JP2020112905A - Property image classification device, property image classification method, learning method and program - Google Patents

Abstract

To classify images related to real estate properties into correct categories, and to determine whether the images have appropriate quality to introduce the properties.SOLUTION: There is provided a property image classification device 100 including: an image reception unit 11 that receives images related to properties; a classifier 13 that classifies the images received by the image reception unit into predetermined categories; a discriminator 14 that discriminates qualities of the images in the predetermined categories with respect to the images classified by the classifier; and an output unit 15 that outputs the images and quality scores of the images discriminated by the discriminator.SELECTED DRAWING: Figure 2

Description

The present invention relates to a property image classification device, property image classification method, learning method and program for classifying images related to real estate properties.

Conventionally, a technique for classifying an image into a predetermined category has been proposed. For example, Patent Document 1 discloses an image classification device capable of accurately classifying various images collected from the Internet in arbitrary categories. This image classification device is provided with a plurality of image determination units having a machine learning function, and these image determination units are arranged in a tree structure (tree structure). Each image determination unit uses image data belonging to different sections. The classification of the target image is determined based on the constructed learning data.

JP, 2017-168057, A

Recently, due to the progress of image recognition technology using artificial intelligence, many technologies for classifying images into predetermined categories are known in addition to the above-described prior art. However, it may be difficult to classify images including similar subjects in a specific field into the correct categories. Further, even if the classified category is correct, whether or not the image has a quality suitable for the field or the category is another problem. For example, in the case of a website that introduces real estate properties (condominiums, apartments, detached houses, etc.), it may be difficult to correctly classify entrances and gardens, and living and dining into their respective categories. In addition, even if the image of the living room is correctly classified into the living room category, on the website that posts the image for the purpose of selling or renting a real estate property, whether the image is attractive as a living room, Whether or not it has the quality suitable for the purpose, such as its appearance and whether or not it is reflected, is an important point in posting.

The present invention has been devised in view of such circumstances, and in a technique of classifying images related to real estate for rent or sale, while classifying images in which similar subjects are photographed into correct categories, There is provided a property image classification device, property image classification method, learning method and program for determining whether or not the product has appropriate quality for introducing the property.

In order to solve the above problems, a property image classification device that classifies images related to properties for rent or sale, and an image reception unit that receives images related to properties and an image that the image reception unit has received. A classifier that classifies into a predetermined category, a classifier that determines the quality of an image in a predetermined category with respect to the images classified by the classifier, and a score of the quality of the image determined by the classifier together with the image. An object image classification device including an output unit for outputting is provided.
According to this, by providing a classifier that classifies an image into a predetermined category and a classifier that classifies the quality of the image in the predetermined category with respect to the classified image, an image in which a similar object is captured is correct. It is possible to provide a property image classification device that classifies the property into categories and determines whether or not the image has appropriate quality for introducing the property.

Further, the classifier may be characterized by being trained by a first teacher image that should be classified into a predetermined category and a second teacher image that should not be classified into a predetermined category.
According to this, a classifier is trained by using both a teacher image that should be classified into a predetermined category and a teacher image that is not so, so that an image including a confusing subject in classification can be classified into a correct category. You can

Further, the output unit further includes a response receiving unit that receives a response to what is output by the output unit, and when the output unit outputs the image and the score of the quality of the image for each category, the category classified by the user by the classifier. So that the response accepting unit associates an image whose user's evaluation is different from the category classified by the classifier with the user's evaluation and causes the classifier to learn as a first teacher image. It may be a feature.
According to this, the teacher image to be accepted from the user for the classified category, and the image whose evaluation is different from the category classified by the classifier is associated with the user's evaluation to be classified into a predetermined category. As a result of learning the classifier again as, it is possible to perform more accurate classification.

Further, when outputting the image and the quality score of the image for each category, the output unit outputs so that the user can evaluate the quality score, and the response acceptance unit determines the user's evaluation by the discriminator. The discriminator may be made to learn as the first teacher image by associating an image having a different quality with the evaluation of the user.
According to this, it is suitable to introduce the property by learning the discriminator as the first teacher image by associating the image whose user evaluation is different from the quality discriminated by the discriminator with the user evaluation. It is possible to support posting images with various qualities.

In order to solve the above problem, a property image classification method for classifying images related to properties for rent or sale, which accepts images related to properties, classifies the accepted images into a predetermined category, and classifies them. A property image classification method is provided that determines the quality of an image in a predetermined category for the generated image and outputs the score of the determined quality of the image together with the image.
According to this, by classifying the image into a predetermined category and determining the quality of the image in the predetermined category with respect to the classified image, the image in which the confusing subject is photographed is classified into the correct category, and the image is It is possible to provide a property image classification method for determining whether or not the property has an appropriate quality for introducing the property.

