JP5046159B2 - MATERIAL SEARCH METHOD, PROGRAM FOR EXECUTING MATERIAL SEARCH METHOD, AND RECORDING MEDIUM CONTAINING THE PROGRAM - Google Patents

Description

  The present invention relates to a material search method for searching for materials belonging to different types of media information.

  In search engines on the Internet, the search target material is mainly text. Further, when searching for an image with a search engine, text such as an article uploaded together with the image is searched, and the image itself cannot be used. There is also a method for indexing and searching for images, but the indexing must be performed manually, and the contents of the index have been affected by the subjectivity of the indexer.

Therefore, a technique for predicting an optimal media pair between different types of media has been disclosed (see, for example, Patent Document 1). The technology disclosed in Patent Document 1 predicts an optimal media pair using a fitness rule that defines the correlation between objective information and sensitivity information of different media pairs and the fitness.
Japanese Patent Laid-Open No. 11-66028

  However, the technique disclosed in Patent Document 1 defines a correlation with a decision tree and needs to be a discrete value (qualitative). Further, the technique disclosed in Patent Document 1 has a problem that the correlation between the sensitivity information of different media pairs cannot be directly analyzed.

  An object of the present invention is to provide a material search method that is highly accurate in search and that can statistically analyze an expression term representing a material.

  In order to solve the above-mentioned problems, a material search method according to the present invention is characterized in that highly consistent expression terms are selected by a canonical correlation analysis method, and a material corresponding to the selected expression terms is obtained.

  Specifically, the material search method according to the present invention includes a plurality of materials for each media information type, a material physical quantity obtained by quantifying the material, and material information having an expression term representing the material. A material search method for searching for the material belonging to a material acquisition step for acquiring a search material, a material physical quantity extraction step for extracting the material physical quantity of the search material acquired in the material acquisition step, The material information that belongs to the same media information type as the search material acquired in the material acquisition step and has the material physical quantity highly correlated with the material physical quantity extracted in the material physical quantity extraction step is subjected to multiple regression from the material information. An expression term acquisition step for selecting the analysis term and acquiring the expression term of the selected material, and acquiring the expression term The expression term that is highly harmonized with the expression term acquired in step is selected by the canonical correlation analysis method from the material information belonging to a different media information type from the search material acquired in the material acquisition step, A search step of acquiring the material corresponding to the selected expression term and making it a search result.

  The material retrieval method uses a quantitative canonical correlation analysis method, and the expression terms have no correlation and high independence, so that the retrieval accuracy can be increased. Further, the material search method can statistically analyze the expression terms by a canonical correlation analysis method, and the important expression terms and the unimportant expression terms in the material belonging to different media information types Can be distinguished.

  In order to solve the above-mentioned problems, a material search method according to the present invention searches for an expression term from harmony information having a highly consistent expression term selected by the canonical correlation analysis method, and selects a material corresponding to the searched expression term. It is characterized by acquiring.

  Specifically, the material search method according to the present invention includes a plurality of materials for each media information type, material information having a material physical quantity obtained by quantifying the material and an expression term representing the material, and a canonical correlation analysis method. The material search method for acquiring the material belonging to the different media information type from the harmony information having the expression term that is highly harmonized between the materials belonging to the different media information types selected by Material acquisition step for acquiring a material for use, a material physical quantity extraction step for extracting the material physical quantity of the search material acquired in the material acquisition step, and media information of the same type as the search material acquired in the material acquisition step The material having the material physical quantity that belongs to the type and has a high correlation with the material physical quantity extracted in the material physical quantity extraction step. The material terms information is selected by multiple regression analysis method, the expression term acquisition step for acquiring the expression terms possessed by the selected material, and the expression terms that are highly harmonized with the expression terms acquired in the expression term acquisition step. A search step of searching from information and acquiring the material corresponding to the searched expression term to obtain a search result.

  The material retrieval method uses a quantitative canonical correlation analysis method, and the expression terms have no correlation and high independence, so that the retrieval accuracy can be increased. Further, the material search method can statistically analyze the expression terms by a canonical correlation analysis method, and the important expression terms and the unimportant expression terms in the material belonging to different media information types Can be distinguished.

