KR101933696B1 - Vod service system based on ai video learning platform - Google Patents

Abstract

According to an embodiment of the present invention, an AI image learning platform-based VOD service system includes: a first server storing a VOD file, and providing a service for transmitting the VOD file in response to a request from a user device; a second server receiving the VOD file from the first sever to execute AI image learning for the individual VOD file, calculating a weight corresponding to the VOD file as a result of the AI image learning, and transmitting the weight corresponding to VOD file in response to a VOD transmission request generated from the user device; a DB storing an ID of the VOD file learned by the second server and the weight calculated as a result of the AI image learning; and the user device requesting the first server for the transmission of an image file, receiving the image file from the first server, and receiving a weight corresponding to the requested image file from the second server to execute the image file done with the application of the weight. Thus, the present invention is capable of enabling mobile VOD viewers to enjoy high-resolution videos at a lower communication expense.

Description

{VOD SERVICE SYSTEM BASED ON AI VIDEO LEARNING PLATFORM}

The present invention relates to a VOD service system based on an AI image learning platform. More specifically, the present invention performs image learning according to an AI-based SRCNN to calculate a weight required for high-quality reconstruction from a high-quality VOD file, and then calculates a low-quality VOD file using a weight corresponding to the VOD file, To a VOD service system capable of restoring to a file.

 The penetration rate of smart devices is increasing all over the world. In particular, the penetration rate of smartphones in Korea is 91%, which is the number one ranking among 56 countries in the world. Also, the number of video viewing through mobile devices continues to increase. In Korea, mobile video traffic increased by 74% in 2017 compared to 58% .

In recent years, artificial intelligence technology has been actively studied and developed in mobile devices. In particular, there has been a remarkable advancement in the performance of mobile devices that can utilize artificial intelligence technology starting from 2015. AI Open Source, which is open to global corporations such as Google, Facebook, Apple, and MS, has begun to support mobile devices, enabling the use of artificial intelligence in mobile devices.

Meanwhile, in order to provide adaptive streaming according to the user environment, the existing VOD service has to prepare video formats for each bitrate (270p, 360p, 480p, 720p, 1080p, etc.), terminals and browsers. As a result, content providers who provide VOD services incurred encoding costs and traffic costs, which resulted in user charges.

The increasing size of the video size from the viewpoint of the mobile VOD viewers is a burden of the communication fee, which burdens the viewers and does not lead to the profit of the provider providing the high-definition VOD service. Accordingly, it is required to reduce the burden on the communication capacity experienced by the user and to provide a video with a high-definition VOD file.

On the other hand, the prior art related to the improvement of the resolution of the moving image is disclosed in Japanese Laid-Open Patent Publication No. 10-2014-0081481 (Method and apparatus for restoring super-resolution image using block unit image matching).

The present invention develops an artificial intelligent VOD player capable of playing a high-resolution image from a multi-platform and a multi-browser using only a low-resolution image file using an artificial intelligent learning algorithm that restores a low-resolution image to a high resolution, The goal is to develop a learning platform.

A VOD service system based on an AI image learning platform according to an embodiment of the present invention includes a first server that stores a VOD file and provides a transmission service of the VOD file in response to a request from a user device, A weight corresponding to an individual VOD file is calculated as a result of the AI image learning, and a corresponding VOD file corresponding to a VOD transmission request generated by the user device is received A DB for storing the ID of the VOD file learned by the second server and the weight calculated as a result of the AI image learning of the VOD file, Receives the image file from the first server, receives the weight corresponding to the image file requested to be transmitted from the second server, and applies the weight It may include a user device to execute an image file.

The present invention has an effect of allowing a mobile VOD viewer to enjoy a high-resolution image with less communication cost than the conventional mobile VOD viewer.

It is possible to reproduce a high-resolution image with only a relatively short communication time, thereby enabling a high-quality image to be reproduced even in a foreign country having a poor network infrastructure.

The AI learning platform and the high-resolution image reproducing technology according to the embodiment of the present invention can be applied to various business models, and accordingly high-resolution images can be received in various industrial fields related to image such as IPTV, VOD, and CCTV.