In order to solve the above problems, a program that can be executed in an information processing system that classifies images related to properties to be rented or sold, wherein the information processing system is an image reception unit that receives images related to the properties. A classifying unit that classifies the image received by the image receiving unit into a predetermined category; a determining unit that determines the quality of the image in the predetermined category with respect to the images classified by the classifying unit; A program that functions as an output unit that outputs the score of the quality of the determined image is provided.
According to this, by providing the classification means for classifying the image into the predetermined category and the judgment means for judging the quality of the image in the predetermined category with respect to the classified image, the image in which the confusing subject is photographed is classified into the correct category. It is possible to provide a program capable of executing, in an information processing system, a property image classification method for classifying and determining whether or not an image has an appropriate quality for introducing a property.

In order to solve the above problems, a classifier that classifies images related to properties to be rented or sold into a predetermined category and a quality of the image in the predetermined category with respect to the images classified by the classifier is determined. A first learning step of learning a classifier and a classifier with a first teacher image that should be classified into a predetermined category and a second teacher image that should not be classified into a predetermined category And a first acceptance step for accepting images related to the property, a classification step for classifying the images accepted in the first acceptance step in a first learning step into a predetermined category, and a classification step for classification. The discriminator learned in the first learning step discriminates the quality of the image in a predetermined category, and the discriminator discriminates between the images classified by the classifier and the discriminator. The output process that outputs so that the user can evaluate it, the second reception process that receives the user's evaluation of the image output in the output process, and the category that the evaluation received in the second reception process is classified in the classification process The classifier is made to learn the image as a first teacher image by associating an image that is different from the image with the user's evaluation, and an image whose evaluation received in the second receiving process differs from the quality judged in the judging process is evaluated by the user. And a second learning step of causing the discriminator to learn in association with.
According to this, in addition to the two classification learning steps, there is provided a discrimination learning step of associating an image whose user's evaluation is different from the quality discriminated by the discriminator with the user's evaluation and learning it as a first teacher image. , An efficient classifier that classifies images related to real estate for rent or sale into a predetermined category and a classifier that distinguishes the quality of the image in a predetermined category from the images classified by the classifier Can be learned.

As described above, according to the present invention, in the technique of classifying images related to real estate properties for rent or sale, images in which similar subjects are photographed are classified into correct categories, and images are introduced to the property. It is possible to provide a property image classification apparatus, property image classification method, learning method, and program for determining whether or not the property image classification device has appropriate quality.

1 is a schematic configuration diagram of a property image classification system according to a first embodiment of the present invention. FIG. 1 is a block diagram of a property image classification device according to a first embodiment of the present invention. FIG. The hardware block diagram of the property image classification device of 1st Example which concerns on this invention. The flowchart which shows the learning method of the property image classification device of 1st Example which concerns on this invention. The flowchart which shows the classification method of the property image classification device of 1st Example which concerns on this invention. The table example of the image database of the property image classification device of 1st Example which concerns on this invention. The output display example which the user of the property image classification device of the first embodiment according to the present invention outputs so that the classified images can be evaluated. The output display example output so that the user of the property image classification device of the first embodiment according to the present invention can evaluate the quality of the determined image. The schematic model figure which shows the schematic model of the neural network of the classifier of 1st Example which concerns on this invention. The schematic model figure which shows the schematic model of the neural network of the discriminator of 1st Example which concerns on this invention.

Embodiments according to the present invention will be described below with reference to the drawings.
<First embodiment>
The property image classification system 1 and property image classification device 100 in this embodiment will be described with reference to FIGS. 1 to 10. As shown in FIG. 1, the property image classification system 1 is a system including a property image classification device 100 and one or more terminals 200 that can communicate with each other by connecting to a network 300. The property image classification device 100 receives an image related to a real estate property for rent or sale from a user, classifies the image into a predetermined category, and outputs a quality score.

The network 300 is any communication line or communication network that enables communication between information processing and data between the property image classification device 100 and the terminal 200. Accordingly, the network 300 may be a wide access network (WAN), a wired network, a fiber network, a wireless network (eg, mobile or cellular network), a cellular or telecommunications network (eg, WIFI, Long Term Evolution (LTE) network), or It can be any suitable combination thereof. Network 300 may include a private network, one or more portions of a public network (eg, the Internet), or any suitable combination thereof. Note that these are merely examples, and the present invention is not limited to these.

