KR102533239B1 - Image library system and counterfeit detection method for artificial intelligence-based counterfeit product detection - Google Patents

Abstract

An image library system and a counterfeit detection method for artificial intelligence-based online counterfeit product detection provide an image library for determining a counterfeit product based on image data confirmed as a counterfeit product to identify a counterfeit product image more precisely on the web. In addition, the system identifies a counterfeit product image through the similarity comparison of a representative image for each counterfeit product seller and an image to be compared to reduce data processing processes, thereby improving an identification speed of an image to be analyzed. Moreover, an embodiment extracts a median value of images used by the counterfeit product seller to set a representative image and uses the extracted median value to identify counterfeit product image, thereby reducing a data capacity required for identifying counterfeit product images. Accordingly, an image detection system can be operated more efficiently.

Description

Image library system and counterfeit detection method for artificial intelligence-based counterfeit product detection

The present disclosure relates to an image library system for detecting counterfeit products and a method for detecting counterfeit products using the image library, and more specifically, to a method for detecting counterfeit products by constructing a library of fake product images used for detecting counterfeit products and using the counterfeit product detection method.

Unless otherwise indicated herein, material described in this section is not prior art to the claims in this application, and inclusion in this section is not an admission that it is prior art.

Luxury brands, such as Chanel and Hermes, suffer the greatest damage from counterfeit sales among all industries, and the size of the counterfeit market is also growing as the size of the luxury goods market increases. The size of the counterfeit goods market is steadily increasing worldwide, and the fashion industry is suffering the most damage. In addition, extensive damage is occurring to all premium brands such as electronic products, watches, and medical devices, but the results of crackdowns are very insufficient compared to the total damage caused by counterfeit products.

Compared to the rapidly growing counterfeit market, there is a lack of enforcement manpower and support, and platform operators are passive about cracking down on trading of infringing items. In addition, counterfeit sellers are avoiding crackdowns by taking advantage of loopholes in the law to punish counterfeit goods according to the amount of counterfeit goods, selling only a small amount online in the form of a purchasing agent, or by setting up a replica business with small businesses at the forefront.

Recently, even small and medium-sized brands with an absolute lack or no professional counterfeiting personnel are becoming targets of counterfeit products, beyond the distribution of counterfeit products of traditional luxury brands. In addition, the online shopping transaction amount, which had been continuously increasing, has rapidly increased due to the influence of the corona pandemic, and the distribution of counterfeit products has also moved to the online center, expanding the scope and complicating the distribution method. In addition, counterfeit products are produced and distributed in large quantities by exploiting the time difference between official distribution. In particular, the distribution of counterfeit products by small and medium-sized brands is directly inflowed into Korea in the form of a purchasing agent as well as overseas markets such as China and Southeast Asia, suffering double damage.

However, due to the unique characteristics of online distribution, a large number of sellers and buyers gather online, and it is not easy to identify the distribution stage of a product because it is free to open and close stores. From the brand's point of view, distribution in areas beyond control is being carried out on a large scale, so it is urgent to understand the actual situation and design a management strategy based on this.

In addition, the number of purchases by mistake is increasing due to the quality of counterfeit products that are so sophisticated that it is impossible to distinguish them from genuine products with the naked eye in photos, but countermeasures to crack down on them are insufficient. In particular, it is almost impossible to select only counterfeit products from hundreds of thousands of units of products circulated online, so there is a problem in that conventional counterfeit product detection systems cannot even begin to respond to counterfeit products circulated online.

1. Korean Patent Registration No. 10-1929025 (2018.12.07) 2. Korean Patent Publication No. 10-2021-0049409 (2021.05.06)

The artificial intelligence-based image library system and method for detecting counterfeit products online based on artificial intelligence according to the embodiment builds an image library for determining counterfeit products that minimizes data engineer's interference and enables online market to be mechanically analyzed, It is possible to determine a fake product on a web page only with the searched image. In addition, through the embodiment, it is possible to identify counterfeit distribution routes by classifying sellers using counterfeit images.

In the embodiment, a similarity is calculated by comparing an image used by a specific fake product seller with an image on a web page, and it is possible to determine whether the image of the web page is used by the fake product seller through the calculated similarity.

In the embodiment, through cosine similarity analysis using artificial intelligence, the entire image that has not been subjected to image preprocessing is compared with a fake product image to calculate the similarity, and based on the calculated similarity, it is possible to determine whether the image to be analyzed includes a fake product.