The present invention can provide an AI learning platform that can be used in all industrial fields from an artificial intelligence image processing service specialized in a specific area learned by upgrading the image reproducing technology.

FIG. 1 and FIG. 2 are diagrams showing a configuration of a VOD service system according to an embodiment of the present invention.
3 is a block diagram illustrating a configuration of a second server according to an embodiment of the present invention.
4 is a flowchart showing a sequence of an AI image learning operation according to an embodiment of the present invention.
5 is a flowchart showing a sequence of operations for performing resolution restoration in a second server according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail.

It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, the terms "comprises", "having", and the like are intended to specify the presence of stated features, integers, steps, operations, components, parts, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

FIG. 1 and FIG. 2 are diagrams showing a configuration of a VOD service system according to an embodiment of the present invention.

1, the VOD service system according to the embodiment of the present invention may include a first server 100, a second server 200, and a DB 220 interlocked with a second server, So that the AI image learning can be performed. In this case, the AI image learning may be an operation of calculating a weight (weight parameter) required for reconstructing a high-resolution image from a low-resolution image, based on SRCNN-based high-resolution image restoration technology, which is a deep learning learning algorithm.

First, the first server 100 may mean a general VOD server. The first server 100 may include a separate DB for storing VOD files, though not shown in the figure, and the first server 100 may transmit the VOD file stored in the DB to the user device.

The second server 200 may mean a server that performs AI image learning according to an embodiment of the present invention. At this time, the second server 200 can perform AI image learning on the VOD files uploaded to the first server 100. In addition, the second server 200 may separately store the weight generated as a result of AI image learning in the DB 220 interlocked with the second server, and the generated weight may be stored together with the ID of the corresponding VOD file have. Accordingly, when the first server 100 receives a request for transmission (e.g., streaming) of a specific VOD file, the VOD ID information is transmitted to the second server 200 side, and the second server 200 detects a weight corresponding to the ID And can be provided to the user equipment side.

At this time, the second server 200 can set the capacity of the calculated weight to be less than a predetermined size (for example, 10 Mb or less) at the time of AI image learning.

Referring to the right side of the second server 200 of FIG. 1, an example of the operation of the AI image learning performed by the second server 200 is schematically shown. The AI image learning performed by the second server 200 will be briefly described as follows.

The AI image learning performed by the second server 200 according to the embodiment of the present invention uses a deep learning algorithm (for example, SRCNN, Super Resolution Convolution Neural Networks) to calculate a weight for restoring a low resolution image to a high resolution image . The weight extracted in this way is used to convert a low-resolution VOD file into a high-resolution VOD file through an artificial intelligent testing process (a learning process is performed in a second server and a test process can be performed by a user device) And may be a weight parameter used.

The basic CNN operation method uses a method of scanning a partial area of an image through a filter one by one and finding a value therefor and learning. CNN's goal is to find a filter with the appropriate weight.

The DB 220 can store the weight, which is a learning result of an individual VOD file, for each VOD ID.

The weight stored in the DB 220 may be designated for each VOD ID as shown in Table 1. For example, in the Weight stage, a weight (parameter of weight) required for restoring image resolution on the user equipment, (E.g., improved from 480p to 720p).

VOD ID Learning Weight A000001 480p → 720p A000002 720p → 1080p A000003 1080p → UHD .... ....... Z025432 1080p → UHD

According to various embodiments, the VOD ID stored in the DB 220 may be assigned based on content related information such as a content item (e.g., a movie, a music video, a drama, etc.), a playing time, plot, characters, . For example, an ID giving method can be determined so that the movie-related VOD is A and the drama-related VOD is B and the ID is started.

The AI image learning operation performed in the second server 200 has been described with reference to FIG. Hereinafter, an operation of providing a high-definition VOD performed through a user equipment will be described with reference to FIG.

Referring to FIG. 2, a first server 100, a second server 200, and a DB 220, which are basic components according to an embodiment of the present invention, are shown together with a user device 300. As described in FIG. 1, the first server 100 may refer to a server that transmits VOD (including streaming) in a conventional general manner. When a request for a VOD transmission is received from the user device 300, the VOD file is transmitted to the user device 300 Via wireless communication. At this time, the VOD file transmitted from the first server 100 to the user device 300 may be a VOD file having a resolution lower than a reference value.