  In the material search method according to the present invention, before the material acquisition step, the learning step of acquiring the expression term corresponding to the material, the material physical quantity and the material learned in advance and registering it as the material information is further included. It is preferable to provide.

  In the material search method described above, by registering the learned expression terms in the material information, the search accuracy can be further increased.

  In the material search method according to the present invention, the material information includes, as the material, two-dimensional information in which the material physical quantity is represented by at least shape and color, and three-dimensional information in which the material physical quantity is represented by at least shape and color. It is preferable to contain.

  The material search method described above can search for the material that can be recognized by human vision such as two-dimensional information and three-dimensional information.

  In the material search method according to the present invention, an image obtained by combining the material acquired in the material acquisition step and the material acquired in the search step is displayed on an output unit, and a combined image display step and the combined image display step. The correlation acquisition step for acquiring harmony or incongruity of the displayed image, and when the harmony is acquired in the correlation acquisition step, it is common from the material physical quantity of the material acquired in the material acquisition step and the material acquired in the search step. The search includes: a correlation analysis step of extracting a material physical quantity to be extracted, acquiring an expression term representing the extracted material physical quantity; and an expression term learning step of adding the expression term acquired in the correlation analysis step to the material information Preferably further after the step.

  According to the present invention, since a plurality of materials having different media information types can be synthesized and displayed, it is possible to acquire a properly determined harmony or inconsistency. Further, since the search service has a learning function, it is possible to learn an expression term that is a basis for determining “harmonic”.

  The material search method according to the present invention includes a program for causing a computer to execute one of the material search methods.

  The material search method according to the present invention includes a computer-readable recording medium in which a program for causing a computer to execute any one of the material search methods is recorded.

  The present invention can provide a material search method with high search accuracy and capable of statistically analyzing an expression term representing a material.

  Embodiments of the present invention will be described with reference to the accompanying drawings. The embodiment described below is an example of the configuration of the present invention, and the present invention is not limited to the following embodiment. The same step and the same device are denoted by the same reference numerals.

(First embodiment)
FIG. 1 shows an example of a material search system that executes the material search method according to the first embodiment. The material search system 100 includes, for example, a computing device 110, a material information database 120, a material acquisition unit 130, an output unit 140, and a learning support unit 150. The computing device 110 executes, for example, a material search method that will be described later. The material information database 120 is, for example, a database that stores material information (not shown) described later. Examples of the material acquisition unit 130 include a scanner that acquires two-dimensional information and three-dimensional information, a microphone and a recorder that acquire auditory information, a tactile sensor that acquires tactile information, an odor sensor that acquires odor information, and taste information. There is a taste sensor to acquire. The output unit 140 is, for example, a printer or a display that outputs search results. The learning support means 150 supports, for example, psychological experiments necessary for learning expression terms.

  FIG. 2 shows a first flowchart of the material search method according to the first embodiment. The material search method according to the first embodiment searches for materials belonging to different media information types from a plurality of materials for each media information type, material physical quantities obtained by quantifying the materials, and material information having expression terms expressing the materials. The material retrieval method is a material acquisition step S110 for acquiring a search material, a material physical quantity extraction step S120 for extracting a material physical amount of a search material acquired in the material acquisition step S110, and acquired in a material acquisition step S110. A material that belongs to the same media information type as the search material and has a material physical quantity that is highly correlated with the material physical quantity extracted in the material physical quantity extraction step S120 is selected from the material information by the multiple regression analysis method, and the selected material has Acquired in the expression term acquisition step S130 for acquiring the expression term and the expression term acquisition step S130 An expression term that is highly harmonized with the selected expression term is selected by the canonical correlation analysis method from the material information that belongs to the media information type different from the search material acquired in the material acquisition step S110, and corresponds to the selected expression term. A search step S140 that acquires a material and uses it as a search result is sequentially provided.

  The material information will be described. The material belongs to one of the media information types of two-dimensional information, three-dimensional information, auditory information, tactile information, olfactory information, or taste information.

  Two-dimensional information is information that can be recognized two-dimensionally (two-dimensionally) by human vision.

