KR20210122586A - Apparatus, method and computer program for recommending photo - Google Patents

Abstract

The present invention provides a photo recommendation device, method, and computer program for collecting feedback information on a user's photo and recommending a photo, which can reflect personal preference information by performing reinforcement learning on a policy table based on feedback information. The photo recommendation device comprises: a collection unit for collecting user reaction information and quality standard information for a plurality of photos; a score derivation unit for deriving scores for the plurality of photos based on the collected user reaction information and the collected quality reference information; an object extraction unit for extracting an object from the plurality of photos; and a recommendation unit for recommending at least one among the plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object.

Description

PHOTO RECOMMENDATION DEVICE, METHOD AND COMPUTER PROGRAM FOR RECOMMENDING PHOTO

The present invention relates to a photo recommendation apparatus, method and computer program for recommending photos.

There are several methods for recommending photos to a user. As an example, in a method of recommending a photo based on a user's photo selection pattern, data from which a user has selected a photo in the past is patterned and analyzed, and information about the color and direction of a photo preferred by the user is extracted and used to select a photo. the recommended method.

Since it is difficult to distinguish several photos using only the color information of the photo, it is difficult to recommend a photo preferred by the user. For example, it is difficult to distinguish a photograph with similar colors, such as a photograph including the sky and a photograph including the sea, and there is a limit to determining which person is preferred for a photograph including a person.

In addition, there is a photo recommendation method using face recognition information, photo-related temporal information, and spatial information. However, conventional methods have a limitation in that it is difficult to recommend by effectively reflecting the photo information preferred by each user.

US Patent Publication No. 2017-0185669 (published on June 29, 2017)

The present invention is to solve the problems of the prior art described above, by collecting user reaction information and quality criterion information for a plurality of photos to derive scores for a plurality of photos, extract objects from the plurality of photos, and score and recommending at least one photo among a plurality of photos based on the recommendation policy table learned based on the object.

An object of the present invention is to provide a photo recommendation apparatus, method, and computer program for recommending photos that can reflect personal preference information by collecting feedback information on a user's photo and performing reinforcement learning on a policy table based on the feedback information.

However, the technical problems to be achieved by the present embodiment are not limited to the technical problems described above, and other technical problems may exist.

As a means for achieving the above-described technical problem, an embodiment of the present invention provides a photo recommendation device for recommending a photo, a collecting unit collecting user reaction information and quality reference information for the plurality of photos, the collecting A score derivation unit for deriving a score for the plurality of photos based on the user reaction information and the collected quality reference information, an object extracting unit for extracting an object from the plurality of photos, and the derived score and the extracted object and a recommender for recommending at least one of the plurality of photos based on the learned recommendation policy table.

In one embodiment, the score deriving unit reaction score deriving unit for deriving a reaction score for each of the plurality of photos based on the user reaction information; and

and a quality score deriving unit for deriving a quality score for each of the plurality of photos based on the quality reference information.

In one embodiment, the collection unit further collects a reaction image of the user to the plurality of photos, and the reaction score derivation unit includes gaze time information for each of the plurality of photos and facial expression information recognized from the reaction image The reaction score may be derived based on the user reaction information.

In an embodiment, the quality score deriving unit may derive the quality score based on the quality criterion information including at least one of composition, sharpness, brightness, and color for each of the plurality of photos.

In an embodiment, the object extraction unit extracts an object corresponding to a preset class from the plurality of photos, and the preset class is at least one of a person class including a face and an object class including an object excluding the face. may include.

In an embodiment, the recommendation unit includes a bitstream generation unit for generating a bitstream for each of the plurality of photos based on the derived score and the extracted object, and the plurality of pieces based on the generated bitstream. It may include an arranging unit for arranging the photos of the matrix form, and a table management unit for managing the recommendation policy table by mapping the plurality of photos arranged in the matrix form to the recommendation policy table.

In an embodiment, the collecting unit may further collect feedback information on the recommended photo.