The second server 200 receives the information on the specific VOD file transmitted from the first server 100 to the user device 300, searches the DB 220 for the weight information corresponding to the specific VOD file, . Alternatively, the second server 200 may receive the VOD file request signal directly from the user device 300 together with the first server 100, and may search for the weight information corresponding to the requested VOD file and provide the weight information to the user device 300.

That is, the first server 100 transmits a VOD file (hereinafter, the VOD received from the first server is referred to as a first VOD file) to the user device 300 through the Internet, and the second server 200 searches Weight information to the user device 300 through the Internet.

As a result, the user device 300 can receive both the low-resolution VOD file and the weight, and the high-resolution VOD file (hereinafter, referred to as the high-resolution VOD file restored on the user device 300) 2 VOD file) can be played back.

The user device 300 plays back the first VOD file received from the first server through the AI high resolution VOD playback program 301. At this time, in the AI high definition VOD playback program 301, weight information corresponding to the program is present in the user device 300 (Or whether it is streamed from the second server 200). If there is a corresponding weight, the second VOD file (high-resolution VOD) can be generated through the restoration process with the original VOD file and the weight. The AI high-definition VOD playback program 301 can be converted into an HD video if the original VOD file is an SD video and converted into an FHD video if it is an HD video. Furthermore, the AI high-resolution VOD playback program 301 may restore the SD image to the FHD image according to the weight information type. The resolution (size) of the second VOD file to be reconstructed at this time may be determined based on size information (for example, items described in the 'Learning weight' section of Table 1) included in the weight information.

3 is a block diagram illustrating a configuration of a second server according to an embodiment of the present invention.

3, the controller 230 includes a resolution determining unit 231, an AI image learning unit 232, an object image extracting unit 233, a content information collecting unit 234, a user viewing information collecting unit A user viewing preference information calculation unit 236, and a user viewing preference information calculation unit 236. [

The communication unit 210 can use a network for transmitting / receiving data between a user device and a server, and the type of the network is not particularly limited. The network may be, for example, an IP (Internet Protocol) network that provides a large capacity data transmission / reception service through an Internet Protocol (IP) or an All IP network that integrates different IP networks. The network may be a mobile communication network including a wired network, a Wibro (Wireless Broadband) network, a mobile communication network including a WCDMA, a High Speed Downlink Packet Access (HSDPA) network and an LTE (Long Term Evolution) network, ), A 5G (Five Generation) mobile communication network, a satellite communication network, and a Wi-Fi network, or a combination of at least one of them.

The communication unit 210 according to the embodiment of the present invention may receive a request signal for searching the user device 300 or the first server 100 for a weight corresponding to a specific VOD file and transmitting the weight to the user device 300 side. Also, the communication unit 210 may transmit a weight corresponding to the requested VOD to the user equipment 300 side. In addition, the communication unit 210 may collect user personal information and content information related to the VOD from a first server, a user device, and other web servers.

The DB 220 stores weight information calculated by the second server 200 in cooperation with the second server 200. According to various embodiments, the DB 220 may be configured separately from the second server 200, and may receive and store weight information transmitted from the second server 200 through wired / wireless communication.

The DB 220 may store weight information for each VOD file. The weight information includes a weight parameter, which is a weight parameter required to convert a low-quality image into a high-quality image according to an AI-based image processing operation, (E.g., HD-UHD (information meaning resolution-corrected from HD to UHD) together with the final resolution information (e.g., HD-UHD).

The control unit 230 may include a resolution determination unit 231, an AI image learning unit 232, an object image extraction unit 233, a content information collection unit 234, a user viewing information collection unit 235, and a user viewing preference information calculation unit 236.

The control unit 230 uses the deep learning algorithm called SRCNN (SRCNN) to reconstruct a low-resolution image into a high-resolution image, Can be calculated. In addition, the controller 230 collects information on the user's viewing information and VOD moving picture information obtained from the first server 100, classifies the collected information according to a predetermined criterion, and provides VOD (e.g., VOD recommendation per user interest).