FIG. 3 shows an example of the hardware configuration of the property image classification device 100 that is a server. The property image classification device 100 is, for example, a server computer, is accessed from one or more terminals 200, receives an image related to a real estate property, classifies it into a predetermined category, and realizes a function of outputting a quality score. The terminal 200 is an information processing device such as a personal computer or a smartphone, and is operated by the user of the property image classification system 1. The server has a function as a web server, for example, and provides the user of the terminal 200 with a service regarding property image classification through a browser.

The property image classification device 100 includes a computing device 110 (control unit 10) including a CPU (Central Processing Unit), a memory 111 such as a ROM (Read Only Memory) and a RAM (Random Access Memory), and an external device such as a hard disk and a storage. A storage unit 20 including a storage device 120, a communication device 130 (communication unit 30) having a network interface connected to the network 300, an input device 150 such as a keyboard or a mouse used by a server administrator, and the like. An output device 140 such as a display or a printer for displaying various information and a non-transitory computer-readable recording medium such as a DVD (Digital Versatile Disk) or a CD-ROM (Compact Disk Read Only Memory) are driven, A medium driving device 122 that reads and writes software and programs recorded on the recording medium, and these are connected to each other via a transmission path 101 such as a system bus including an internal bus, an external bus, an expansion bus, and the like. Is.

The function of the control unit 10 is realized by the arithmetic device 110 (CPU), and the control unit 10 comprehensively controls the property image classification device 100. The control unit 10 has an internal memory for storing a control program such as an OS (Operating System), a program that defines various processing procedures, and required data, and various programs based on these stored programs. Information processing of.

The storage unit 20, whose function is realized by the memory 111 and the external storage device 120, stores various databases, tables, files, and the like. A computer program for giving instructions to the control unit 10 to perform various processes in cooperation with the OS is recorded in the storage unit 20. These computer programs are various programs used to provide websites and programs that realize functions provided as services of the property image classification service.

For example, the processing functions of the property image classification apparatus 100, particularly the respective processing functions performed by the control unit 10, are realized in whole or in part by a CPU and a program interpreted and executed by the CPU. Alternatively, it may be realized as hardware by wired logic. Note that the program may be recorded in a non-transitory computer-readable recording medium including programmed instructions for causing the CPU to execute the processing described in the present embodiment, and the medium driving device as necessary. It is read by the property image classification device 100 via 122. The recording medium includes any "portable physical medium" such as a DVD. Therefore, the recording medium that stores the program for executing the processing or the processing method described in the present specification also constitutes the present invention. This computer program is executed by being loaded in the RAM, and constitutes the control unit 10 in cooperation with the CPU.

In addition, the program may be stored in an application program server connected to the property image classification apparatus 100 via an arbitrary network, and it is possible to download all or part of the program as necessary. .. The "program" is a data processing method described in any language or description method, and may be in any format such as source code or binary code. Note that the "program" is not necessarily limited to a single configuration, and it achieves its function in cooperation with a plurality of modules or libraries that are distributed and configured, or with a separate program represented by an OS. Including things. Note that known configurations and procedures can be used for a specific configuration and a reading procedure for reading the recording medium in each of the above-described devices, an installation procedure after the reading, and the like.

Further, the property image classification device 100 may be configured as an information processing device such as a known personal computer or workstation, or may be configured as the information processing device to which any peripheral device is connected. Further, the property image classification device 100 may be realized by mounting software (including a program, data, etc.) that realizes the processing as described above. The specific form of the property image classification device 100 is not limited to that shown in the figure, and all or part of the property image classification device 100 may be functionally or physically dispersed/integrated in arbitrary units according to the load of the function. it can.

The communication device 130 receives data or the like from the terminal 200 via the network 300 and sends the data or the like to the arithmetic device 110, and transmits the data or the like generated by the arithmetic device 110 to the terminal 200 via the network 300. The communication device 130 is a network interface for connecting to the terminal 200 via the network 300, and more specifically, is a communication device such as a LAN (Local Area Network) card.

The input device 150 and the output device 140 are controlled by the arithmetic device 110 via an input/output interface. The arithmetic device 110 acquires data and the like from the input device 150 and outputs the generated data and the like to the output device 140. The input device 150 is typically a device such as a keyboard, a mouse, and a touch panel that is directly locally connected to the server, but is a device that remotely accesses via the network 300 to operate the property image classification device 100. May be. Similarly, the output device 140 is typically a device directly connected locally to a server such as a display or a printer, but may be a device remotely connected via the network 300.