  The three-dimensional information can be recognized three-dimensionally (three-dimensionally) by human vision.

  FIG. 3 is an image showing an example of a material, where (a) is an example of two-dimensional information, and (b) and (c) are examples of three-dimensional information. Examples of the two-dimensional information include a photograph, a picture, a video, or a computer graphic (CG) expressed in a plane. Examples of the three-dimensional information include a three-dimensional model, a three-dimensional photograph, a picture, a video, or a CG.

  Auditory information can be recognized by human hearing. As the auditory information, for example, there is music.

  Tactile information is information that can be recognized by human tactile sense. The tactile information includes, for example, the temperature of a material such as hot, cold, warm, and soft, the hardness of the material such as soft and hard, and the flatness of the material surface such as rough and slip.

  The olfactory information is information that can be recognized by human odor. As the olfactory information, for example, there is a smell from the material.

  Taste information is information that can be recognized by human taste. Taste information includes the taste of ingredients such as sweetness, bitterness, spiciness, sourness, and deliciousness.

  In the material search method according to the first embodiment, the material information includes, as materials, two-dimensional information in which the material physical quantity is represented by at least shape and color and three-dimensional information in which the material physical quantity is represented by at least shape and color. It is preferable. The material search method according to the first embodiment can search 2D information and 3D information that can be recognized by the most developed visual sense among human senses, and is highly convenient. Here, the color includes lightness, hue, and saturation.

  FIGS. 4A and 4B are examples of extracting material physical quantities of two-dimensional information by contrast analysis. FIG. 4A is a diagram illustrating two-dimensional information, and FIG. 4B is an enlarged view of a section into which the two-dimensional information is divided. . As shown in FIG. 4A, in the contrast analysis, for example, the two-dimensional information 210 is divided into nine sections 212 to obtain features, and the material physical quantity of the two-dimensional information 210 is extracted.

  If the material is three-dimensional information, for example, the shape and color of the material are extracted by using both vertex information processing for extracting the shape of the material from the position of the vertex and image processing such as contrast analysis (not shown).

  When the material is auditory information, the material physical quantity is, for example, sound pressure, frequency, timbre, or distinction between a consonant and a dissonant. When the material is tactile information, the material physical quantity is, for example, smoothness, roughness, temperature, hardness, or elasticity. When the material is odor information, the material physical quantity is, for example, an odor index. When the material is taste information, the material physical quantity is, for example, a sugar concentration, acidity, or salt concentration.

  The expression terms are learned in advance by a psychological experiment, for example. The learning of expression terms will be described later.

  FIG. 5 shows a second flowchart of the material search method according to the first embodiment. As shown in FIG. 5, in the material search method according to the first embodiment, before the material acquisition step S <b> 110, learning that acquires the material, the material physical quantity, and the expression terms corresponding to the material learned in advance and registers it as material information Preferably, step S150 is further provided. Note that the material acquisition step S110, the material physical quantity extraction step S120, the expression term acquisition step S130, and the search step S140 in FIG. 5 are the same as those in FIG.

  In learning step S150 of FIG. 5, the material, the extracted material physical quantity, and the material term learned in advance are acquired and registered in the material information. In the material search method according to the first embodiment, the more material information is registered, the higher the search accuracy can be.

  Learning of expression terms can be performed by, for example, learning support means (not shown). For example, the learning support means displays a word pair in which words having opposite meanings are combined to the subject, and causes the subject to select one word considered to represent the material from the word pair. The learning support means acquires the learned expression term by acquiring this word. Moreover, it is good also considering the word selected many from several test subjects as an expression term among word pairs. FIG. 6 shows an example of word pairs. In FIG. 6, the word pair is, for example, a combination of a word A and a word B. For example, a word pair is a combination of “like” and “dislike” in the first line, and a combination of “bright” and “dark” in the second line.

  The expression terms may be learned by scoring, for example. With scoring, for example, a subject scores a word under the rule that a word suitable for the material has a high score and a word not suitable for the material has a low score. A plurality of subjects score a word pair, and one word having a higher average score from the word pair is used as an expression term. Or it is good also considering the word which has a certain average score or more as an expression term.