More specifically, the control unit 230 may calculate weight information required to restore a low-quality file to a high-quality file through the resolution determination unit 231, the AI image learning unit 232, and the object image extraction unit 233, The user viewing information collecting unit 235, and the user viewing preference information calculating unit 236. The VOD service can be provided according to the user characteristics.

First, the resolution determination unit 231 can determine whether the resolution of the VOD file acquired from the first server 100 is higher than a predetermined level. The AI image learning unit 232 can determine an image learning method for the acquired image according to the resolution determination result. If the resolution determining unit 231 determines that the resolution is higher than a predetermined level as a result of the determination of the resolution of the first VOD file, the AI image learning unit 232 extracts a low-resolution image based on the first VOD file AI image learning (eg, SRCNN) can be performed using the extracted low-resolution image and high-resolution image patches. At this time, the AI image learning unit 232 calculates weight information, which is information required to implement a high-resolution image from a low-resolution image as a result of image learning. The AI image learning unit 232 may control to store the calculated weight information together with the ID information of the corresponding VOD file in the DB 220 separately.

On the other hand, if the resolution of the first VOD file obtained from the first server 100 is a low-quality image with a resolution lower than a predetermined resolution, the weight can be calculated by learning based on a plurality of images extracted from the VOD file. At this time, all of the plurality of images extracted from the first VOD file are low-quality images of less than predetermined resolution, and the weight that can improve the resolution can be calculated by using the self-similarity of the plurality of extracted low-quality images. In this case, the AI image learning unit 232 can perform AI image learning in a manner other than SRCNN according to various embodiments of the present invention, and calculate a weight based on the AI image learning.

The resolution determination operation of the resolution determination unit 231 may be omitted and the same video learning method may be performed for all the VOD files at once.

The AI image learning unit 232 performs AI image learning according to the SRCNN scheme. However, the AI image learning unit 232 may perform AI image learning according to other methods (e.g., VDSR, DRCN) according to various embodiments. And calculate weight information for each VOD file.

When extracting the weight from the first VOD file, the object image extracting unit 233 selectively extracts only some of the images of the first VOD file when the size of the entire image is equal to or higher than a predetermined level, or the capacity of the weight is equal to or greater than a predetermined capacity level Can be sampled. Some images sampled by the object image extracting unit 233 may correspond to, for example, an image of a center region excluding corner portions. In addition, the object image extracting unit 233 may perform sampling (e.g., object image sampling) on an area in which a predetermined object (e.g., a face) of the images included in the first VOD file is detected. The sampled image extracted by the object image extracting unit 233 can be used to calculate the object image weight through the AI image learning performed by the AI image learning unit 232. [ The weight for the object image may be provided when requesting center-based resolution correction in the user equipment 300. The weight for the object image may be a resolution of a predetermined region (region sampled by the object image) of the image region of the entire VOD file Lt; / RTI > Since the weight for the object image is a value that is applied only to a part of the image rather than the entire image, it has the effect of reducing the size of the weight capacity and saving resources and time required for restoration of the resolution of the image.

In one embodiment, when resolution correction using an object-based weight is provided according to the above-described embodiment, the user device 300 can request and perform object-based weight search and resolution correction.

For example, if there is no information about the image file to be reproduced by the user device 300, the user device 300 can not acquire the weight for the image file, and therefore it may be difficult to perform the resolution correction. For example, the video file may be a video program without title (or ID), or a clip video that edits a partial scene of the movie.

In this case, the user device 300 may extract one or more objects from the image file and identify the extracted objects. The previously learned model may be utilized for identification of the object, and the learning of the model may be performed in advance in the user device 300 or at least one server. For example, the user device 300 can identify and recognize a specific object such as a character, an actor, a building, and a small object appearing in a video file, and can search for and acquire the type of the corresponding video and the scene of the corresponding video.

The user equipment 300 acquires a weight from the user equipment 300 or the second server 200 based on the acquired information. According to an exemplary embodiment, when there is no corresponding weight, the user device 300 may search the server for a high-quality image corresponding to the image file, and perform AI image learning to acquire a weight using the searched high-quality image. Since the learning for weight acquisition is time consuming, the user device 300 may perform the above-described operations on the at least one image file stored in the streaming target play list of the user or stored in the user device 300, You can advance through the server in advance and acquire and save the weight when learning is completed.