In this specification, “outputting” includes outputting by the arithmetic device 110 to a device other than the arithmetic device 110. For example, “to output” means that the arithmetic device 110 (control unit 10) outputs to the network 300 to which the terminal 200 is connected via the transmission path 101 and the communication device 130, and the arithmetic device 110 (control unit 10). Output to a remote device via the transmission line 101 and the output device 140, and the arithmetic unit 110 (control unit 10) loads the external storage device 120 or a writable recording medium via the transmission line 101. Output to the recorded medium driving device 122.

The function of the property image classification device 100 will be described with reference to FIG. The property image classification device 100 includes a communication unit 30 that communicates with the terminal 200 via the network 300, a control unit 10 that exchanges images with the terminal 200 via the communication unit 30 and a process related to the image, and a control unit 10. And a storage unit 20 that stores an image to be processed. The control unit 10 receives an image related to a property from the terminal 200 via the communication unit 30, an image reception unit 11, a classifier 13 that classifies the image received by the image reception unit 11 into a predetermined category, and a classifier 13. The classifier 14 that determines the quality of the image in a predetermined category with respect to the image classified by the above, and the image of the property, together with the score of the quality of the image that the classifier 14 has determined through the communication unit 30, the terminal 200 The output unit 15 that outputs the output to the device and the response reception unit 12 that receives a response from the terminal 200 via the communication unit 30 to the output of the output unit 15. The storage unit 20 includes an external storage device 120, and includes an image storage unit 21 that stores the image received by the image reception unit 11.

The image reception unit 11 receives an image related to a real estate property for rent or sale. An image related to a real estate property is an image of a photo related to a real estate property such as an apartment for rent, an apartment for sale or a detached house, and in the indoor photo of the real estate property, for example, a living room or a dining room. , Living/dining room, entrance, veranda, closet, Japanese-style room, closet, bath, toilet, interior, storage facility, water supply facility, common space such as lobby, etc.In outdoor photos, for example, garage, garden, gate, Exterior, planting, building appearance, nearby facilities and parks. It is important that these photographic images are good-looking photos that are attractive to rent or sell. The image reception unit 11 may receive an image uploaded from the terminal 200 operated by the user via the communication unit 30, and directly input the image to the classifier 13 described later, or the image storage unit 21 of the storage unit 20. May be stored.

The classifier 13 classifies the photographic image received by the image receiving unit 11 or the photographic image stored in the image storage unit 21 into a predetermined category by performing learning by machine learning. Various machine learning methods can be applied to the classification method of the classifier 13, but an example in which deep learning is applied to a machine learning algorithm will be described below with reference to FIG. 9, for example. The classifier 13 decomposes the two-dimensional image data into color information (color tone and gradation) for each pixel, inputs the image to the input layer corresponding to each pixel, and then determines what the image is in the output layer. It is designed to classify and output the captured images.

In the example shown in the figure, when a photo image of a real estate property is input to the input layer, the output layer displays 8 types of "living room", "dining room", "Japanese room", "entrance", "closet", "bath", and "toilet". We classify two categories (objects and places) and show the probability of falling into each category. In the example of this figure, this image is determined to have a probability of being a living image of 81% and a probability of having a dining image of 22%, and is therefore generally classified as a living image.

The classification processing in the classifier 13 is based on the learning content in the learning phase of the neural network of the classifier 13, and the neural network of the classifier 13 correlates the photo image of the real estate property and the category shown in the image. We are learning the feature quantity that represents. In the learning phase of the classifier 13, the multi-layer neural network designed as described above is implemented by software on the object image classifying apparatus 100, and then the learning data sets of the many classifiers 13 stored in the storage unit 20 are stored. Is used to make the classifier 13 learn by so-called supervised learning. The learning data set of the classifier 13 associates the photographed images of each category (in this example, the photographed images of each of the eight categories) with the categories photographed in each image. Create as one learning data set. Then, for each category, a large number of such learning data sets are created with various photographic images.

In the learning phase of the classifier 13 in the present embodiment, the input image and the output of the multi-layer neural network of the classifier 13 are used by using the teacher data of the combination of the photograph image of the real estate property as the input data and the output data indicating the classification. The weighting coefficient of each link connecting each node is adjusted so that the relationship between the two is established in the intermediate layer between layers. The adjustment method is not particularly limited. For example, processing accuracy may be improved by using various known learning methods such as back propagation, automatic encoder, and Boltzmann machine. The algorithm for classifying the photographic images related to the real estate property in the classifier 13 may be, for example, another algorithm using a support vector machine, a Bayesian network, or the like other than the illustrated multilayer neural network. Even in that case, the basic configuration of classifying and outputting the categories captured from the input photographic image is the same.

