KR102578169B1 - System for Providing Auto Complete Service YouTube Thumbnail - Google Patents

Abstract

The YouTube thumbnail auto-complete service system automatically completes a thumbnail image with a high click-through rate from the viewer's perspective when a YouTube video address is entered, allowing the uploaded video to achieve the highest number of views possible.
The present invention provides an easy and convenient service dedicated to creating YouTube thumbnails, and automatically completes differentiated thumbnail images, which can be expected to increase the number of views through continuous clicking by viewers entering the channel.

Description

YouTube Thumbnail Auto Complete Service System {System for Providing Auto Complete Service YouTube Thumbnail}

The present invention relates to a YouTube thumbnail auto-completion service system. More specifically, when a YouTube video address is entered, a thumbnail image with a high click-through rate is automatically completed from the viewer's perspective, allowing the uploaded video to achieve the highest number of views achievable. This is about the YouTube thumbnail auto-complete service system.

There is one channel per 529 people in Korea, and if the number of small-scale general YouTubers is included, the number of YouTube channels is estimated to be more than 10 times the size.

As of 2022, the size of the creator economy market, including YouTubers, is approximately 120.5 trillion won, and the economic growth of YouTube creators is expected to continue.

The total number of YouTube channels in Korea in 2023 was found to be approximately 1.7 million, the number of YouTubers making money in 2021 is 110,654 in 2021, 125,043 in 2022, and the number of YouTube channels with more than 10,000 subscribers is 22,056 in 2021 and 2022. There are 24,924 people per year.

When we survey YouTube viewers and YouTubers about the factors that determine clicks on YouTube videos, thumbnail videos influence the decision to watch YouTube by 86.6%.

Likewise, YouTube thumbnails are a key factor in determining the number of video views.

However, many YouTubers have the problem of not being able to create clickable thumbnails due to obvious copy that repeats content descriptions, use of design templates that are difficult to read, and miscellaneous editing that does not take the viewer's perspective into consideration.

Thumbnails are created using design tools such as Photoshop, Precanvas, and Mangoboard, but design tools have the disadvantage of requiring creative work from YouTubers. In particular, it was complicated and difficult for digitally vulnerable YouTubers, such as the elderly, to use design tools.

Currently, there is almost no customized thumbnail creation service for YouTubers in the market.

Korean Patent Registration No. 10-1990089

In order to solve this problem, the present invention provides a YouTube thumbnail auto-completion service that, when a YouTube video address is entered, automatically completes a thumbnail image with a high click-through rate from the viewer's perspective, thereby achieving the highest number of views that the uploaded video can reach. The purpose is to provide a system.

The YouTube thumbnail automatic completion service system according to the characteristics of the present invention to achieve the above object is,

A thumbnail image extraction unit that searches an external server or the Internet for a YouTube channel with a number of subscribers equal to or greater than a preset standard, and captures and extracts the thumbnail image of the searched YouTube channel as a YouTube video screenshot frame;

Classifies a plurality of YouTube video screenshot frames received from the thumbnail image extraction unit by thumbnail type, stores the YouTube video screenshot frames classified by thumbnail type in a database, and stores them in a YouTube video screenshot frame by thumbnail type. an object image extraction unit that extracts an object from the object image extraction unit, an object image analysis unit that analyzes the object detected by the object image extraction unit to generate a thumbnail background image included in the video screenshot frame, and a YouTube video screen for each thumbnail type. A thumbnail image database unit comprising a text detection unit that extracts a text area within a shot frame, and a text analysis unit that analyzes the extracted text area to generate text information included in the YouTube video screenshot frame; and

It includes a control unit that generates a final thumbnail image by combining the thumbnail background image generated by the object image analysis unit and the text information generated by the text analysis unit.

A copywriting analysis generation unit that receives a thumbnail type and text content (YouTube video keyword) from the control unit and outputs a thumbnail text (thumbnail copy) corresponding to the thumbnail type using the text content;

a background imaging analysis generation unit that receives a thumbnail type and a plurality of YouTube video screenshot frames from the control unit and outputs an optimal thumbnail background image corresponding to the thumbnail type; and

and a thumbnail combination unit that combines the thumbnail text generated by the copywriting analysis generation unit and the optimal thumbnail background image generated by the background imaging analysis generation unit to complete a final thumbnail image.