As another example, if the weight corresponding to the image file to be reproduced by the user device 300 is not stored in the user device 300 or the second server 200, the user device 300 can perform object-based resolution restoration without a separate learning procedure have.

For example, the user device 300 may extract one or more objects from an image file and identify the extracted objects according to the method described above. For example, the user device 300 can identify and recognize a specific object such as a character, an actor, a building, and a small object appearing in the image file, and acquire the corresponding object image weight from the user device 300 or the second server 200 . For example, even though the weight for a particular movie is not stored, the weight for the actor in the movie may already be stored. In this case, the user device 300 may acquire a weight for the corresponding actor, and perform resolution restoration on the image of the actor based on the obtained weight.

In addition, an object may include a generic object that is not specified. For example, objects such as a sidewalk block, a telephone pole, a building, a car, etc. are not specified, and a weight classified according to a predetermined criterion can be stored in advance. Accordingly, when the user device 300 is identified as having a general object such as a roadmap, a building, or an automobile, the user device 300 acquires a corresponding weight and then performs resolution restoration for each general object to improve the overall quality of the image .

The content information collecting unit 234 can collect content information provided from the first server 100 and based on the VOD file having the history of performing the AI image learning. The content information is information required to classify the VOD files according to items. For example, the content information is a content type (e.g., movie, drama, M / V, entertainment, animation, (Name of the movie), season and time, story, characters, and the like.

According to various embodiments, the content information collecting unit 234 may collect content information (content information for VOD with a large number of views) frequently requested by the user from the user so that the user can determine the property of the content to be requested later As shown in FIG. The content information collecting unit 234 can collect internal content information such as a name and a turn included in the first VOD file. Also, the content information collecting unit 234 may further collect external content information such as plot, characters, and the like based on the internal content information included in the first VOD file.

The content information collecting unit 234 may control the operation of the AI image learning unit 232 based on the collected content information according to various embodiments. For example, if there are 20 VODs classified as 'Drama / Kimchi / 1 time' (eg, if there are a total of 1 hour of content, there may be a number of 5 minute divided video clips) VODs can be classified into one bundle. The AI image learning unit 232 further adds a VOD that holds similar thumbnail images among the same classified contents based on the content information among the AI image learning operations for calculating a weight for any of the 20 contents, Thereby performing image learning.

The user viewing information collection unit 235 may collect information based on the VOD request signal received from the user device 300. In addition, the user viewing information collecting unit 235 may collect viewing information (e.g., view number information) of the user from the first server 100. The user viewing information collecting unit 235 may collect user information (sex, age group, country, etc.) from the user account. The user viewing information collecting unit 235 may collect information related to the number of accesses per user account (for example, the average number of accesses per month), an ID of the viewing VOD, or classification item information (for example, may be recorded in the form of a viewing VOD list per user) Can be collected. According to various embodiments, the user viewing information collection unit 235 may further collect the average number of times of repeated playback of the same VOD content, main resolution selection information, and the like.

The user viewing preference information calculation unit 236 may calculate information related to the user's VOD preference based on the information collected by the content information collection unit 234 and the user viewing information collection unit 235. [ The user viewing preference information calculation unit 236 calculates a viewing preference preference value for a user having a predetermined number of times (for example, three times / one week) or more of the number of VOD viewing times or the number of accesses to the first server 100 (VOD service web page or application) Information can be calculated. Alternatively, the user viewing preference information calculation unit 236 may calculate viewing preference information based on the VOD file having the viewing history and the content information related to the VOD file with respect to all users. Alternatively, the user viewing preference information calculation unit 236 may calculate viewing preference information for a specific age group or gender based on the user information.

The user viewing preference information calculation unit 236 may utilize the audience information collected from the user viewing information collection unit 235 to determine a user's interest VOD. The user viewing preference information calculation unit 236 can calculate the characteristics of the VOD file preferred by the user based on the viewing information such as the viewing VOD file list for each user. At this time, the user viewing preference information calculation unit 236 classifies the information of each of the viewing VOD files of the user according to the content information, and then the information of the viewed items over the preset number (or rate) It can be judged. For example, the user viewing preference information calculation unit 236 may assign identification values for the classification items to each of the viewing VOD lists of the user.