In addition, in the embodiment, the counterfeit distribution platform reporting process is automated and an advanced dashboard is provided to intuitively check the image detection results collected online, so that the manager directly inputs data and counterfeit detection results so that you can check

In addition, through the image library built in the embodiment, not only counterfeit products are detected, but also abnormal distribution channels of normal products are identified, and furthermore, the online distribution route of normal products is identified so that the brand can determine the distribution status and route of products and counterfeit products in detail. enable you to understand In particular, in the embodiment, it is possible to acquire abnormal distribution information including fake products.

An image library system for detecting artificial intelligence-based online counterfeit products according to an embodiment includes a counterfeit product sales information collection module that collects counterfeit product images used by counterfeit sellers and seller data handling counterfeit goods; Calculate the average hash value of each image using the pixel structure and color information of the image used by the counterfeit seller and the image to be analyzed, and calculate the similarity by comparing the average hash value of the image to be analyzed with the hash value of the image used by the counterfeit seller an image analysis module that; a determination module for determining whether an analysis target image is a fake image according to the calculated similarity; includes

An artificial intelligence-based online counterfeit product detection method according to another embodiment includes the steps of (A) collecting counterfeit product images used by counterfeit sellers and seller data handling counterfeit goods in a counterfeit sale information collection module; (B) In the image analysis module, the average hash value of each image is calculated using the pixel structure and color information of the image used by the counterfeit seller and the image to be analyzed, and the average hash value of the image to be analyzed and the hash of the image used by the counterfeit seller Comparing the values to calculate a degree of similarity; and (C) determining whether the analysis target image is a fake image according to the calculated similarity in a determination module. includes

The artificial intelligence-based image library system and counterfeit detection method for online counterfeit product detection as described above enable small and medium-sized businesses to identify the image of an infringing product and store the counterfeit product image. and products so that counterfeit infringement response can be carried out quickly.

In addition, based on the image data confirmed as fake, an image library for determining fake is provided so that fake images on the web can be more accurately identified.

In addition, in the embodiment, it is possible to improve the determination speed of the image to be analyzed by reducing the data processing process by discriminating the fake product image through similarity comparison between the representative image of each counterfeit seller and the comparison target image.

In addition, in the embodiment, the representative image is set by extracting the median value of the images used by the counterfeit seller, and the extracted median value is used to determine the counterfeit image, thereby reducing the amount of data required for counterfeit image identification. Through this, the image detection system can be operated more efficiently.

In addition, through the embodiment, it is possible to accurately determine the counterfeit image and to respond quickly to counterfeit sellers, thereby minimizing the company's loss and contributing to the protection of the brand of small and medium-sized enterprises and securing corporate competitiveness.

The effects of the present invention are not limited to the above effects, and should be understood to include all effects that can be inferred from the detailed description of the present invention or the configuration of the invention described in the claims.

1 is a diagram showing the configuration of an image library system for artificial intelligence-based counterfeit product detection according to an embodiment.
2 is a diagram showing a data processing configuration of a counterfeit detection server 100 according to an embodiment
3 is a diagram showing an embodiment of a user interface for detecting a fake image;
4 is a diagram showing an embodiment of a user interface for determining whether an image displayed on a user terminal includes a counterfeit product;
5 is a diagram showing a data processing process of a counterfeit image detection server according to an embodiment;

Advantages and features of the present invention, and methods for achieving them, will become clear with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and only the present embodiments make the disclosure of the present invention complete, and common knowledge in the art to which the present invention belongs It is provided to fully inform the holder of the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numerals designate like elements throughout the specification.

In describing the embodiments of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description will be omitted. In addition, terms to be described below are terms defined in consideration of functions in the embodiments of the present invention, which may vary according to the intention or custom of a user or operator. Therefore, the definition should be made based on the contents throughout this specification.

1 is a diagram showing the configuration of an image library system for artificial intelligence-based counterfeit product detection according to an embodiment.

Referring to FIG. 1, the library system for detecting artificial intelligence-based counterfeit products according to an embodiment includes a counterfeit image detection server 100, a genuine manufacturer 200, a manager terminal 300, and a user terminal 400. can be configured.

The counterfeit image detection server 100 builds a database of images already determined to be counterfeit and provides indexing and storage services for counterfeit images. Through this, in the embodiment, it is possible to find a counterfeit product by analyzing images on a page composed only of images retrieved from the web. In addition, sellers who use the counterfeit image are classified so that the counterfeit distribution route can be identified.

The artificial intelligence-based image library system and method for detecting counterfeit products online based on artificial intelligence according to the embodiment builds an image library for determining counterfeit products that minimizes data engineer's interference and enables online market to be mechanically analyzed, It is possible to determine a fake product on a web page only with the searched image. In addition, through the embodiment, it is possible to identify counterfeit distribution routes by classifying sellers using counterfeit images.