The thumbnail image database unit sets reference object information consisting of the location, size, number, and composition of objects within the YouTube video screenshot frame for each thumbnail type, and sets standard object information consisting of the location, size, number, and composition of objects within the YouTube video screenshot frame for each thumbnail type, and provides text content, size, and location within the YouTube video screenshot frame for each thumbnail type. It sets the standard text information and further includes a thumbnail type classification unit that classifies the thumbnail type corresponding to each YouTube video screenshot frame using the reference object information and the reference text information.

A video input unit that obtains a plurality of video frames when a YouTube video address (web address) is input; and

It further includes a scene change detection unit that determines a scene change of the image frame by calculating a difference value between macro blocks composed of N × N pixel size for each image frame,

The thumbnail image extractor finds the scene change point in the scene change detection unit, then captures and extracts a YouTube video screenshot frame in which the scene change occurred using a preset duration for each scene,

The YouTube video screenshot frame in which the scene change occurred is classified into a thumbnail type in the thumbnail image database unit under the control of the control unit, and the copywriting analysis generation unit outputs a thumbnail text (thumbnail copy) corresponding to the thumbnail type. The background imaging analysis generation unit outputs an optimal thumbnail background image corresponding to the thumbnail type, and the thumbnail combination unit automatically generates the final thumbnail image by combining the thumbnail text with the optimal thumbnail background image for each thumbnail type. Complete.

With the above-described configuration, the present invention has the effect of automatically completing a thumbnail image with a high click-through rate from the viewer's perspective when a YouTube video address is entered, thereby achieving the highest number of views that the uploaded video can reach.

The present invention provides an easy and convenient service dedicated to creating YouTube thumbnails, and automatically completes differentiated thumbnail images, which can be expected to increase the number of views through continuous clicking by viewers entering the channel.

Figure 1 is a diagram showing the configuration of a YouTube thumbnail automatic completion service system according to an embodiment of the present invention.
Figure 2 is a block diagram briefly showing the internal configuration of a thumbnail image database unit according to an embodiment of the present invention.
Figure 3 is a diagram showing an example of object analysis and text analysis of a YouTube thumbnail image according to an embodiment of the present invention.
Figure 4 is a diagram showing automatic completion of the final thumbnail image by combining thumbnail text with an optimal thumbnail background image according to an embodiment of the present invention.
Figure 5 is a block diagram briefly showing the internal configuration of the artificial neural network engine unit according to an embodiment of the present invention.
Figure 6 is a diagram illustrating the concept of recommending an optimal thumbnail background image according to an embodiment of the present invention.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present invention. While describing each drawing, similar reference numerals are used for similar components.

Terms such as first, second, A, B, etc. may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be named a second component, and similarly, the second component may also be named a first component without departing from the scope of the present invention. The term “and/or” includes any of a plurality of related stated items or a combination of a plurality of related stated items.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless explicitly defined in the present application, should not be interpreted in an ideal or excessively formal sense. No.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate overall understanding when describing the present invention, the same reference numerals are used for the same components in the drawings, and duplicate descriptions for the same components are omitted.

Figure 1 is a diagram showing the configuration of a YouTube thumbnail automatic completion service system according to an embodiment of the present invention, Figure 2 is a block diagram briefly showing the internal structure of a thumbnail image database unit according to an embodiment of the present invention, and Figure 3 is A diagram showing an example of object analysis and text analysis of a YouTube thumbnail image according to an embodiment of the present invention, and Figure 4 shows a diagram showing an example of automatically completing the final thumbnail image by combining the thumbnail text with an optimal thumbnail background image according to an embodiment of the present invention. It is a diagram showing the appearance, Figure 5 is a block diagram briefly showing the internal structure of the artificial neural network engine unit according to an embodiment of the present invention, and Figure 6 shows the concept of recommending the optimal thumbnail background image according to an embodiment of the present invention. It is a drawing.
The YouTube thumbnail auto-complete service system 10 according to an embodiment of the present invention includes an external server 20 and a YouTube thumbnail auto-complete service device 100.