In an embodiment, the table manager may perform reinforcement learning on the recommendation policy table based on the collected feedback information.

In an embodiment, the recommendation unit may recommend at least one photo among the plurality of photos based on a recommendation policy table on which the reinforcement learning is performed.

Another embodiment of the present invention provides a method for recommending a photo, comprising: collecting user reaction information and quality standard information for the plurality of photos; based on the collected user reaction information and the collected quality standard information Deriving a score for the plurality of photos, extracting an object from the plurality of photos, and at least one of the plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object It may include a step of recommending a photo.

Another embodiment of the present invention is a computer program stored in a computer readable recording medium including a sequence of instructions for recommending a picture, when the computer program is executed by a computing device, a user reaction to the plurality of pictures Collect information and quality criterion information, derive scores for the plurality of photos based on the collected user reaction information and the collected quality criterion information, extract an object from the plurality of photos, and the derived score and a sequence of commands for recommending at least one photo among the plurality of photos based on a recommendation policy table learned based on the extracted object.

The above-described problem solving means are merely exemplary, and should not be construed as limiting the present invention. In addition to the exemplary embodiments described above, there may be additional embodiments described in the drawings and detailed description.

According to any one of the above-described problem solving means of the present invention, the present invention collects user reaction information and quality reference information for a plurality of photos to derive scores for a plurality of photos, extracts objects from the plurality of photos, , a score, and at least one photo among a plurality of photos may be recommended based on the learned recommendation policy table based on the object.

In addition, by collecting feedback information on a user's photo and performing reinforcement learning on a policy table based on the feedback information, a photo may be recommended by reflecting personal preference information.

In addition, a service for generating new content by using the recommended photo information may be provided.

1 is a block diagram of a photo recommendation apparatus according to an embodiment of the present invention.
2 is a diagram for explaining a method of deriving a quality score according to an embodiment of the present invention.
3 is a diagram for explaining a method of deriving a quality score according to an embodiment of the present invention.
4 is a flowchart of a photo recommendation method according to an embodiment of the present invention.
5 is a flowchart of a photo recommendation method according to an embodiment of the present invention.
6 is a diagram for explaining a method of recommending a photo according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is "connected" with another part, this includes not only the case of being "directly connected" but also the case of being "electrically connected" with another element interposed therebetween. . Also, when a part "includes" a component, it means that other components may be further included, rather than excluding other components, unless otherwise stated, and one or more other features However, it is to be understood that the existence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded in advance.

In this specification, a "part" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. In addition, one unit may be implemented using two or more hardware, and two or more units may be implemented by one hardware. Meanwhile, '~ unit' is not limited to software or hardware, and '~ unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example, '~' denotes components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, and procedures. , subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. The functions provided in the components and '~ units' may be combined into a smaller number of components and '~ units' or further separated into additional components and '~ units'. In addition, components and '~ units' may be implemented to play one or more CPUs in a device or secure multimedia card.

The "network" referred to below means a connection structure in which information can be exchanged between each node, such as terminals and servers, and a local area network (LAN), a wide area network (WAN) , the Internet (WWW: World Wide Web), wired and wireless data networks, telephone networks, wired and wireless television networks, and the like. Examples of wireless data communication networks include 3G, 4G, 5G, 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), Wi-Fi, Bluetooth communication, infrared communication, ultrasound Communication, Visible Light Communication (VLC), LiFi, etc. are included, but are not limited thereto.

Some of the operations or functions described as being performed by the terminal or device in the present specification may be instead performed by a server connected to the terminal or device. Similarly, some of the operations or functions described as being performed by the server may also be performed in a terminal or device connected to the server.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a photo recommendation apparatus according to an embodiment of the present invention. Referring to FIG. 1 , the photo recommendation apparatus 100 may include a photo storage unit 110 , a collection unit 120 , a score derivation unit 130 , an object extraction unit 140 , and a recommendation unit 150 . .

