture is used for real-time monitoring, a lightweight model is used for preliminary screening at a data source end including but not limited to an edge node of an e-commerce platform server, low-risk data is quickly filtered, suspected infringement data is uploaded to a cloud end, deep analysis is performed in combination with a heavy model and large-scale data storage, monitoring time delay is shortened, early warning information is customized, and targeted content is pushed; The early warning information is customized, a trademark holder is highlighted, infringement behavior, brand reputation, market share influence estimation is provided, a brand value evaluation model is constructed, a current brand value is set , a value loss rate caused by infringement is r, and an estimated loss value is: ; for a supervision department, detailed law enforcement clues such as infringement commodity warehouse logistics information and social network association of personnel involved in a case are provided, precision strikes are assisted, trademark rights and interests are effectively maintained, and precision control over trademark infringement risks is realized.

[0038] Finally, it should be noted that: the above only describes the preferred embodiments of the present application and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for trademark infringement risk analysis based on big data, characterized in that, include: Based on the trademarks to be evaluated, including registered and planned trademarks, clarify the service categories and goods groups covered by the trademarks; Expand data collection channels based on service marks, optimize data connection with the Trademark Office database, strengthen collaboration with market supervision departments, participate in their online and offline integrated sampling inspections, share data in real time, monitor online sales of infringing goods, and introduce a dynamic price elasticity model. Image data preprocessing is performed on the target trademark logo to construct an infringement risk assessment model, similarity calculation is performed on the infringing goods being sold, and the calculation results are used to determine the risk of infringement. Build a consumer behavior prediction model, integrate e-commerce platform search data and social media sentiment, construct a consumer cognition graph, and track the product production chain based on IoT and blockchain information; Build a real-time processing architecture based on the collaboration of edge computing and cloud computing, perform real-time monitoring, and personalize early warning information to push targeted content.

2. The trademark infringement risk analysis method based on big data according to claim 1, characterized in that, The data collection channels include, but are not limited to, niche but precise data sources such as e-commerce, social media, corporate websites, online service platforms, industry forums, and professional service review websites. The data obtained in the integrated online and offline sampling inspection includes, but is not limited to, physical product information, store experience details, and sourcing information. It also includes analytical data on the sales network topology, tracing the flow path of infringing goods in multi-level sales chains, and locating key dissemination nodes. For online monitoring, a sales data prediction model is used to estimate sales volume, and the sales prediction formula used is: ,in Indicates time Prediction of time data values This represents the sum of data values ​​at n time points counting backwards from time t, where n is the length of the selected time window, compared to actual sales. Initially, the dynamic threshold of k is set to 5. When the following occurs... and These are marked as potentially infringing products.

3. The trademark infringement risk analysis method based on big data according to claim 1, characterized in that, In image data preprocessing, the target trademark's text data is converted into vector form for service mark text feature extraction, service mark image feature extraction, and infringing goods feature extraction. The integrated service mark text feature vector, image feature vector, and corresponding registered trademark information are used as training samples, labeled with infringing or non-infringing tags. The service mark text feature extraction uses TF-IDF technology to convert trademark names and product names into vectors, with the following formula: in ； The service mark image feature extraction employs CNN technology to extract features from both the service mark image and the product image. The convolution operation used is as follows: 。 4. The trademark infringement risk analysis method based on big data according to claim 3, characterized in that, The feature extraction of infringing goods involves refining the price of the goods sold, introducing a price dispersion coefficient. The calculation formula is: ,in This displays prices for different stores of the same brand, where L represents the number of stores. The average price is represented by the price dispersion coefficient, which is used to measure the reasonableness of price fluctuations.

5. The trademark infringement risk analysis method based on big data according to claim 3, characterized in that, The extraction of features for infringing products also includes an evaluation of product sales volume, with the addition of the median sales volume over the past 14 days as a new sales feature. Sales skewness coefficient over the past 30 days The skewness coefficient is calculated by the ratio of the third central moment to the cube of the standard deviation. Let the sales data be... The mean is: The second-order central moment is: The third-order central moment is: The skewness coefficient is calculated as follows: The skewness coefficient reflects the symmetry of sales distribution and is used in conjunction with existing indicators to comprehensively assess sales anomalies. Brand keyword feature extraction incorporates semantic role labeling technology to identify the roles of keywords in the text, including but not limited to subject, object, and modifier. Feature vectors are constructed by weighting the importance of each role to strengthen the targeting of infringement.

6. The trademark infringement risk analysis method based on big data according to claim 3, characterized in that, The vector machine used in the service mark infringement risk assessment model employs an adaptive parameter adjustment method based on radial basis functions. Through cross-validation and gradient descent algorithms, it dynamically searches for the optimal kernel function parameters and penalty coefficients, achieving the best balance between classification accuracy and generalization ability on the training set. The model evaluation adopts a comprehensive indicator system, including accuracy, recall, F1 score, and Matthews correlation coefficient, to evaluate model performance from multiple dimensions for the model's ability to identify infringement.

7. The trademark infringement risk analysis method based on big data according to claim 6, characterized in that, The accuracy rate is: ； The recall rate was: ; Among them, (TP) is a true positive, (TN) is a true negative, (FP) is a false positive, and (FN) is a false negative; The correlation coefficient of Matthews is: 。 8. The trademark infringement risk analysis method based on big data according to claim 3, characterized in that, The text similarity calculation for the sale of infringing goods is performed using cosine similarity to calculate the trademark name and product name. The cosine similarity calculation is as follows: ,in It is the dot product of vectors A and B. It is the magnitude of vector A.

9. The trademark infringement risk analysis method based on big data according to claim 1, characterized in that, The process of tracing the product production chain includes the following specific steps: Raw material procurement: Supplier information is recorded using RFID tags; In the production process: the unique ID of the trademark printing equipment is bound to a blockchain-based notarized record; In the logistics process: GPS data is linked with smart contracts, and if the transportation route deviates from the preset trajectory, a risk check will be automatically triggered.

10. The trademark infringement risk analysis method based on big data according to claim 1, characterized in that, The real-time processing architecture includes, but is not limited to, the data source end of the e-commerce platform server edge node. It uses a lightweight model for preliminary screening to quickly filter low-risk data, uploads suspected infringing data to the cloud, and combines a heavyweight model with large-scale data storage for in-depth analysis to shorten monitoring latency. The personalized early warning information forecasts the impact of prominent infringements by trademark holders, on brand reputation, and on market share. This is achieved by constructing a brand value assessment model and setting the current brand value. The value loss rate due to infringement is r, and the estimated loss value is: .

Images