In addition, in the embodiment, the image used by the counterfeit seller is collected, and the similarity between the image used by the counterfeit seller and the image to be analyzed collected on the web is calculated. can figure it out In the embodiment, cosine similarity comparison of images and histogram of color meta-information are compared to compare both the structure and color of analyzed images, thereby increasing image detection accuracy and enabling easy grouping of image data on a database.

In addition, in the embodiment, counterfeit representative images for each seller are set using artificial intelligence, and it is possible to determine whether a product included in the comparison target image is a counterfeit product by comparing the similarity between the set image and the comparison target image. In addition, in the embodiment, cosine similarity analysis using artificial intelligence is performed to compare the entire image that has not undergone image preprocessing with a fake product image to calculate the similarity, and based on the calculated similarity, it can be determined whether the image to be analyzed contains a fake product there is. In addition, the embodiment automates the counterfeit distribution platform reporting process and provides an advanced dashboard that allows intuitive checking of image detection results collected online, so that administrators or users can directly enter data and counterfeit goods Allows you to check detection results. In addition, the embodiment provides an entry service through which a user or administrator can directly input counterfeit-related data and check a counterfeit detection result.

The genuine product manufacturer 200 receives counterfeit seller information and fake product information illegally distributed online from the counterfeit product detection server 100, checks the counterfeit product information, and takes legal action and a series of measures to delete online exposure.

When the manager terminal 300 accesses a counterfeit seller or a counterfeit seller page while surfing the web, the administrator directly controls online counterfeit sale by receiving the accessed page information and counterfeit seller information and transmitting the information to the counterfeit detection server 100 make it possible

The user terminal 400 downloads and installs the counterfeit image detection application from the counterfeit detection server 100 in a distributed manner. The fake image detection application outputs a fake image determination icon to the user terminal in the form of a widget. When the fake image determination icon displayed on the user terminal 400 is touched, the counterfeit image detection application extracts an analysis target image from the output screen of the user terminal 400 . Thereafter, by calculating the similarity between the extracted image to be analyzed and the representative image of the counterfeit product for each seller, it is possible to determine whether the image on the web displayed on the user terminal is genuine.

2 is a diagram showing a data processing configuration of the counterfeit detection server 100 according to an embodiment.

Referring to FIG. 2 , the fake product detection server 100 according to the embodiment includes a counterfeit sales information collection module 110, an image setting module 120, an image analysis module 130, a determination module 140, and a direct response module ( 150) may be configured. The term 'module' used in this specification should be interpreted as including software, hardware, or a combination thereof, depending on the context in which the term is used. For example, the software may be machine language, firmware, embedded code, and application software. As another example, the hardware may be a circuit, processor, computer, integrated circuit, integrated circuit core, sensor, micro-electro-mechanical system (MEMS), passive device, or combination thereof.

The fake product sales information collection module 110 collects fake product images used by fake product sellers and seller data handling fake products. The seller data may include detailed information of the seller, such as the type of counterfeit product sold by the seller, the cumulative number of sales, the selling price, the URL of the seller's web page, the seller's name, and the activity name. In the embodiment, the counterfeit sales information collection module 110 collects image data already determined to be counterfeit and counterfeit images used by counterfeit sellers, classifies the collected images by seller, product type, and brand to create an image library for counterfeit detection generate In the embodiment, through the generated image library, the manager can analyze the seller handling the fake product included in the fake product image detected online, the distribution route of the fake product, and the frequency and use method of the corresponding fake product image. In addition, in the embodiment, through the built image library, not only counterfeit products are detected, but also abnormal distribution channels of normal products are identified, and furthermore, the online distribution route of normal products is identified so that the brand can determine the distribution status and route of products and counterfeits in detail. enable you to understand In particular, abnormal distribution information including counterfeit products can be identified.

In addition, in the embodiment, a web page generated for the sale of counterfeit goods is crawled, counterfeit seller information is extracted, and a counterfeit seller list is generated using the previously stored counterfeit seller information. Thereafter, fake product images used by sellers included in the counterfeit seller list may be collected, and the collected images may be databased for each seller, brand, and product.

The image setting module 120 measures histograms of several fake product images used by each seller and extracts a median value of the measured histograms. Then, a representative image of a counterfeit product for each seller reflecting the median value of the extracted histogram is set. In the embodiment, the fake product representative image for each seller is an image generated by extracting an average value or a median value of various fake product image meta data used by each seller. In the embodiment, the fake product representative image corresponds to an image sample or reference image used by the seller to sell fake products.