The classifier 13 is preferably trained by the first teacher image that should be classified into a predetermined category and the second teacher image that should not be classified into the predetermined category. For example, a photograph image (first teacher image) that should be classified as “dining” is learned as a correct answer in association with the “dining” category, and a photograph image that should not be classified as “dining”, for example, “living” (Second teacher image) is learned as an incorrect answer in association with the "dining" category. In this way, by learning the classifier 13 using both the teacher images that should be classified into a predetermined category and the teacher images that are not so, it is possible to classify an image including a subject that is confusing in classification into the correct category. it can.

Further, the classifier 13 has almost the same number of photographic images that should be classified as "dining" (first teacher image) and that which should not be classified as "dining", for example, "living" photo images (second teacher image). Is preferably used for learning. That is, it is preferable that the classifier 13 be trained using approximately the same number of correct answer data sets and incorrect answer data sets. In this way, more accurate classification can be performed by learning the classifier 13 with the teacher images that should be classified into almost the same number of predetermined categories and the teacher images that are not.

As described above, in the property image classifying apparatus 100, the classifier 13 should correctly classify the photographic image related to the real estate property into which category based on the learning content in the machine learning process (deep learning and other machine learning). It can be classified as an image. By providing the classifier 13 that classifies such an image into a predetermined category, the property image classifying apparatus 100 that classifies into a correct category an image showing a subject that is similar but should be classified into a different category is provided. be able to.

The discriminator 14 discriminates the image quality in a predetermined category from the images classified by the classifier 13. For example, the discriminator 14 classifies an image in which the living room is a subject into the “living” category by the classifier 13 as an image in the “living” category. The rank, the C rank, and the like are determined. It should be noted that the A rank, B rank, and C rank respectively show the quality that is most suitable, suitable, and not suitable when used as an image for introducing real estate properties to be sold or rented. On websites that post images for the purpose of selling or renting real estate properties, whether the image is attractive as a living room, whether it looks good, whether it is reflected, etc., and whether it has a quality suitable for that purpose Whether or not it becomes an important point in posting. The discriminator 14 plays an important role in this respect.

Similar to the classifier 13, the classifier 14 classifies the photographic image classified by the classifier 13 or the photographic image once stored in the image storage unit 21 by performing learning by machine learning. As the classification method of the classifier 14, various machine learning methods can be applied, like the classifier 13. However, an example in which deep learning is applied to a machine learning algorithm will be described below with reference to FIG. 10. In addition, in order to avoid redundant description, description of the same points as the classifier 13 will be omitted. The discriminator 14 decomposes the two-dimensional image data into color information for each pixel and inputs the image to the input layer corresponding to each pixel, and when the image is input to the output layer, the quality of the image is suitable for the purpose of sale or the like. It is designed to determine whether or not to output and output.

In the illustrated example, when a photo image of a real estate property is input to the input layer, the output layer determines three ranks (quality) of “A rank”, “B rank”, and “C rank” in each category. , Probability corresponding to each discrimination result is shown. In the example of this figure, this image is classified as an image of “living room”, but as the rank (quality) of this image in the “living room” category, the probability of being “A rank” is 80%, and “B rank”. It was determined that the probability of being “C” was 19% and the probability of being “C rank” was 1%. Therefore, this image is generally judged to be an A-rank image in the “living room” category, that is, an image of good quality that matches the purpose.

The discrimination processing in the discriminator 14 is based on the learning content in the learning phase of the neural network of the discriminator 14, and the neural network of the discriminator 14 determines the correlation between the photographic image of the real estate property and the quality of the category of the image. We are learning the feature quantity to represent. In the learning phase of the discriminator 14, similar to the classifier 13, the discriminator 14 is trained by so-called supervised learning using the learning data set. The learning data set of the discriminator 14 corresponds to one learning data by associating the photographic images (three photographic images of three respective qualities in this example) showing each category with the quality of each image. Create as a set.

In the learning phase of the discriminator 14 in the present embodiment, the input and output of the multi-layer neural network of the discriminator 14 are used by using the teacher data in combination with the photographic image of the real estate property as the input data and the output data indicating the quality thereof. The weighting coefficient of each link connecting each node is adjusted so that the relationship between the two is established in the intermediate layer between layers. The adjustment method is not particularly limited.