The YouTube thumbnail automatic completion service device 100 according to an embodiment of the present invention includes a video receiving unit 110, a video input unit 111, a scene change detection unit 112, a thumbnail image extracting unit 120, and a thumbnail preference research unit 130. ), a thumbnail image preference analysis unit 140, a thumbnail image database unit 150, a control unit 160, a copywriting analysis generation unit 170, a background imaging analysis generation unit 180, and a thumbnail combination unit 190. do.

The video receiver 110 connects to the external server 20 and the Internet and provides a communication interface for receiving video information from the external server 20 and the Internet.

The thumbnail image extraction unit 120 searches an external server or the Internet for YouTube channels whose number of subscribers is greater than a preset standard number of subscribers, captures and extracts the thumbnail image of the searched YouTube channel as a YouTube video screenshot frame, and extracts the extracted YouTube video screenshot. The frame is stored in the thumbnail image database unit 150. The reason for extracting thumbnail images based on the number of subscribers is to analyze thumbnail images that people prefer.

The thumbnail preference research unit 130 classifies YouTube channels that match the type of content handled by the YouTube channel among YouTube channels with the number of subscribers exceeding a preset standard.

The thumbnail preference research unit 130 generates survey request information in a form in which the user directly inputs and selects the classified YouTube channel information and transmits it over the Internet. Here, the survey request information includes information on multiple YouTube channels, and includes an input window for selecting preference or disfavor for the YouTube channel.

The thumbnail preference research unit 130 receives and stores a response signal corresponding to the survey request information from the Internet, and cumulatively calculates and stores the number of preferred and non-preferred YouTube channels for each YouTube channel according to the response signal.

The thumbnail image preference analysis unit 140 receives the number of views of each content belonging to the classified YouTube channel information, and calculates the preference index through Equation 1 using the number of views of each content and the number of preferences and dislikes of the YouTube channel. Calculate and determine whether the calculated preference index is greater than or equal to the preset standard preference index.

The thumbnail image preference analysis unit 140 extracts a thumbnail image (YouTube video screenshot frame) of content whose preference index is greater than or equal to a preset standard preference index and transmits it to the thumbnail image database unit 150.

Here, N is the total number of YouTube channels with matching content types, i represents the order of YouTube channels with matching content types, PI is the preference index indicating preference for a specific YouTube channel, and W is It is a weight set differently depending on the type of content handled by the YouTube channel, V is the number of views for the content of a specific YouTube channel, P represents the preference number of the YouTube channel, and NP is the non-preference of the YouTube channel. Indicates the number (Non-Preference Number). Depending on the type of content, there are topics that people are interested in and topics that people are not interested in, so the weights are set differently.

The thumbnail image database unit 150 receives and stores more than 15,000 thumbnail images (YouTube video screenshot frames) from the thumbnail image extraction unit 120 or the thumbnail image preference analysis unit 140, and when saving data, the thumbnail images are The YouTube video screenshot frames classified by thumbnail type are stored in a database, and the optimal thumbnail background image and text information are extracted from the YouTube video screenshot frames.

The thumbnail image database unit 150 includes a thumbnail type classification unit 151, an object image extraction unit 152, an object image analysis unit 153, an object image storage unit 154, a text detection unit 155, and a text analysis unit ( 156) and a text information storage unit 157.

The thumbnail type classification unit 151 divides the thumbnail type into a first thumbnail that the user can conveniently request, a second thumbnail that increases the number of views of the video, a third thumbnail that effectively expresses the video content, a fourth thumbnail that the viewer wants to click, and It is divided into the 5th thumbnail, which is more differentiated than other video thumbnails of the same topic, and the 6th thumbnail, which allows users to quickly receive results. Of course, thumbnail types can be added or changed.

The thumbnail type classification unit 151 sets standard object information consisting of the location, size, number, and composition template of objects within the YouTube video screenshot frame for each thumbnail type, and sets the text content, size, and content within the YouTube video screenshot frame for each thumbnail type. Setting reference text information consisting of location.

The thumbnail type classification unit 151 classifies the thumbnail type corresponding to each YouTube video screenshot frame using reference object information and reference text information.

The object image extraction unit 152 can extract objects within YouTube video screenshot frames (thumbnail images) for each thumbnail type.