In the embodiment, the image analysis module 130 calculates the average hash value of each image using the image used by the fake seller and the pixel structure and color information of the image to be analyzed, and calculates the average hash value of the image to be analyzed and the pixel structure used by the fake product seller The similarity is calculated by comparing the hash values of the images. In an embodiment, the image analysis module 130 measures a histogram of an image used by a counterfeit seller, and compares the measured histogram with the histogram of an image to be analyzed to calculate a first degree of similarity. In addition, through the analysis of the cosine similarity between the image used by the counterfeit seller and the image to be analyzed, the second similarity is calculated, and based on the calculated similarity, the first similarity, and the second similarity, whether or not the image to be analyzed includes a fake product can judge In the embodiment, cosine similarity comparison of images and histograms of color meta-information are compared to increase image detection accuracy and to easily group image data on a database.

The image analysis module 130 calculates an average hash value using the pixel structure and color information or characteristic information of the image used by the fake product seller and the image to be analyzed, and calculates the average hash value of the image to be analyzed and the hash of the image used by the fake product seller The similarity can be calculated by comparing the values. In an embodiment, the image characteristic information may include an image hash conversion value and the like.

In addition, the image analysis module 130 compares the histogram of the fake product representative image for each seller with the histogram of the analysis target image to calculate a first similarity between the analysis target image and the fake product representative image for each seller. In the embodiment, when detecting a fake product image online, it is not to detect whether or not the image to be analyzed includes a fake product through all the preprocessing of the image to be analyzed, but rather to detect the fake product representative image set to the median value of several images used by the seller and the image to be analyzed Counterfeit detection is performed through histogram comparison. Accordingly, it is possible to reduce the data processing process for image analysis.

In addition, the image analysis module 130 calculates a second similarity through cosine similarity analysis between the analysis target image and the fake product representative image for each seller. The cosine similarity is the degree of similarity between vectors measured using the cosine value of the angle between two vectors in the dot product space. When the angle is 0°, the cosine value is 1, indicating that the two vectors agree, and the cosine value of all other angles is less than 1.

In addition, in an embodiment, the image analysis module 130 may extract a product object and a background included in an analysis target image, and calculate a first similarity and a second similarity for each of the extracted product object and background. For example, the image analysis module 130 extracts a product object image and a background from the analysis target image, compares the extracted product object image with the product object extracted from the fake product representative image for each seller, and calculates a first similarity and a second similarity. . Thereafter, the first similarity and the second similarity may be calculated by comparing the background of the image to be analyzed with the background extracted from the representative image of the counterfeit product for each seller.

In this case, the first similarity is the result of comparing the histogram between the image to be analyzed and the fake product image for each seller, and the second similarity is the result of analyzing the cosine similarity between the image to be analyzed and the fake product image for each seller. Specifically, the image analysis module 130 extracts the product object and background included in the analysis target image, and compares the first similarity and cosine similarity of the extracted product object and the histogram of the fake product representative image product object for each seller. Calculate the second similarity. Thereafter, a first similarity comparing the background included in the image to be analyzed and the histogram of the background of the fake product representative image for each seller may be calculated, and a second similarity comparing the cosine similarity may be calculated.

In addition, in the embodiment, the image analysis module 130 may determine the similarity between the two images through a histogram, MSE calculation, SSIM comparison, and image hash conversion of the analysis target image and the counterfeit representative image for each seller. MSE (mean square error) calculation is to calculate the average of the distance difference between each pixel of the image and set the similarity according to the calculated average value. SSIM (structural similarity index) comparison is a method for increasing accuracy rather than similarity by MSE calculation, and calculates similarity by comparing structures of pixels constituting images. Image hash conversion is to convert an image compressed in various ways into a representative value by using a hash function in order to give unique data (eg, fingerprint) to the image. In an embodiment, a similarity may be calculated by comparing values obtained by converting two images using a hash function.

In addition, the image analysis module 130 extracts an analysis target image from the output screen of the user terminal when the fake product image determination icon displayed on the user terminal is touched, and determines the similarity between the extracted analysis target image and the counterfeit representative image for each seller. By calculating, it is possible to determine whether the web image output to the user terminal is genuine.

In the embodiment, the image analysis module 130 recognizes the image used by the counterfeit seller through an image recognition process through artificial intelligence machine learning, and recognizes the object and background included in the image to obtain image information for each counterfeit seller can figure it out In the embodiment, for image recognition, a data processing process of classifying and detecting a counterfeit object and background included in an image, and identifying and segmenting the object in pixel units may be performed. In addition, in an embodiment, the image analysis module 130 may perform an out of distribution detection process for non-learning pattern processing other than noise response. Distribution data detection outside learning is to identify whether an image input to artificial intelligence is learned probability distribution data or not. In the embodiment, stability and reliability can be increased by filtering out images that are difficult for an artificial neural network to judge through distribution data detection other than learning or processing them as exceptions. In addition, in the embodiment, in order to detect non-learning distribution data, a probability value indicating how confident a deep learning decision is is calibrated, or non-learning distribution data is generated with a generative adversarial network (GAN). and learn to improve detection accuracy.