The output unit 15 receives the image received by the image reception unit 11, the category classified by the classifier 13 for the image, and the image quality score determined by the classifier 14 for the image via the communication unit 30 from the terminal. Output to 200. The quality score may be the rank itself discriminated by the discriminator 14 as described above (the rank with the highest probability or all ranks with probability), or the rank is quantified and the average value thereof is shown. The image quality may be any value as long as it indicates the quality/badness of the image. FIG. 8 is a display example output by the output unit 15. In this display example, in the images classified into the “living” category, the respective probabilities of “A rank”, “B rank”, and “C rank” are shown. It is shown.

As described above, by providing the classifier 13 that classifies an image into a predetermined category and the classifier 14 that classifies the quality of an image in a predetermined category with respect to the classified image, when introducing the property, the category is classified into different categories. To provide a property image classification device 100 that classifies images in which similar subjects are captured, which should be classified, into a correct category and determines whether or not the images have appropriate quality for introducing the property. You can

The property image classification apparatus 100 may further include a response receiving unit 12 that receives a response to the output from the output unit 15. In this case, the output unit 15 includes an image and a score of the quality of the image for each category. Is output so that the user can evaluate whether or not the image is correctly classified. For example, as shown in FIG. 7, the output unit 15 can display three images classified into the “living room” category and evaluate whether or not each image is really an image of the “living room” as a subject. The output is displayed. When the user clicks the send button after evaluating this output display, the user indicates that the left and center images are images to be classified into the “living room” category, and vice versa. Indicates that the image should not be classified in the "living" category.

The response accepting unit 12 causes the classifier 13 to learn as a first teacher image by associating an image whose evaluation by the user is different from the category classified by the classifier 13 with the evaluation by the user. Then, the weighting factors of the links connecting the nodes of the multilayer neural network shown in FIG. 9 are readjusted, and by repeating this, erroneous category classification is gradually reduced and correct category classification is increased. It becomes possible to go, and more accurate classification becomes possible.

Further, when outputting the image and the quality score of the image for each category, the output unit 15 outputs so that the user can evaluate the quality score. For example, as shown in FIG. 8, the output unit 15 displays three images classified into the “living” category and the probability (quality score) of each rank of the images determined by the discriminator 14 for the images. At the same time, it is output and displayed so that the user can evaluate which rank the image should belong to. When the user clicks the send button on this output display after evaluation, the user indicates that the left and center images are images to be classified into “A rank” as the result of the discriminator 14. On the contrary, the right image indicates that the discrimination result of the discriminator 14 is wrong. The response receiving unit 12 causes the discriminator 14 to learn again based on the evaluation of the user. It should be noted that the evaluation method shown in this figure is an example, and the method is not limited to this. For example, the user may simply select the output image and evaluate it as a high-quality image.

By this re-learning, the discriminator 14 learns, as a first teacher image, an image in which the user's evaluation is different from the quality discriminated by the discriminator 14 in association with the user's evaluation. Then, the weighting factor of each link connecting the nodes of the multilayer neural network shown in FIG. 10 is readjusted, and by repeating this, it becomes possible to gradually increase the correct quality discrimination for the image. , It is possible to support the posting of images with the appropriate quality to introduce the property.

The above is also a method of further learning the learned models of the classifier 13 and the discriminator 14. A method of retraining the trained models of the classifier 13 and the discriminator 14 will be described with reference to FIG. In the figure, S means a step. In S100, the control unit 10 classifies the classifier 13 using a plurality of first teacher images that should be classified into a predetermined category and a plurality of second teacher images that should not be classified into a predetermined category according to a user's instruction. The discriminator 14 is trained to create a trained model.

The image reception unit 11 of the control unit 10 receives an image of a real estate property to be classified from the user's terminal 200 via the communication unit 30 in S102. In S104, the classifier 13 classifies the images received by the image receiving unit 11 into categories such as living and dining. The discriminator 14 determines whether or not the image classified by the classifier 13 in S106 corresponds to the quality of the image in the classified category, for example, “A rank”.

As illustrated in FIG. 6, at this stage, the image storage unit 21 stores the initial classification category and the initial quality score for each received image. For example, an image with an image ID of P0001 is stored with an initial classification category of "dining" and an initial quality score of "A rank". An image with an image ID of P0002 is stored as an initial classification category of "dining" and an initial quality score of "B rank". An image with an image ID of P0003 is stored with the initial classification category of “apartment” and the initial quality score of “A rank”. An image with an image ID of P0004 is stored as an initial classification category of "townhouse" and an initial quality score of "B rank".

In step S108, the output unit 15 outputs the images classified by the classifier 13 for each category and the quality score determined by the classifier 14 to the terminal 200 via the communication unit 30 so that the user can evaluate the score. The user inputs the evaluation of the output image on the output display of the terminal 200 (for example, the output display illustrated in FIGS. 7 and 8) as illustrated in FIGS. 7 and 8. In step S110, the image receiving unit 11 receives the evaluation of the classification and quality determination of the image by the user from the user's terminal 200 via the communication unit 30, and stores the evaluation in the image storage unit 21.