The photo recommendation apparatus 100 may recommend a photo from a plurality of photos. The photo recommendation apparatus 100 may recommend a photo by reflecting the user's preference information by selecting one or more photos from among a plurality of photos, for example. The photo recommendation apparatus 100 may be a user terminal capable of storing a plurality of photos of the user and recommending photos from the plurality of photos being stored, receiving a plurality of photos or information on the plurality of photos from the user terminal, and , it may be a server capable of recommending a photo based on the received information.

The photo storage unit 110 may store a plurality of photos. The photo storage unit 110 may receive and store a plurality of photos, for example, from a user terminal, an external server, a cloud, or an external device.

The collection unit 120 may collect user reaction information and quality reference information for a plurality of photos.

The collection unit 120 may further collect the reaction image of the user to the plurality of photos. The reaction image of the user may be, for example, a face image of the user captured using a camera attached to the user terminal while a plurality of photos are output through the display of the user terminal.

The user reaction information may be derived based on the user's reaction image to the plurality of photos. The user reaction information may include, for example, gaze time information for each of a plurality of photos and facial expression information recognized from a reaction image.

In another embodiment, the photo recommendation apparatus 100 may collect reaction images of the user to a plurality of photos, and the photo recommendation apparatus 100 may derive user reaction information from the user reaction images.

For example, the photo recommendation apparatus 100 may detect the user's face image from the user's reaction image to a plurality of photos. The photo recommendation apparatus 100 may derive facial expression information of the user included in the user's face part. The photo recommendation apparatus 100 may derive expression information of the user from the user's face image based on a probability value corresponding to any one of a plurality of expressions. The plurality of facial expressions may include, for example, one or more of an expressionless expression, a smile, a smile, an anger, and a sadness.

The quality reference information may include, for example, at least one of composition, sharpness, brightness, and color of each of the plurality of photos.

The score derivation unit 130 may derive scores for a plurality of photos based on the collected user reaction information and the collected quality reference information.

Referring back to FIG. 1 , the score deriving unit 130 may include a response score deriving unit 131 and a quality score deriving unit 132 .

The reaction score deriving unit 131 may derive a reaction score for each of the plurality of photos based on user reaction information. The user reaction information may include, for example, gaze time information for each of a plurality of photos and facial expression information recognized from a reaction image.

The response score derivation unit 131 may give a higher response score as the user gazes at the picture longer, for example, based on gaze time information for each of the plurality of pictures.

The reaction score derivation unit 131, for example, based on the expression information recognized from the reaction image for each of the plurality of photos, the user's expression is strongly expressed or the more positive expressions appear, the higher the reaction score. can be given

The quality score derivation unit 132 may derive a quality score for each of the plurality of photos based on the quality criterion information.

The quality reference information may include, for example, at least one of composition, sharpness, brightness, and color of each of the plurality of photos.

2 and 3 are diagrams for explaining a method of deriving a quality score according to an embodiment of the present invention.

For example, the quality score derivation unit 132 may derive a quality score for the picture as shown in FIG. 2A on the basis of the quality criterion information. The quality score derivation unit 132 may analyze the position of the horizontal line and the vanishing point by detecting the contour as shown in (b) of FIG. 2 . The quality score deriving unit 132 may analyze that the photo is a template of a three-part composition as shown in FIG. 2C .

For example, the quality score derivation unit 132 may derive a quality score based on whether a horizontal line and a vanishing point of the picture exist at 1/3 of the picture having a composition of three equal parts. The quality score derivation unit 132 may derive a quality score (S ₁ ) for a photograph of a composition divided into thirds by using Equation 1 below.

As another example, the quality score deriving unit 132 may derive a quality score based on the quality criterion information for the picture as shown in FIG. 3A . The quality score deriving unit 132 may detect an outline as shown in FIG. 3B . The quality score deriving unit 132 may analyze that the photo is a template of a symmetric composition as shown in FIG. 3C .