In addition, in the embodiment, in order to reduce the size of the model while maintaining image recognition accuracy, a lightweight deep learning technology that simplifies calculation is used to determine the image used by the counterfeit seller and to expand it. In the embodiment, for image recognition, a convolution filter is modified in a Convolution Neural Network (CNN) to reduce the operation dimension or pruning and weight values to delete the weight of a neural network that has no significant effect. By reducing the floating point of , a quantization process that simplifies calculation is performed to enable data weight reduction. In addition, in the embodiment, the output of the previously trained large neural network is imitated by the small neural network to simplify calculation and maintain accuracy.

The determination module 140 determines whether the analysis target image includes a fake product according to the calculated similarity. In an embodiment, the determination module 140 may determine the analysis target image as a fake image when at least one of the similarity, the first similarity, and the second similarity exceeds a preset value.

Also, in an embodiment, the determination module 140 may calculate the probability that the analysis target image contains a counterfeit according to the first similarity and the second similarity. In an embodiment, the determination module 140 calculates the possibility of including a fake product in the image to be analyzed with a first similarity calculated through color comparison of the images, and calculates the possibility of including a fake product in the image to be analyzed with a second similarity. Thereafter, the probability that the image to be analyzed may contain a fake product may be finally determined by summing up the possibility of including a fake product of the image to be analyzed calculated by the first similarity and the second similarity.

In addition, in an embodiment, the determination module 140 may calculate a total similarity obtained by adding the first similarity and the second similarity, and calculate the probability that the analysis target image contains a fake product in proportion to the total similarity. In addition, in the embodiment, the determination module 140 assigns weights to the first similarity and the second similarity according to the shape and color characteristics of the counterfeit product, and then obtains an overall similarity obtained by summing the weighted first similarity and the first similarity. and the probability of including a counterfeit in the image to be analyzed may be calculated in proportion to the total degree of similarity.

In addition, in the embodiment, the determination module 140 determines the first similarity and the second similarity of the product object extracted from the analysis target image and the first similarity and second similarity of the background extracted from the analysis target image according to the type, background, and color of the fake product. A weight is assigned to each to calculate the total similarity. Thereafter, the possibility of including a counterfeit in the image to be analyzed may be calculated in proportion to the total similarity.

The direct response module 150 collects counterfeit seller and counterfeit report information from registered administrator terminals, and provides a counterfeit seller report form including a complaint and report form for legal processing of counterfeit sellers.

In addition, in an embodiment, the direct response module 150 provides an entry service that allows a user or manager to directly input counterfeit-related data and check a counterfeit detection result. For example, when a user terminal logs in to the web and inputs detection purpose, brand, and product information, the image analysis module 130 crawls the online market based on the input data, performs unsupervised clustering, and performs primary classification. User feedback on the result is received and detection conditions are set. In an embodiment, the detection condition may be set through a user input condition or by searching for seller products included in the counterfeit seller list. Thereafter, counterfeit detection is performed according to the detection conditions. In addition, in the embodiment, when a user confirms that a product is fake and requests a platform report from a user terminal, the platform can be automatically reported after learning with artificial intelligence by reflecting the confirmed information.

In order to provide the entry service, unsupervised clustering and clustering verification of text and images must be possible without developer intervention, so the image analysis module 150 according to the embodiment detects distribution products based on user input information and automatically Perform unsupervised clustering to cluster. In an embodiment, the image analysis module 130 advances while applying an existing clustering logic to product categories for which data has been accumulated, and enables clustering of new product categories based on this. In an embodiment, model learning may be performed by reflecting user feedback on the primary classification. Also, the image analysis module 130 extracts outliers from the list classified by clustering. For example, singularities of text and images can be extracted.

In addition, the image library system for detecting counterfeit products based on artificial intelligence according to the embodiment is used for the purpose of finding patterns similar to images used by counterfeit sellers in order to fingerprint images and perform functions necessary for counterfeit product management. can be utilized In the embodiment, even if the background and object included in the analysis target image are different from the image used by the counterfeit seller, the image pattern used by the counterfeit seller is compared with the analysis target image, and the similarity is calculated to determine whether the analysis target image includes a counterfeit product. do.