The user viewing preference information calculation unit 236 may then determine preference VOD item information of the user based on the identification value information provided in the viewing VOD list of the specific user. For example, when it is determined that a predetermined ratio or more of the identification values assigned to the viewing VOD list of the arbitrary user A corresponds to the drama (kind) and the reason (genre), the user viewing preference information calculation unit 236 calculates Can be judged as (1) drama and (2) reason. The user viewing preference information calculation unit 236 may provide recommendation information to the first server side to recommend the VOD classified as the reason and drama to the user. Accordingly, when the user accesses the first server, the first server displays a screen for recommending the corresponding VOD on the connection page screen for the corresponding user when there is a drama that is an object of the newly updated VOD file, Information can be delivered as push notifications.

In addition, when the VOD (e.g., a series) determined to be preferred by a plurality of users is updated in the first server 100, the user viewing preference information calculation unit 236 acquires the VOD in the second server in preference to the VOD, Learning can be controlled. That is, the controller 230 may determine the priority of the AI video learning of the first VOD obtained from the first server 100 based on the user's viewing preference information calculated by the user viewing preference information calculator 236.

According to various embodiments, when setting the priority of the AI video learning of the first VOD obtained from the first server 100, the controller 230 may give preference to the preference images of the loyal users. The loyalty may be determined based on the number of connections to the VOD server and the number of viewing times. For example, if a user A having a preference for the number of times of connection of the first server 100 and a number of times of watching the VOD through the first server 100 is higher than a predetermined level, The controller may assign a priority to the AI drama in order to perform AI image learning.

Also, according to various embodiments of the present invention, the controller 230 may further include a keyword-based VOD recommendation unit (not shown). The keyword-based VOD recommender may learn a story and a user review of the VOD on the learned VOD file by artificial intelligence, and recommend a VOD file related to a trend keyword recommended in a web server or the like. For example, the keyword-based VOD recommender may determine that the content of the corresponding name or the content appearing on the actor is a content that is an issue on the basis of the drama name, actor name, .

4 is a flowchart showing a sequence of an AI image learning operation according to an embodiment of the present invention.

The second server 200 according to the embodiment of the present invention may perform the operation 405 of acquiring the VOD file from the first server. Then, the second server 200 may perform an operation 410 for calculating the weight required for AI image learning and high-quality reconstruction of the acquired VOD file. At this time, the second server 200 calculates a high image quality restoration weight of the VOD file obtained through the controller 230.

Then, the controller 230 of the second server 200 may perform the operation 415 of classifying the calculated weight according to the content ID and storing the classified weight in the DB 220.

The second server 200 can calculate the resolution restoration weight corresponding to the VOD file held in the first server 100 and retain the resolution restoration weight in the separate DB 220 according to the above procedure. The larger the parameter size of the weight, the higher the resolution recovery rate will be, but the VOD service availability can be expected to decrease. The size of the weight thus calculated can be specified to maintain a constant capacity (e.g., less than 10 MB).

5 is a flowchart showing a sequence of operations for performing resolution restoration in a second server according to an embodiment of the present invention.

The controller 230 of the second server 200 may receive a transmission request signal for a specific VOD file. At this time, the second server 200 can directly receive the transmission request signal for the specific VOD file from the user device 300. In this case, when the user device 300 accesses the first server 100 providing the VOD service, the user device 300 automatically receives the information on the second server 200 side, and when requesting the VOD on the first server 100 side, And can also be set to request the server 200 side.

Alternatively, when the first server 100 receives the VOD request signal from the user device, the second server 200 receives the information of the user device 300 connected to the VOD server from the first server, To the server 200 side. Accordingly, the second server 200 can receive information on the VOD file requested to be transmitted and the information of the user device 300.

Then, the controller 230 of the second server 200 may perform an operation 510 for searching for a weight corresponding to the requested VOD file.

The control unit 230 can perform an operation 515 for determining whether a corresponding weight exists in the corresponding VOD file in the DB. If the corresponding weight exists, the controller 230 transmits a corresponding weight to the user device 300 Control 520 operation. On the other hand, if it is determined that the weight corresponding to the selected VOD does not exist, the controller 230 may perform the operation 525 of assigning importance weight to the AI image learning sequence for the corresponding VOD file, and then end this operation.