For example, the quality score deriving unit 132 may derive a quality score based on the degree of symmetry of the photo with respect to the photo having a symmetric composition. The quality score derivation unit 132 may derive a quality score based on the degree of symmetry by using the detected outline as shown in FIG. 3B . The quality score derivation unit 132 may derive a quality score (S ₂ ) for a symmetrical photograph using Equation 2 below.

The score derivation unit 130 may derive a final score for each of the plurality of photos based on the response score and the quality score.

The object extraction unit 140 may extract an object from a plurality of photos.

The object extraction unit 140 may extract an object corresponding to a preset class from a plurality of photos. The preset class may include, for example, at least one of a person class including a face and an object class including an object other than the face.

For example, the object extractor 140 may extract an object corresponding to a face class from a plurality of photos by using the face detection model. The object extraction unit 140 may further extract landmark information included in the face from the plurality of photos. The object extractor 140 may further extract feature value information for recognizing a face included in a face from a plurality of photos.

For example, the object extractor 140 may extract an object corresponding to each object class from a plurality of photos by using a plurality of object detection models. The thing class may mean, for example, one or more classes including things excluding faces such as a building class, a furniture class, and a car class.

The recommendation unit 150 may recommend at least one photo among a plurality of photos based on the recommendation policy table. The recommendation policy table may be derived based on the derived score and the extracted object.

Referring back to FIG. 1 , the recommendation unit 150 may include a bitstream generation unit 151 , an alignment unit 152 , and a table management unit 153 .

The bitstream generator 151 may generate a bitstream for each of the plurality of photos based on the derived score and the extracted object.

The bitstream generator 151 may perform quantization on, for example, a response score and a quality score. The bitstream generator 151 may, for example, perform binarization on the class ID of each class including the extracted object.

The bitstream generator 151 may generate a bitstream for each of the plurality of photos based on a result value obtained by performing quantization on the response score and the quality score and a result value obtained by performing binarization on the class ID.

The sorting unit 152 may align a plurality of photos in a matrix form based on the generated bitstream. The sorting unit 152 may sort the plurality of photos in a matrix form, for example, by arranging bitstreams in a descending order or an ascending order. In the case of simple sorting of the bitstream, it is most affected by the value located in the upper bit. Therefore, the influence may be increased depending on which score or class ID-related value in the bitstream is located in the upper bit. .

The table manager 153 may manage the recommendation policy table by mapping a plurality of photos arranged in a matrix form to the recommendation policy table.

The collection unit 120 may further collect feedback information on the recommended photo. In an embodiment, the collection unit 120 may collect feedback information generated based on whether the user has selected or rejected the recommended photo.

In another embodiment, the collection unit 120 may collect feedback information generated based on photo information selected by a user from among a plurality of photos.

The table manager 153 may perform reinforcement learning on the recommendation policy table based on the collected feedback information. The table manager 153 may, for example, update a reward based on the collected feedback information, and may perform reinforcement learning on the recommendation policy table based on the updated reward.

The recommender 150 may recommend at least one photo among a plurality of photos based on a recommendation policy table on which reinforcement learning is performed. For example, the recommender 150 may recommend a picture of a bitstream corresponding to the bitstream in a matrix in which bitstreams are arranged in an order of increasing values of a recommendation policy table on which reinforcement learning is performed.

The recommender 150 may recommend the number of photos desired by the user from among the plurality of photos, for example, based on a recommendation policy table on which reinforcement learning is performed. For example, when a plurality of bitstreams having the same value of the recommendation policy table exist, the recommendation unit 150 may select and recommend one or more of the plurality of photos based on the bitstream.

4 is a flowchart of a method for recommending a photo according to an embodiment of the present invention. The method 400 for recommending a photo performed by the photo recommending device 100 shown in FIG. 4 includes steps that are time-series processed by the photo recommending device 100 according to the embodiment shown in FIG. 1 . Accordingly, even if the description is omitted below, it is also applied to the method of recommending a photo performed by the photo recommendation apparatus 100 according to the embodiment shown in FIG. 1 .