The object image extraction unit 152 divides each thumbnail image into macroblocks of N×N pixel size and converts the divided macroblocks into a black-and-white image indicating a boundary line.

The object image extractor 152 converts each thumbnail image frame into an edge image using an edge detection algorithm for a plurality of macro blocks.

The edge detection algorithm means that only the section in which the pixel value of the boundary line is found to have a larger difference than the preset reference pixel value is identified as an edge. As another embodiment, the edge detection algorithm may detect edges using the rate of change of brightness values between pixels using the Sobel operation method.

The edge detection algorithm can use any method for the thumbnail image as long as it is an edge image that can extract the boundary line of the person in the video.

The object image extraction unit 152 uses deep neural networks (DNN), convolutional deep neural networks (CNN), recurrent neural networks (RNN), and deep belief networks (DBN). Extract the feature map of the thumbnail image frame using one of the neural networks.

The object image extraction unit 152 can generate a feature map using a model that has already been trained by the learning unit based on deep learning. Deep learning is a set of machine learning algorithms that attempt high-level abstraction (summarizing key content or functions in large amounts of data or complex data) through a combination of various non-linear transformation techniques. is defined.

The object image extraction unit 152 extracts an area where an object is estimated to exist from a YouTube video screenshot frame and extracts a feature map representing features from the extracted area.

Here, an object is a concept that includes people, animals, and inanimate objects (buildings, bicycles, etc.).

The object image extraction unit 152 extracts at least one area in which the presence of an object is estimated from the thumbnail image based on the extracted feature map. Methods for extracting the region may include, for example, faster RCNN, SSD (Single Shot MultiBox Detector), YOLO (You Only Look Once), etc.

The object image extraction unit 152 detects objects using a deep learning model that segments objects in YouTube video screenshot frames.

As shown in FIG. 2, the object image analysis unit 153 analyzes the object detected by the object image extraction unit 152 and determines the object size, position, and composition template within the YouTube video screenshot frame (thumbnail image). After making the determination, the video screenshot frame including the object size, position, and composition template is stored in the object image storage unit 154 as a thumbnail background image.

The text detection unit 155 can extract a text area within a YouTube video screenshot frame (thumbnail image) for each thumbnail type.

The text detection unit 155 may detect an area (text detection area) corresponding to text within the thumbnail image stored in the thumbnail image database unit 150.

The text detection unit 155 provides an image-based text translation function and can detect text parts to be translated within the image based on character recognition technology such as OCR (Optical Character Reader).

The text detection unit 155 can designate a text portion within a YouTube video screenshot frame as a rectangular box area, and at this time, can output text box information including coordinate values on the image for the text detection area. For example, text box information may include upper left coordinates, upper right coordinates, lower left coordinates, and lower right coordinates.

As shown in FIG. 3, the text analysis unit 156 analyzes the text area extracted from the YouTube video screenshot frame, determines the text shape, size, and location, and generates text information included in the YouTube video screenshot frame. and stored in the text information storage unit 157.

The control unit 160 generates a final thumbnail image by combining the thumbnail background image stored in the object image storage unit 154 corresponding to the thumbnail type and the text information stored in the text information storage unit 157.

The control unit 160 stores the location, size, number, composition, text content, size, and position of objects in the final thumbnail image corresponding to the thumbnail type together with the final thumbnail image in the thumbnail image database unit 150.

Here, the text content represents YouTube video keywords, commercial copy, and Twitter trend topics included in the YouTube channel content.
As shown in FIG. 5, the artificial neural network engine unit 200 includes a copywriting analysis generation unit 170 and a background imaging analysis generation unit 180.

delete

The control unit 160 inputs the thumbnail type and text content (YouTube video keywords) stored in the text information storage unit 157 to the copywriting analysis generation unit 170, and uses the text content from the copywriting analysis generation unit 170. to output thumbnail text (thumbnail copy) corresponding to the thumbnail type.

The copywriting analysis generation unit 170 applies a thumbnail text generation artificial intelligence algorithm to the artificial neural network engine unit to generate thumbnail text according to YouTube video keywords, and the first artificial neural processing network 210 and the first learning unit ( 211) is further included.

The copywriting analysis generation unit 170 receives the thumbnail type and text content (YouTube video keyword) from the control unit 160, and outputs thumbnail text (thumbnail copy) corresponding to the thumbnail type using the text content.