  A material searching method will be described. In the material acquisition step S110 of FIG. 5, for example, a search material is acquired by a material acquisition unit (not shown).

  The search material is a material that belongs to one of the media types and serves as a search condition in the material search method according to the first embodiment.

  In the material physical quantity extraction step S120, for example, if the search material is two-dimensional information, the material physical quantity is extracted by contrast analysis. If the search material is three-dimensional information, the vertex physical information and image processing are used in combination. To extract.

  In the expression term acquisition step S130, for example, if the search material is two-dimensional information or three-dimensional information, a material having a highly correlated shape and color is selected from the material information by multiple regression analysis. For example, the expression terms can be acquired from the selected material information as follows. When the search material and the material physical quantity of the material match, all of the expression terms of the material may be acquired. For example, when the material has the expression terms “large”, “soft”, and “light”, all of “large”, “soft”, and “light” are acquired as the expression terms of the search material. Further, for example, when the shapes of the search material and the material physical quantity of the material are different, “soft” and “light” are acquired as the expression terms of the search material, and “large” is not acquired.

  In the search step S140, an expression term that is highly harmonized with the expression term acquired in the expression term acquisition step S130 is selected by a canonical correlation analysis method. For example, a material corresponding to the expression term can be acquired from the selected expression term as follows. In the expression term acquisition step S130, when “large”, “soft”, and “light” are acquired as expression terms, a material having all of “large”, “soft”, and “light” as expression terms is registered in the material information. If so, get that material. Here, if multiple materials that have all of the terms “large”, “soft”, and “light” are registered in the material information, any material may be obtained, and all materials You may win. In addition, when a material having all of the terms “large”, “soft” and “light” is not registered, a material having any two of “large”, “soft” and “light” You may win. Furthermore, when a material cannot be acquired, a material having any one of “large”, “soft”, and “light” may be acquired.

  As described above, the material search method according to the first embodiment uses a quantitative canonical correlation analysis method, and the expression terms have no correlation and high independence, so that the search accuracy is increased. be able to. In addition, the material retrieval method according to the first embodiment can statistically analyze the expression terms by the canonical correlation analysis method, and the important expression terms and the unimportant expression terms in the materials belonging to different media information types can be obtained. Can be distinguished.

(Second Embodiment)
Regarding the material search method according to the second embodiment, the material search method according to the first embodiment will be mainly described. FIG. 7 shows an example of a material search system that executes the material search method according to the second embodiment. The material search system 101 includes, for example, a computing device 110, a material information database 120, a material acquisition unit 130, an output unit 140, a learning support unit 150, and a harmony information database 160. The harmony information database 160 is, for example, a database that stores harmony information (not shown) described later.

  FIG. 8 shows a flowchart of the material search method according to the second embodiment. The material search method according to the second embodiment is selected by a canonical correlation analysis method, a material information having a plurality of materials for each media information type, a material physical quantity obtained by quantifying the material, and an expression term representing the material, and A material search method for acquiring a material belonging to a different media information type from harmonized information having a highly harmonized expression term between materials belonging to different media information types, and acquiring a search material S110, a material physical quantity extraction step S120 for extracting the material physical quantity of the search material acquired in the material acquisition step S110, and a material physical quantity extraction that belongs to the same media information type as the search material acquired in the material acquisition step S110. A material having a material physical quantity highly correlated with the material physical quantity extracted in step S120 is converted into a multiple regression analysis method from the material information. The expression term acquisition step S130 for acquiring the expression term possessed by the selected material and the expression term having high harmony with the expression term acquired in the expression term acquisition step S130 are searched from the harmony information, and the searched expression term is obtained. A search step S141 that obtains a corresponding material and uses it as a search result is sequentially provided.

  The material acquisition step S110, the material physical quantity extraction step S120, and the expression term acquisition step S130 are the same as the material search method according to the first embodiment. Further, it is preferable that a learning step (not shown) is provided before the material acquisition step S110 as in the first embodiment.