In step S410, the photo recommendation apparatus 100 may store a plurality of photos.

In step S420 , the photo recommendation apparatus 100 may collect user reaction information and quality reference information for a plurality of photos.

In step S430 , the photo recommendation apparatus 100 may derive scores for a plurality of photos based on the collected user reaction information and the collected quality reference information.

In step S440 , the photo recommendation apparatus 100 may extract an object from a plurality of photos.

In step S450 , the photo recommendation apparatus 100 may recommend at least one photo among a plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object.

In the above description, steps S410 to S450 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be switched.

5 is a flowchart of a method for recommending a photo according to an embodiment of the present invention. The method 500 for recommending a photo performed by the photo recommending device 100 shown in FIG. 5 includes steps that are time-series processed by the photo recommending device 100 according to the embodiment shown in FIG. 1 . Accordingly, even if the description is omitted below, it is also applied to the method of recommending a photo performed by the photo recommendation apparatus 100 according to the embodiment shown in FIG. 1 .

In operation S510 , the photo recommendation apparatus 100 may generate an M×N matrix based on a plurality of photos.

The photo recommendation apparatus 100 generates an M×N matrix from M×N photos by, for example, generating a bitstream for each of a plurality of photos, and arranging the plurality of photos in a matrix form based on the generated bitstream. can create

In step S520 , the photo recommendation apparatus 100 may generate a random number to select the i-th photo.

The photo recommendation apparatus 100 may collect feedback information on a plurality of photos by generating a random number and recommending a random number before reinforcement learning is performed on the recommendation policy table.

In step S530, the photo recommendation apparatus 100 may output the selected i-th photo to the user.

In step S540 , the photo recommendation apparatus 100 may collect feedback information on the output i-th photo.

The photo recommendation apparatus 100 may collect, for example, feedback information generated based on whether the user selects or rejects the output i-th photo.

In operation S550 , the photo recommendation apparatus 100 may perform reinforcement learning on the recommendation policy table based on the collected feedback information.

The photo recommendation apparatus 100 may, for example, update a reward based on the collected feedback information, and may perform reinforcement learning on the recommendation policy table based on the updated reward.

In step S560, the photo recommendation apparatus 100 may recommend a photo based on a recommendation policy table on which reinforcement learning is performed (not shown).

In the above description, steps S510 to S560 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be switched.

A method of recommending a photo according to an embodiment of the present invention will be described with reference to FIG. 6 . For the sake of brevity, it is assumed that 2 pictures out of 4 pictures are recommended using a 2×2 matrix.

6A shows four photos newly stored in the user terminal. For example, in the case of a photo having an ID of 001, a bitstream "0101 0010 0010 0000" may be generated based on a quality score of 82 points, a response score of 32 points, and the extracted object. In the case of a photo having an ID of 002, a bitstream "0101 0101 0001 1101" may be generated based on a quality score of 85 points, a reaction score of 29 points, and the extracted object. In the case of a picture having an ID of 003, a bitstream "010 0000 0100 1110" may be generated based on a quality score of 32 points, a response score of 78 points, and the extracted object. In the case of a photo having an ID of 004, a bitstream "0100 0001 0101 1100" may be generated based on a quality score of 65 points, a response score of 92 points, and the extracted object.

6B exemplarily illustrates a current recommendation policy table. The current recommendation policy table may be, for example, a recommendation policy table in which reinforcement learning is performed based on feedback information on previously recommended photos.

For example, the photo recommendation apparatus 100 may arrange bitstreams of a plurality of photos stored in a previous branch in a matrix form, and map the bitstreams arranged in a matrix form to respective values of the recommendation policy table. The photo recommendation apparatus 100 may update a reward based on feedback information on a recommended photo among a plurality of photos stored in the previous branch, and may perform reinforcement learning on the recommendation policy table based on the updated reward. That is, the location (coordinate value) to which the bitstream of the user's preferred picture is mapped in the recommendation policy table may have a higher value.