The control unit 160 inputs the thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images) stored in the object image storage unit 154 to the background imaging analysis generation unit 180, and Outputs the optimal thumbnail background image corresponding to the thumbnail type.

The background imaging analysis generation unit 180 applies a thumbnail background image recommendation artificial intelligence algorithm to the artificial neural network engine unit to generate an optimal thumbnail background image according to the thumbnail type, and the second artificial neural processing network 220 and the second learning unit. It further includes part 221.

The background imaging analysis generation unit 180 receives a thumbnail type and a plurality of YouTube video screenshot frames from the control unit 160, and outputs an optimal thumbnail background image corresponding to the thumbnail type.

As shown in FIG. 4, the thumbnail combining unit 190 combines the thumbnail text generated by the copywriting analysis generating unit 170 and the optimal thumbnail background image generated by the background imaging analysis generating unit 180 to create a final thumbnail image. Complete .

As shown in FIG. 5, the artificial neural network engine unit 200 according to an embodiment of the present invention further includes a first learning unit 211 that trains the first artificial neural processing network 210, and a second artificial neural processing network 210. It further includes a second learning unit 221 that trains the neural processing network 220.

delete

The artificial neural network engine unit 200 can input a thumbnail type and text content (YouTube video keyword) as input, and output thumbnail text corresponding to the thumbnail type using the text content as an output.

The artificial neural network engine unit 200 can input a thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images) as input, and output an optimal thumbnail background image corresponding to the thumbnail type as an output.

The first learning unit 211 can learn through the first artificial neural processing network 210 using learning data stored in the text information storage unit 157.

The second learning unit 221 can learn through the second artificial neural processing network 220 using learning data stored in the object image storage unit 154.

The first and second learning units (211, 221) can detect feature points from the learning data, and determine connection weights applied to the first and second artificial neural processing networks (210, 220) based on the learning results of the detected feature points. It can be adjusted.

The first and second learning units 211 and 221 calculate the error by comparing the output value generated from the output layer of the neural network with the desired expected value for the learning data, and adjust the connection weight applied to the recognition model of the neural network in the direction of reducing the error. It can be adjusted.

The first learning unit 211 is composed of an input layer, a hidden layer, and an output layer, and uses text content to learn the first artificial neural processing network 210 to output thumbnail text (thumbnail copy) corresponding to the thumbnail type to the output layer. You can do it.

The second learning unit 221 consists of an input layer, a hidden layer, and an output layer, and uses a plurality of YouTube video screenshot frames (thumbnail images) to output the optimal thumbnail background image corresponding to the thumbnail type to the output layer. The processing network 220 can be trained.

The first learning unit 211 detects feature points from the thumbnail type and text content (YouTube video keywords), which are learning data received from the control unit 160, and trains the first artificial neural processing network 210 using the feature points. You can.

In other words, the first learning unit 211 can detect feature vectors from the thumbnail type and text content (YouTube video keywords) and train the first artificial neural processing network 210 using the detected feature vectors.

The second learning unit 221 detects feature points from the thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images), which are learning data received from the control unit 160, and uses the feature points to generate a second artificial neural processing network (220). ) can be learned.

In other words, the second learning unit 221 detects feature vectors from the thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images), and learns the second artificial neural processing network 220 using the detected feature vectors. You can do it.

The first and second learning units (211, 221) train the first and second artificial neural processing networks (210, 220), and when the weights of the first and second artificial neural processing networks (210, 220) are adjusted and learning is completed, , the first and second artificial neural processing networks 210 and 220 may become the artificial neural network engine unit 200.

The first learning unit 211 performs learning of the first artificial neural processing network 210 on a training set consisting of thumbnail types and text contents (YouTube video keywords) from the control unit 160, and performs deep learning for each type. Learn thumbnail text corresponding to the thumbnail type using feature values and text content.

The first learning unit 211 learns the thumbnail text (thumbnail copy) corresponding to the thumbnail type from the first artificial neural processing network 210 using text content.

The second learning unit 221 uses a plurality of YouTube video screenshot frames (thumbnail images) to learn the optimal thumbnail background image corresponding to the thumbnail type in the second artificial neural processing network 220.