The controller 230 according to the embodiment of the present invention can determine the priority of the AI image learning for the VOD file and can perform the AI image learning operation of the VOD file based on the set order information. For example, the order of AI image learning performed by the controller 230 may be performed according to a score calculation method. For the newly updated VOD images, the following number of pieces of content whose existing number of views is equal to or higher than a preset reference value The VOD file having the different information can be judged as a succeeding work of the same content), the score can be given based on the view number information of the previous VOD contents. As described in the operation 525, when there is a history of the user's request, the controller 230 may assign a weight in calculating the score for determining the order of the AI image learning.

In addition, the controller 230 can determine the VOD content having a high degree of interest by the users according to various methods, and can preferentially perform the AI video learning on the VOD contents with high interest of the users.

According to various embodiments of the present invention, the second server 200 may re-use the learned pretrained model to perform additional learning (e.g., perform additional learning in response to receiving a transmission request signal for a specific VOD file Lt; / RTI > Accordingly, the AI image learning operation according to the embodiment of the present invention can shorten the learning time.

Although not shown in the figure, on the side of the user device 300 after the above operation in the second server 200, the received first VOD file (low-resolution image) and the weight corresponding to the VOD are driven on the AI high- (Second VOD file) and reproduce it.

Although the present invention has been described in detail with reference to the above examples, those skilled in the art will be able to make adaptations, changes and modifications to these examples without departing from the scope of the present invention. In other words, in order to achieve the intended effect of the present invention, all the functional blocks shown in the drawings are separately included or all the steps shown in the drawings are not necessarily followed in the order shown, It can be in the range.

100: first server
200: second server
210:
220: DB
230:
231: resolution determination unit
232: AI image learning section
233: object image extracting unit
234: Content information collecting unit
235: User viewing information collecting section
236: User's viewing preference information calculating section
300: User device

Claims (4)

A system for providing a VOD service,
A first server for storing a VOD file and providing a transmission service of the VOD file in response to a request from a user device;
A VOD file is received from the first server, AI video learning is performed on an individual VOD file, a weight corresponding to an individual VOD file is calculated as a result of the AI image learning, and a VOD transmission request A second server for transmitting a corresponding weight to the user device in response to the requested VOD file;
A DB storing an ID of a VOD file learned by the second server and a weight calculated as a result of AI image learning of the VOD file; And
A user device for requesting transmission of an image file to the first server, receiving an image file from the first server, receiving a weight corresponding to the image file requested to be transmitted from the second server, and executing the image file to which the weight is applied; , ≪ / RTI &
The second server
Collects content information including at least one of a type, a name, a plot, and a character of content for each VOD file,
The user viewing information is calculated based on the viewing information of the user and the content information if the viewing ratio of the user is equal to or greater than a predetermined reference value, and the viewing preference information of the user is calculated based on the viewing information of the user, When the content is updated, the VOD is recommended to the user,
The AI video learning is performed in preference to the new VOD file corresponding to the viewing preference information of a plurality of users at the time of AI video learning,
Calculating the viewing preference information by age group and sex on the basis of the calculated viewing preference information and assigning a priority of AI image learning to VODs determined to be preferred by a plurality of users based on the calculated viewing preference information, And a control unit configured to set a priority of AI image learning so that a preference image of a user having a high loyalty is given, and the loyalty is determined based on a connection count and a VOD viewing count for executing a VOD,
The control unit
The method comprising: acquiring an arbitrary VOD file from the first server and selectively sampling only a part of the images of the VOD file when the size of the entire image is equal to or greater than a preset level during weight extraction; Performs AI image learning on the basis of the sampled image region, designates a weight calculated from an image region including the object separately as a weight for an object image,
And an object image extracting unit for controlling the weight of the object image to be provided to the user equipment when the object center resolution restoration is selected by the user,
Wherein the object is set to include a face image of a specific person. ◈ Claim 2 is abandoned due to payment of registration fee. The method according to claim 1,
The AI image learning
And a learning operation according to a Super Resolution Convolution Neural Networks (SRCNN) technique.
delete delete

Images