Subsequently, the photo recommendation apparatus 100 may arrange bitstreams of a plurality of photos stored in this branch in a matrix form in the same manner. At this time, if the bitstreams of the plurality of photos of the previous branch and the plurality of photos of the current branch are arranged in a matrix form, the arrangement of the bitstreams in the matrix may exhibit the same tendency. The photo recommendation apparatus 100 may map a bitstream arranged in a matrix form to each value of a current recommendation policy table in which reinforcement learning is performed.

That is, the bitstreams of the plurality of photos of this branch and the bitstreams of the plurality of photos of the previous branch are arranged to have the same tendency. Therefore, the photo recommendation apparatus 100 maps the recommendation policy table, which has been reinforcement-learned in the previous branch, to the bitstream of a plurality of photos of the current branch, so that the photo corresponding to the bitstream corresponding to the position where the value of the recommendation policy table is high is displayed to the user. can recommend to

FIG. 6(c) shows that the bitstreams of the four pictures shown in FIG. 6(a) are arranged in a descending order so that IDs of each picture are arranged in a matrix form. As described above, based on (b) and (c) of FIG. 6 , the photo recommendation device 100 sets the ID corresponding to the position 601 where the value of the recommendation policy table is 0.8, in the order of increasing the value of the recommendation policy table. A photo 611 having a value of 004 and a photo 613 having an ID of 003 corresponding to a location 603 having a value of 0.3 in the recommendation policy table may be recommended.

6 (d) and (e) exemplarily show a recommendation policy table in which reinforcement learning is performed based on feedback information on a recommended photo. When the user selects both the recommended photos (ID 004 (611) and ID 003 (613)), the value of the recommendation policy table corresponding to the selected photo may increase as shown in (d) of FIG. 6 . For example, the value 621 of the recommendation policy table for the ID 004 (611) selected by the user is '(Value of the existing recommendation policy table + (reward value (=1) if selected - value of the existing recommendation policy table) ))/number of attempts', and specifically (0.8+(1-0.8))/1=1. As another example, the value 623 of the recommendation policy table for the ID 003 ( 613 ) selected by the user may increase to (0.3+(1-0.3))/1=1. When the user selects the photo 611 with ID 004 from among the recommended photos and rejects the photo 613 with ID 003, the value of the recommendation policy table corresponding to the selected photo increases as shown in FIG. The value of the recommendation policy table corresponding to the rejected photo may decrease. For example, the value 631 of the recommendation policy table for the ID 004 (611) selected by the user may increase to (0.8+(1-0.8))/1=1. As another example, the value 633 of the recommendation policy table for ID 003 (613) selected by the user is '(Value of the existing recommendation policy table + (reward value (= 0) when rejected) - of the existing recommendation policy table value))/number of attempts', and specifically, it can be reduced to (0.3+(0-0.3))/1=0.

The method for recommending a photo in the photo recommendation apparatus described with reference to FIGS. 1 to 6 may be implemented in the form of a computer program stored in a medium executed by a computer or a recording medium including instructions executable by the computer.

Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. Also, computer-readable media may include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention. do.

100: photo recommendation device
110: photo storage unit
120: collection unit
130: score derivation unit
140: object extraction unit
150: recommendation

Claims (19)

In the photo recommendation device for recommending a photo,
a collection unit for collecting user reaction information and quality reference information for a plurality of photos;
a score derivation unit for deriving scores for the plurality of photos based on the collected user reaction information and the collected quality reference information;
an object extraction unit for extracting objects from the plurality of photos; and
A recommendation unit for recommending at least one photo among the plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object
Which will include, a photo recommendation device.
The method of claim 1,
The score derivation unit
a reaction score deriving unit for deriving a reaction score for each of the plurality of photos based on the user reaction information; and
A quality score derivation unit for deriving a quality score for each of the plurality of photos based on the quality reference information
Which will include, a photo recommendation device.
3. The method of claim 2,
The collection unit further collects the user's reaction image to the plurality of photos,
The reaction score deriving unit will derive the reaction score based on the user reaction information including gaze time information for each of the plurality of photos and facial expression information recognized from the reaction image.