The first learning unit 211 uses a training set consisting of a thumbnail type and text content (YouTube video keywords) as an input vector, and when it passes through the input layer, hidden layer, and output layer, it uses the text content to determine the thumbnail type. It is trained through supervised learning to generate the corresponding thumbnail text (thumbnail copy) as an output vector.

The second learning unit 221 uses a training set consisting of a thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images) as an input vector, and when it passes through the input layer, hidden layer, and output layer, a plurality of YouTube videos are generated. It is learned through supervised learning to use a screenshot frame (thumbnail image) to generate the optimal thumbnail background image corresponding to the thumbnail type as an output vector.

The first and second learning units 211 and 221 input the learning data stored in the object image storage unit 154 and the text information storage unit 157 into the first and second artificial neural processing networks 210 and 220 to produce correct answer data. It plays the role of adjusting the connection weights of the first and second artificial neural processing networks (210, 220) so that the loss function can be output, that is, the loss function becomes the minimum value, and the first and second artificial neural processing networks for which learning has been completed are generated through this. It becomes a neural processing network (210, 220).

When training data, test data, or new data not used for learning are input to the first and second artificial neural processing networks 210, 220 having updated connection weights, the first and second learning units (211, 221) input Layer - Hidden layer - Polling connected layer - The results output through the output layer can be obtained and output as response data.

The first learning unit 211 uses the input text content to create a deep learning-based recognition model (thumbnail copy creation artificial intelligence algorithm) through optimization based on whether the thumbnail text corresponding to the thumbnail type is predicted.

The first artificial neural processing network 210, which has completed learning, can predict the thumbnail text corresponding to the thumbnail type using the input text content using a deep learning-based recognition model.

The second learning unit 221 performs deep learning-based recognition through optimization based on the prediction result of the optimal thumbnail background image corresponding to the thumbnail type using the input thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images). Create a model (thumbnail background image recommendation artificial intelligence algorithm).

The second artificial neural processing network 220, which has completed learning, uses a deep learning-based recognition model to create an optimal thumbnail background image corresponding to the thumbnail type using the input thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images). It is predictable.

The first artificial neural processing network 210 generates a recognition model of the neural network from the first learning unit 211, receives the thumbnail type and text content (YouTube video keywords) from the first learning unit 211, and Predict thumbnail text corresponding to the thumbnail type using a recognition model.

The second artificial neural processing network 220 generates a recognition model of the neural network from the second learning unit 221, and receives a thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images) from the second learning unit 221. And, using a neural network recognition model, the optimal thumbnail background image corresponding to the thumbnail type is predicted.

The first artificial neural processing network 210 extracts feature points by quantizing the received thumbnail type and text content (YouTube video keyword), and converts the extracted feature points into feature vectors of the training data learned in the first learning unit 211. and transmits the thumbnail text corresponding to the thumbnail type to the control unit 160.

The first artificial neural processing network 210 outputs the thumbnail type and text content (YouTube video keyword), which are test data, as the result of response data (the result of the thumbnail text corresponding to the thumbnail type) using a deep learning-based recognition model. do.

The second artificial neural processing network 220 extracts feature points by quantizing the received thumbnail type and a plurality of YouTube video screenshot frames (thumbnail images), and uses the extracted feature points as learning data learned in the second learning unit 221. The optimal thumbnail background image corresponding to the thumbnail type is transmitted to the control unit 160 by comparing the feature vectors of .

The second artificial neural processing network 220 uses the thumbnail type, which is test data, and a plurality of YouTube video screenshot frames (thumbnail images) using a deep learning-based recognition model to determine the result of the response data (optimal thumbnail background corresponding to the thumbnail type). output as an image result).

The background imaging analysis generation unit 180 may add a movie still shot image to the learning data portion. Here, a movie still cut may be an image that attracts high viewer attention.

The background imaging analysis generation unit 180 can recommend the optimal thumbnail background image by developing an artificial intelligence algorithm through labeling of movie still cut images.

As shown in FIG. 6, the background imaging analysis generation unit 180 recommends the optimal thumbnail background image when inputting a large amount of YouTube video screenshot image frames, for example, while learning movie still cut images, and generates thumbnail text. Can be applied when combined with (thumbnail copy).