At this stage, the image storage unit 21 stores the same correction classification category and correction quality score as the initial classification category and the initial quality score for the correctly classified image and the image for which the correct quality score is determined. .. On the other hand, when the evaluation by the user is different from the initial classification category and the initial quality score, the image reception unit 11 stores the category and the score input by the user in the modified classification category and the modified quality score.

For example, as illustrated in FIG. 6, the image with the image ID P0001 has an initial classification category of “dining” and an initial quality score of “A rank”. Since it was evaluated that the quality score is "B rank", "living" and "B rank" are stored as the modified classification category and the modified quality score, respectively. Similarly, an image with an image ID of P0002 had an initial classification category of "dining" and an initial quality score of "B rank", but the user may have this image as "dining", but the quality score is " Since it was evaluated as "A rank", the same "dining" as the initial classification category is stored in the modified classification category, while "A rank" is stored in the modified quality score. Further, the image having the image ID P0003 had the initial classification category of “apartment” and the initial quality score of “A rank”, but the user may classify the image with the quality score of “A rank”. Since the category is evaluated as "townhouse", the modified quality score stores "A rank" which is the same as the initial quality score, while the modified classification category stores "townhouse".

In S112, the control unit 10 associates the images having different evaluations with the category (correction classification category) evaluated by the user and causes the classifier 13 to learn again as the first teacher image. Further, the classifier 13 may be relearned as the second teacher image by associating the images having different evaluations with the initial category. Further, the images having the same evaluation may be associated with the initial category and the classifier 13 may be relearned as the first teacher image. Further, in S114, the control unit 10 associates the images having different evaluations with the quality (corrected quality score) evaluated by the user and causes the images to be learned again.

Summarizing the above, the method of further learning the learned models of the classifier 13 and the classifier 14 is as follows: the classifier 13 that classifies images related to properties for rent or sale into a predetermined category; A method of learning the classifier 14 that discriminates the quality of an image in a predetermined category with respect to the classified images, and should not be classified into the first teacher image and the predetermined category to be classified into the predetermined category. A first learning step of learning the classifier 13 and the discriminator 14 with a second teacher image that is not present, a first acceptance step of accepting an image related to the property, and an image accepted in the first acceptance step of the first learning step The classifying process performed by the classifier 13 to classify the image into a predetermined category, the classifying process performed by the classifier 14 to determine the quality of the image in the predetermined category with respect to the images classified in the classifying process, and the classifier 13 An output step of outputting the images classified into each category and the score of the quality discriminated by the discriminator 14 so that the user can evaluate, and a second reception step of receiving the user's evaluation of the image output in the output step. , The classifier 13 is learned as a first teacher image by associating an image whose evaluation received in the second receiving step is different from the category classified in the classifying step with the evaluation of the user, and the evaluation received in the second receiving step The second learning step of making the discriminator 14 learn by associating an image having a quality different from the quality discriminated in the discrimination step with the user's evaluation.

According to this, in addition to the two classification learning steps, there is a discrimination learning step of associating an image whose evaluation by the user is different from the quality discriminated by the discriminator 14 with the evaluation by the user to learn as a first teacher image. Thus, the classifier 13 that classifies images related to real estate for rent or sale into a predetermined category, and the classifier that determines the quality of the image in the predetermined category with respect to the images classified by the classifier 13. 14 can be learned efficiently.

The above is also a property image classification method for classifying images related to real estate properties for rent or sale. The property image classification method will be described with reference to FIG. The image reception unit 11 of the control unit 10 receives an image of a real estate property to be classified from the user's terminal 200 via the communication unit 30 in S200. The classifier 13 classifies the image received by the image receiving unit 11 into a predetermined category in S202. The discriminator 14 discriminates whether or not the image classified by the classifier 13 in S204 corresponds to the quality of the image in the classified category. In S206, the output unit 15 outputs the images classified by the classifier 13 for each category and the quality score determined by the classifier 14 to the terminal 200 via the communication unit 30.

To summarize the above, the property image classification method that classifies images related to real estate for rent or sale accepts images related to properties, classifies the received images into predetermined categories, and classifies them into classified images. On the other hand, it is a method of determining the quality of an image in a predetermined category and outputting a score of the determined quality of the image together with the image. In this way, by classifying images into predetermined categories and determining the quality of images in the predetermined categories with respect to the classified images, it is possible to classify images with confusing subjects into the correct categories and It is possible to provide a property image classification method for determining whether or not the property image has an appropriate quality for introducing.