Similar images can be grouped through unsupervised clustering, but clustering each time an image is added is inefficient, so in the embodiment, using the hash value of the image, images with patterns similar to those used by counterfeit sellers can be found. let it be A hash value is a numeric value such as a password that abbreviates file characteristics to prove the identity of digital data. and compare it with the images used by counterfeit sellers. In addition, in the embodiment, in order to improve the accuracy of clustering, image clustering may be performed using a logic (perceptual hash) for obtaining a hash value difference from a specific image and a structural similarity index (SSIM) for obtaining a structural difference between images. .

3 is a diagram illustrating an embodiment of a user interface for detecting a fake image.

3, in the embodiment, when an image collected online is loaded into the image input window 10, the input image is compared with the representative image of a counterfeit product by seller, and the match rate, original item found, Seller found information may be provided. In the embodiment, when a seller handling a counterfeit product included in an input image is extracted, the seller group, all items handled by the seller (item total), linked item (item connected), item infringement (item infringement), update date (updated at), and seller threat level (Threat level) information may be extracted and provided to the manager. In an embodiment, the seller threat level may be determined according to threat factors including the sales period of the seller extracted by the determination module, the selling price of the counterfeit product, the price difference with the genuine product, and the total number of items handled by the seller.

4 is a diagram illustrating an embodiment of a user interface for determining whether an image displayed on a user terminal includes a counterfeit product.

Referring to FIG. 4A , in the embodiment, when a counterfeit image detection application is installed in a user terminal and the application is executed, a counterfeit image determination icon (detection) is output in the form of a widget on the output screen of the user terminal. Thereafter, when the user touches a fake product image determination icon (detection), as shown in FIG. 4B, in the embodiment, an analysis result of whether each image included in the output screen is included in a fake product is provided. In the embodiment, a Genuine Probability (G.P) value of each image displayed on the screen may be provided.

Hereinafter, an AI-based online counterfeit product detection method will be described in turn. Since the operation (function) of the artificial intelligence-based online counterfeit product detection method according to the embodiment is essentially the same as the function of the image library system for artificial intelligence-based online counterfeit product detection, overlapping descriptions with FIGS. 1 to 4 will be omitted. .

5 is a diagram illustrating a data processing process of a counterfeit image detection server according to an embodiment.

Referring to FIG. 5 , in step S100, the counterfeit product sales information collection module collects counterfeit product images used by counterfeit sellers and seller data handling counterfeits. Step S100 of the embodiment crawls the web page generated for the sale of fake products, extracts fake product seller information, creates a fake product seller list using the previously stored fake product seller information, collects fake product images used by the fake product seller, and collects the collected Images can be databased by seller, product type, and brand.

In step S200, a histogram of the collected fake product images used by the counterfeit seller is measured in the image setting module, a median value of the measured histogram is extracted, and a counterfeit representative image for each seller reflecting the extracted median value is set.

In step S300, the image analysis module calculates the average hash value using the pixel structure and color information of the image used by the counterfeit seller and the image to be analyzed, and compares the average hash value of the image to be analyzed with the hash value of the image used by the counterfeit seller Thus, the similarity can be calculated. In an embodiment, the image characteristic information may include an image hash conversion value and the like.

In addition, in step S300 , a first similarity between the image to be analyzed and the representative image of fake products by seller is calculated by comparing the histogram of the representative fake product image for each seller with the histogram of the image to be analyzed. In addition, the first similarity may be calculated by comparing the histogram of the image used by the counterfeit seller and the histogram of the image to be analyzed. In step S300, a second similarity is calculated by analyzing the cosine similarity between the representative image of the counterfeit product for each seller and the image to be analyzed. In addition, the second similarity may be calculated through cosine similarity analysis between the image used by the counterfeit seller and the image to be analyzed.

In addition, in the embodiment, the product object and background included in the image to be analyzed are extracted, and the first similarity comparing the extracted product object and the histogram of the product object of the counterfeit representative image by seller Calculate the second similarity comparing the cosine similarity do. Thereafter, a first similarity comparing the background included in the image to be analyzed and the histogram of the background of the fake product representative image for each seller may be calculated, and a second similarity comparing the cosine similarity may be calculated.

In addition, in the embodiment, the objects and backgrounds included in the analysis target image are extracted, and the object included in the image used by the counterfeit seller and the histogram of the object included in the analysis target image are compared. The first similarity and the cosine similarity are compared. 2 The degree of similarity can be calculated. Thereafter, a first similarity comparing the background included in the image to be analyzed and a histogram of the image background used by the counterfeit seller and a second similarity comparing the cosine similarity may be calculated.

In addition, in step S300, when a fake product image determination icon is touched on the user terminal, an analysis target image is extracted from a screen displayed on the user terminal, and a similarity between the extracted analysis target image and a fake product representative image for each seller is calculated, and the user It is possible to determine whether the image printed on the terminal is genuine.