When a YouTube video address (web address) is input, the video input unit 111 obtains a plurality of video frames. Here, the YouTube video address refers to the address of the YouTube channel whose number of subscribers is greater than or equal to the preset standard number of subscribers.

The scene change detection unit 112 divides each image frame into macro blocks composed of N × N pixels (Pixel) in size, and converts the divided macro blocks into a black and white image that indicates a boundary line.

The scene change detection unit 112 converts each image frame into an edge image using an edge detection algorithm for a plurality of macro blocks.

The scene change detection unit 112 divides each image frame into macro blocks with a size of 16 × 16 pixels, converts the divided macro blocks into a black and white image indicating a border, and uses an edge detection algorithm. To detect a scene change, an edge image of the N-1th video frame and an edge image of the Nth video frame are generated 1/24 second ago.

The scene change detection unit 112 calculates the difference value between the macro blocks at the same position in the N-1th image frame and the N-th image frame, and when the difference value between macro blocks is smaller than the preset first threshold, the similar macro It is determined to be a block, and if the difference value between macro blocks is greater than the preset first threshold (T1), the scene is judged to be a different macro block.

The scene change detection unit 112 adds up the number of all similar macro blocks in the image frame, and if the summed number of similar macro blocks is greater than the preset second threshold T2, it determines that a scene change has occurred. Of course, the scene change detection unit 112 may configure various algorithms for determining scene change other than the above method.

The thumbnail image extraction unit 120 finds the scene change point in the scene change detection unit 112, and then captures and extracts only the YouTube video screenshot frames in which the scene change occurred using the preset duration for each scene.

The thumbnail image extraction unit 120 stores the extracted YouTube video screenshot frame in the thumbnail image database unit 150.

As another embodiment, the thumbnail image extractor 120 may separately extract YouTube video screenshot frames including 'popular scenes' from the extracted YouTube video screenshot frames. Here, a popular scene refers to a scene in a section where no users have left the YouTube channel video, that is, a scene that includes all users appearing in the video.

The object image analysis unit 153 analyzes the composition template of the YouTube video screenshot frame including popular scenes in the thumbnail image extraction unit 120, and generates the optimal composition with a square box within the YouTube video screenshot frame or displays the image on the display unit (Mido). (hour) can be displayed.

When creating a composition template, the object image analysis unit 153 may specify in advance a range (reference point) into which text information will be inserted in the created composition template.

The YouTube video screenshot frame with the scene change is classified into thumbnail types in the thumbnail image database unit 150 under the control of the control unit 160, and the copywriting analysis generation unit 170 generates thumbnail text (thumbnail text) corresponding to the thumbnail type. copy), the background imaging analysis generation unit 180 outputs an optimal thumbnail background image corresponding to the thumbnail type, and the thumbnail combination unit 190 combines the thumbnail text for each thumbnail type with the optimal thumbnail background image to create a final thumbnail. Autocomplete the image.

In other words, the YouTube thumbnail auto-complete service system 100 provides a service that automatically completes and displays a thumbnail image when a YouTube video address is entered.

Operations according to embodiments of the present specification can be implemented as a computer-readable program or code on a computer-readable recording medium. Computer-readable recording media include all types of recording devices that store data that can be read by a computer system. Additionally, computer-readable recording media can be distributed across networked computer systems so that computer-readable programs or codes can be stored and executed in a distributed manner.

When the embodiment is implemented in software, the above-described techniques may be implemented as modules (processes, functions, etc.) that perform the above-described functions. Modules are stored in memory and can be executed by a processor. Memory may be internal or external to the processor and may be connected to the processor by a variety of well-known means.

Additionally, computer-readable recording media may include hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Program instructions may include not only machine language code such as that created by a compiler, but also high-level language code that can be executed by a computer using an interpreter, etc.

Although some aspects of the invention have been described in the context of an apparatus, it may also refer to a corresponding method description, where a block or device corresponds to a method step or feature of a method step. Similarly, aspects described in the context of a method may also be represented by corresponding blocks or items or features of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device, such as a microprocessor, programmable computer, or electronic circuit, for example. In some embodiments, one or more of the most important method steps may be performed by such an apparatus.