3. The method of claim 2,
Wherein the quality score derivation unit derives the quality score based on the quality criterion information including at least one of composition, sharpness, brightness, and color for each of the plurality of photos.
The method of claim 1,
The object extraction unit extracts an object corresponding to a preset class from the plurality of photos,
The preset class includes at least one of a person class including a face and an object class including an object excluding the face.
The method of claim 1,
The recommendation is
a bitstream generator for generating a bitstream for each of the plurality of photos based on the derived score and the extracted object;
an alignment unit aligning the plurality of photos in a matrix form based on the generated bitstream; and
A table management unit for managing the recommendation policy table by mapping the plurality of photos arranged in the matrix form to the recommendation policy table
Which will include, a photo recommendation device.
7. The method of claim 6,
Wherein the collecting unit further collects feedback information on the recommended photo.
8. The method of claim 7,
The table manager will perform reinforcement learning on the recommendation policy table based on the collected feedback information, a photo recommendation device.
9. The method of claim 8,
Wherein the recommendation unit recommends at least one photo among the plurality of photos based on the recommendation policy table on which the reinforcement learning is performed.
In the method of recommending a photo,
collecting user reaction information and quality reference information for a plurality of photos;
deriving scores for the plurality of photos based on the collected user reaction information and the collected quality reference information;
extracting an object from the plurality of photos; and
Recommending at least one photo among the plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object
A method for recommending photos, which includes.
11. The method of claim 10,
The step of deriving the score is
deriving a reaction score for each of the plurality of photos based on the user reaction information; and
deriving a quality score for each of the plurality of photos based on the quality reference information
A method for recommending photos, which includes.
12. The method of claim 11,
Further comprising the step of collecting the user's reaction image to the plurality of photos,
The step of deriving the reaction score is to derive the reaction score based on the user reaction information including gaze time information for each of the plurality of photos and facial expression information recognized from the reaction image.

12. The method of claim 11,
The step of deriving the quality score is to derive the quality score based on the quality criterion information including at least one of composition, sharpness, brightness, and color for each of the plurality of photos.
11. The method of claim 10,
The step of extracting the object extracts an object corresponding to a preset class from the plurality of photos,
The preset class includes at least one of a person class including a face and an object class including an object excluding the face.
11. The method of claim 10,
generating a bit stream for each of the plurality of photos based on the derived score and the extracted object;
arranging the plurality of photos in a matrix form based on the generated bitstream; and
managing the recommendation policy table by mapping the plurality of photos arranged in the matrix form to the recommendation policy table
Which will further include, a photo recommendation method.
16. The method of claim 15,
The method further comprising the step of collecting feedback information on the recommended photo, photo recommendation method.
17. The method of claim 16,
The managing of the recommendation policy table may include performing reinforcement learning on the recommendation policy table based on the collected feedback information.
18. The method of claim 17,
The recommending step is to recommend at least one photo among the plurality of photos based on a recommendation policy table on which the reinforcement learning is performed.
In a computer program stored on a computer-readable recording medium comprising a sequence of instructions for recommending a picture,
When the computer program is executed by a computing device,
Collect user reaction information and quality standard information for a plurality of photos,
Derive a score for the plurality of photos based on the collected user reaction information and the collected quality reference information,
Extracting an object from the plurality of photos,
A computer program stored in a computer-readable recording medium comprising a sequence of instructions for recommending at least one of the plurality of photos based on a recommendation policy table learned based on the derived score and the extracted object. .

Images