In step S400, it is determined whether the image to be analyzed is a fake image according to the degree of similarity, the first degree of similarity, and the second degree of similarity calculated by the determination module.

In step S500, the counterfeit seller and counterfeit reporting information are collected from the manager terminal registered in the direct response module, and a counterfeit seller reporting form including complaint and report is provided for legal processing of the counterfeit seller. In addition, in the embodiment, in step S500, an entry service is provided so that the user or manager can directly input counterfeit-related data and check the counterfeit detection result. For example, when a user terminal logs in to the web and enters detection purpose, brand, and product information, the online market is crawled based on the entered data, unsupervised clustering is performed, and user feedback on the primary classification result is received. to set detection conditions. In an embodiment, the detection condition may be set through a user input condition or by searching for seller products included in the counterfeit seller list. Thereafter, counterfeit detection is performed according to the detection conditions. In addition, in the embodiment, when a user determines that a fake product is a counterfeit and requests a platform report from a user terminal, the corresponding confirmed information is reflected, learned by artificial intelligence, and the platform can be automatically reported.

The artificial intelligence-based image library system and counterfeit detection method for online counterfeit product detection as described above enable small and medium-sized businesses to identify the image of an infringing product and store the counterfeit product image. and products so that counterfeit infringement response can be carried out quickly.

In addition, based on the image data confirmed as fake, an image library for determining fake is provided so that fake images on the web can be more accurately identified.

In addition, in the embodiment, it is possible to improve the determination speed of the image to be analyzed by reducing the data processing process by discriminating the fake product image through similarity comparison between the representative image of each counterfeit seller and the comparison target image.

In addition, in the embodiment, the representative image is set by extracting the median value of the images used by the counterfeit seller, and the extracted median value is used to determine the counterfeit image, thereby reducing the amount of data required for counterfeit image identification. Through this, the image detection system can be operated more efficiently. In addition, through the embodiment, it is possible to accurately determine the counterfeit image and to respond quickly to counterfeit sellers, thereby minimizing the company's loss and contributing to the protection of the brand of small and medium-sized enterprises and securing corporate competitiveness.

The disclosed content is only an example, and can be variously modified and implemented by those skilled in the art without departing from the subject matter of the claim claimed in the claims, so the protection scope of the disclosed content is limited to the specific not limited to the examples.

Claims (9)