A program for causing the computer to execute the above-described property image classification method is executed by being loaded in the RAM, and configures the control unit 10 in cooperation with the CPU. This program is a program that can be executed by the property image classification system 1 (property image classification device 100) that classifies images related to real estate properties to be rented or sold. An image receiving unit that receives the image; a classifying unit that classifies the image received by the image receiving unit into a predetermined category; a determining unit that determines the image quality of the image classified by the classifying unit in the predetermined category; At the same time, the program functions as an output unit that outputs the quality score of the image determined by the determination unit.

As described above, by providing the classification means for classifying the image into the predetermined category and the judgment means for judging the quality of the image in the predetermined category with respect to the classified image, the image in which the confusing subject is photographed is classified into the correct category. In addition, it is possible to provide a program capable of executing a property image classification method for determining whether or not an image has an appropriate quality for introducing a property in an information processing system.

It should be noted that the present invention is not limited to the illustrated embodiments, and can be carried out with a configuration within a range that does not deviate from the contents described in each item of the claims. That is, the present invention is mainly illustrated and described mainly with respect to specific embodiments, but without departing from the scope of the technical idea and object of the present invention, with respect to the embodiments described above, the quantity, Those skilled in the art can make various modifications to other detailed configurations.

1 Property Image Classification System 100 Property Image Classification Device 10 Control Unit 11 Image Reception Unit 12 Response Reception Unit 13 Classifier 14 Discriminator 15 Output Unit 20 Storage Unit 21 Image Storage Unit 200 Terminal 300 Network

Claims (7)

A property image classification device for classifying images related to properties for rent or sale,
An image reception unit that receives images related to the property,
A classifier for classifying the image received by the image receiving unit into a predetermined category;
For the images classified by the classifier, a classifier that determines the quality of the image in a predetermined category,
Along with the image, an output unit that outputs the score of the quality of the image determined by the discriminator,
Property image classification device equipped with. The property image according to claim 1, wherein the classifier is trained by a first teacher image that should be classified into a predetermined category and a second teacher image that should not be classified into a predetermined category. Classifier. Further comprising a response receiving unit that receives a response to the output from the output unit,
When outputting the image and the score of the quality of the image for each category, the output unit outputs so that the user can evaluate the category classified by the classifier,
The response receiving unit causes the classifier to learn as a first teacher image by associating an image whose user evaluation is different from the category classified by the classifier with the user evaluation. Property image classification device described in 2. When outputting the image and the score of the quality of the image for each category, the output unit outputs so that the user can evaluate the score of the quality,
The response acceptance unit causes the discriminator to learn an image in which a user's evaluation is different from the quality discriminated by the discriminator as a first teacher image in association with the user's evaluation. Property image classification device described in 3. A property image classification method for classifying images related to properties for rent or sale,
Accept images related to the property,
Classify the received images into predetermined categories,
For the classified images, determine the image quality in a predetermined category,
Output the quality score of the determined image together with the image,
Property image classification method. A program executable in an information processing system for classifying images related to properties for rent or sale,
The information processing system,
Image receiving means for receiving images related to the property,
Classifying means for classifying the image received by the image receiving means into a predetermined category,
Discriminating means for discriminating the image quality in a predetermined category with respect to the images classified by the classifying means,
Output means for outputting a score of the quality of the image discriminated by the discriminating means together with the image,
A program to function as. A method of learning a classifier that classifies images related to properties for rent or sale into a predetermined category, and a classifier that determines the quality of an image in a predetermined category for the images classified by the classifier. There
A first learning step of learning the classifier and the discriminator by a first teacher image to be classified into a predetermined category and a second teacher image that should not be classified into a predetermined category;
A first reception process that receives images related to the property,
A classifying step of classifying the image received in the first receiving step into a predetermined category by the classifier learned in the first learning step;
A discriminating step in which the discriminator learned in the first learning step discriminates the quality of the image in a predetermined category for the images classified in the classification step;
An output step of outputting so that the user can evaluate the image classified by the classifier for each category and the quality judged by the classifier,
A second receiving step for receiving a user's evaluation of the image output in the output step,
An image in which the evaluation received in the second receiving step is different from the category classified in the classifying step is associated with the user's evaluation to cause the classifier to learn as a first teacher image, and is received in the second receiving step. A second learning step in which the discriminator is made to learn by associating an image whose evaluation was different from the quality discriminated in the discrimination step with the user's evaluation,
Including the method.