In embodiments, a programmable logic device (e.g., a field programmable gate array) may be used to perform some or all of the functionality of the methods described herein. In embodiments, a field programmable gate array may operate in conjunction with a microprocessor to perform one of the methods described herein. In general, the methods are preferably performed by some hardware device.

Although the present invention has been described above with reference to preferred embodiments, those skilled in the art may make various modifications and changes to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. You will understand that you can do it.

100: YouTube thumbnail auto-complete service system 110: Video receiver
111: video input unit 112: scene change detection unit
120: Thumbnail image extraction unit 130: Thumbnail preference research unit
140: Thumbnail image preference analysis unit 150: Thumbnail image database unit
151: Thumbnail type classification unit 152: Object image extraction unit
153: object image analysis unit 154: object image storage unit
155: text detection unit 156: text analysis unit
157: text information storage unit 160: control unit
170: Copywriting analysis generation unit 180: Background imaging analysis generation unit
190: Thumbnail coupling part 200: Artificial neural network engine part
210: first artificial neural processing network 211: first learning unit
220: Second artificial neural processing network 221: Second learning unit

Claims (4)

A thumbnail image extraction unit that searches an external server or the Internet for a YouTube channel with a number of subscribers equal to or greater than a preset standard, and captures and extracts the thumbnail image of the searched YouTube channel as a YouTube video screenshot frame;
Classifies a plurality of YouTube video screenshot frames received from the thumbnail image extraction unit by thumbnail type, converts and stores the YouTube video screenshot frames classified by thumbnail type into a database, and stores them in a YouTube video screenshot frame for each thumbnail type. an object image extraction unit that extracts an object from the object image extraction unit, an object image analysis unit that analyzes the object detected by the object image extraction unit and generates a thumbnail background image included in the video screenshot frame, and a YouTube video screen for each thumbnail type. A thumbnail image database unit consisting of a text detection unit that extracts a text area within the shot frame and a text analysis unit that analyzes the extracted text area to generate text information included in the YouTube video screenshot frame;
A control unit that generates a final thumbnail image by combining the thumbnail background image generated by the object image analysis unit and the text information generated by the text analysis unit;
A copywriting analysis generation unit that receives a thumbnail type and text content (YouTube video keyword) from the control unit and outputs a thumbnail text (thumbnail copy) corresponding to the thumbnail type using the text content;
a background imaging analysis generation unit that receives a thumbnail type and a plurality of YouTube video screenshot frames from the control unit and outputs an optimal thumbnail background image corresponding to the thumbnail type; and
YouTube including a YouTube thumbnail auto-completion service device having a thumbnail combination unit that combines the thumbnail text generated by the copywriting analysis generation unit and the optimal thumbnail background image generated by the background imaging analysis generation unit to complete a final thumbnail image. Thumbnail auto-complete service system. delete In claim 1,
The thumbnail image database unit sets reference object information consisting of the location, size, number, and composition of objects within the YouTube video screenshot frame for each thumbnail type, and sets text content, size, and position within the YouTube video screenshot frame for each thumbnail type. A YouTube thumbnail auto-complete service system that sets reference text information consisting of and further includes a thumbnail type classification unit that classifies thumbnail types corresponding to each YouTube video screenshot frame using the reference object information and the reference text information. In claim 1,
The YouTube thumbnail auto-complete service device,
A video input unit that obtains a plurality of video frames when a YouTube video address (web address) is input; and
It further includes a scene change detection unit that determines a scene change of the image frame by calculating a difference value between macro blocks composed of N × N pixel size for each image frame,
The thumbnail image extractor finds the scene change point in the scene change detection unit, then captures and extracts a YouTube video screenshot frame in which the scene change occurred using a preset duration for each scene,
The YouTube video screenshot frame in which the scene change occurred is classified into a thumbnail type in the thumbnail image database unit under the control of the control unit, and the copywriting analysis generation unit outputs a thumbnail text (thumbnail copy) corresponding to the thumbnail type. The background imaging analysis generation unit outputs an optimal thumbnail background image corresponding to the thumbnail type, and the thumbnail combination unit automatically generates the final thumbnail image by combining the thumbnail text with the optimal thumbnail background image for each thumbnail type. YouTube thumbnail auto-complete service system.