In the image library system for artificial intelligence-based online counterfeit product detection,
a counterfeit product sales information collection module that collects counterfeit product images used by counterfeit sellers and seller data handling counterfeit goods;
Counterfeit product representation by seller used for counterfeit sales by measuring metadata including histograms of multiple counterfeit images used by each seller, extracting the median value of the measured metadata, and reflecting the median or average value of the extracted metadata an image setting module that sets an image;
Calculate the average hash value of each image using the pixel structure and color information of the image used by the counterfeit seller and the image to be analyzed, and calculate the similarity by comparing the average hash value of the image to be analyzed with the hash value of the image used by the counterfeit seller an image analysis module that; and
a judgment module for determining whether an analysis target image is a fake product image according to the calculated similarity; contains
The image analysis module
Histogram of the image to be analyzed and representative images of counterfeit products by seller, MSE (mean square error) calculation that calculates the average of the distance difference between each pixel, SSIM (structural similarity index) comparison that compares pixel structures, image hash for unique data assignment Through (image hash) conversion, the similarity between the two images is identified,
Measure the histogram of the representative image of each fake product by seller, compare the measured histogram with the histogram of the image to be analyzed to calculate the first similarity between the image to be analyzed and the representative image of each fake product by seller, and the image used by the counterfeit seller and the image to be analyzed Calculate a second similarity through cosine similarity analysis between
the judgment module; silver
determining the image to be analyzed as a fake image based on at least one of similarity, first similarity, and second similarity;
Calculate the total similarity by adding the first similarity and the second similarity, and calculate the possibility of including fakes in the image to be analyzed in proportion to the total similarity
The image analysis module
Through artificial intelligence machine learning, counterfeit objects and backgrounds included in images are classified and detected, and images are identified and segmented in pixel units to perform image recognition (segmentation) to detect fakes. An image library system for artificial intelligence-based online counterfeit product detection, characterized by identifying images used by sellers and recognizing objects and backgrounds included in images to identify image information for each counterfeit seller.
The method of claim 1, further comprising: the fake product sales information collection module; silver
Crawling the web page created for the sale of counterfeit goods, extracting counterfeit seller information, creating a counterfeit seller list using the pre-stored counterfeit seller information, collecting counterfeit images used by the counterfeit seller, and using the collected images as a seller, An image library system for detecting online counterfeit products based on artificial intelligence, characterized by database creation by product type and brand.
According to claim 1, wherein the image analysis module; silver
Extract the product object and background from the image to be analyzed and the image used by the counterfeit seller, respectively;
Calculate a first similarity by comparing the histogram of the product object of the image to be analyzed and the product object of the image used by the counterfeit seller, and calculate a second similarity by comparing the cosine similarity,
Artificial intelligence-based online counterfeit product detection, characterized by calculating a first similarity by comparing the background included in the image to be analyzed and a histogram of the background included in the image used by the counterfeit seller, and a second similarity by comparing the cosine similarity image library system for
The method of claim 1, wherein the artificial intelligence-based image library system for online counterfeit product detection
A direct response module that collects counterfeit sellers and counterfeit report information from registered administrator terminals, and provides a counterfeit seller reporting form including a complaint and a report for legal processing of the counterfeit sellers; Image library system for artificial intelligence-based online counterfeit product detection, characterized in that it further comprises.
According to claim 1, wherein the image analysis module; silver
When the fake product image determination icon is touched on the user terminal, the analysis object image is extracted from the screen displayed on the user terminal, and the similarity between the extracted analysis object image and the counterfeit representative image for each seller is calculated, and the image displayed on the user terminal An image library system for detecting online counterfeit products based on artificial intelligence that identifies whether the product is genuine or not.
In the artificial intelligence-based online counterfeit product detection method,
(A) collecting counterfeit product images used by counterfeit sellers and seller data handling counterfeit goods in a counterfeit sale information collection module;
(B) In the image analysis module, the average hash value of each image is calculated using the pixel structure and color information of the image used by the counterfeit seller and the image to be analyzed, and the average hash value of the image to be analyzed and the hash of the image used by the counterfeit seller Comparing the values to calculate a degree of similarity;
(C) determining whether the image to be analyzed is a fake image according to the calculated similarity in a judgment module; and
Step (A) above; Is
In the image setting module, metadata including a histogram of several counterfeit images used by each seller is measured, the median value of the measured metadata is extracted, and the median or average value of the extracted metadata is used for sales of counterfeit products. Setting a fake product representative image for each seller; including,
Step (B) above; Is
Histogram of the image to be analyzed and representative images of counterfeit products by seller, MSE (mean square error) calculation that calculates the average of the distance difference between each pixel, SSIM (structural similarity index) comparison that compares pixel structures, image hash for unique data assignment Figuring out the similarity between the two images through (image hash) conversion; and
Measure the histogram of the representative image of each fake product by seller, compare the measured histogram with the histogram of the image to be analyzed to calculate the first similarity between the image to be analyzed and the representative image of each fake product by seller, and the image used by the counterfeit seller and the image to be analyzed Calculating a second similarity through inter-cosine similarity analysis; and
Through artificial intelligence machine learning, counterfeit objects and backgrounds included in images are classified and detected, and images are identified and segmented in pixel units to perform image recognition (segmentation) to become a counterfeit seller. Identifying images used by and recognizing objects and backgrounds included in the images to identify image information for each counterfeit seller; including,
Step (C) above; Is
determining an analysis target image as a fake product image based on at least one of similarity, first similarity, and second similarity; and
calculating a total similarity obtained by summing the first similarity and the second similarity, and calculating a probability that an analysis target image contains a counterfeit product in proportion to the total similarity; Artificial intelligence-based online counterfeit product detection method comprising a.
The method of claim 6, wherein the step of (A); Is
crawling a web page generated for the sale of counterfeit goods, extracting counterfeit seller information, and generating a counterfeit seller list using the pre-stored counterfeit seller information; and
Collecting fake product images used by the counterfeit seller and converting the collected images into a database for each seller, product type, and brand; An artificial intelligence-based online counterfeit product detection method comprising a.
The method of claim 6, wherein the step of (B); Is
Extracting the product object and background included in the analysis target image;
Calculating a first similarity by comparing the extracted product object and the histogram of the image product object used by the fake product seller and a second similarity by comparing the cosine similarity;
Calculating a first similarity by comparing a background included in the analysis target image with a histogram of an image background used by a counterfeit seller and a second similarity by comparing cosine similarity; Artificial intelligence-based online counterfeit product detection method comprising a.
The method of claim 6, wherein the artificial intelligence-based online counterfeit product detection method
(D) collecting counterfeit seller and counterfeit report information from the manager terminal registered in the direct response module, and providing a counterfeit seller report form including a complaint and a report for legal processing of the counterfeit seller; Artificial intelligence-based online counterfeit product detection method further comprising.

Images