  In search step S141, an expression term having high harmony with the expression term acquired in expression term acquisition step S130 is searched from the harmony information. Further, the method for acquiring the corresponding material from the searched expression terms is the same as the material searching method according to the first embodiment.

  The material search method according to the second embodiment uses a quantitative canonical correlation analysis method, and the expression terms have no correlation and high independence, so that the search accuracy can be increased. In addition, the material search method according to the second embodiment can statistically analyze the expression terms by the canonical correlation analysis method, and the important expression terms and the unimportant expression terms in the materials belonging to different media information types can be obtained. Can be distinguished. Furthermore, since the material search method according to the second embodiment prepares the harmony information in advance, the search process can be speeded up.

  The present invention includes a program for causing a computer to execute the material search method according to the first or second embodiment. The present invention also includes a computer-readable recording medium that records a program for causing a computer to execute the material search method according to the first or second embodiment. Here, the computer-readable recording medium is, for example, an electromagnetic or optical disk or tape.

  In order to evaluate the expression terms selected by the canonical correlation analysis, the following experiment was conducted. First, 28 sets of samples in which expression terms of two-dimensional information (landscape) and three-dimensional information (buildings) that the subject evaluated to be in harmony were combined were prepared. Here, 7-stage evaluation (1 is harmonious, 4 is neutral, and 7 is harmonized) was performed for each of the representation terms of the two-dimensional information and the three-dimensional information. The sample was also evaluated in seven stages. Then, a canonical correlation analysis was performed on samples evaluated as harmonious. FIG. 9 shows the results of selection of expression terms by canonical correlation analysis. In FIG. 9, the portion where the canonical load amount is 0.4 or more is shown in bold.

  The following was found from FIG. It was found that expression terms having a canonical load amount of 0.4 or more are grouped in both 2D information and 3D information in terms of competence and modernity. From the above, it was found that the expression terms selected by the canonical correlation analysis method are highly harmonized.

  As a premise of the material search method in the first and second embodiments, a database for material information having a plurality of materials for each media information type, material physical quantities obtained by quantifying the materials, and expression terms expressing the materials is constructed. There is a need. As an example, a case where the owner of a “room” searches for a “chair” suitable for his / her room by the material search method according to this embodiment will be described.

  “Room” and “chair” have different media information types. The media information type varies depending on the material search method. As in this embodiment, in the case of a web site for searching for furniture, furniture such as a chair and a bed and a room have different media information types. In this embodiment, the material “room” is treated as two-dimensional information and the material “chair” is treated as three-dimensional information. However, the media information type may be the material “chair”, or the material “room” may be two-dimensional information. Three-dimensional information may be used.

  In order to construct a database for material information, material physical quantities and expression terms for the material “chair” are acquired. Here, the acquisition of the material physical quantity is preferably performed from information embedded when creating the three-dimensional information of the material “chair” in order to reflect the intention of the creator of the three-dimensional information of the material “chair”. For example, the actual size and scale in the three-dimensional direction are acquired from the CAD information. Alternatively, it may be performed by extracting features of the three-dimensional information of the material “chair”. Furthermore, the material physical quantity may be acquired by learning in the material search method.

  Acquisition of an expression term acquires the image which the person who sees the three-dimensional information of material "chair" feels. The expression terms may be acquired in advance in the material information database or may be acquired by learning in the material search method according to the present embodiment.

  On the other hand, in order to construct a database for material information, material physical quantities and expression terms for the material “room” are acquired. For the material physical quantity, for example, the floor plan and size of the room are acquired by the user input of the material search method. Further, the material physical quantity can be acquired by image processing such as feature extraction from the image data of the material “room” without depending on the user input of the material search method. Acquisition of an expression term acquires the image which the person who sees the two-dimensional information of material "room" feels. For example, it is obtained by inputting an image of a room desired by the owner of the material “room”. Alternatively, a plurality of options are shown on the screen, and one or more expression terms are selected from the options. The acquisition of the expression terms is performed by preparing a table in which the expression terms for the material physical quantity are determined in advance and referring to the table.

(Embodiment 3)
FIG. 10 is a flowchart illustrating an example of the material search method according to the present embodiment. In the material search method according to this embodiment, after the search step S501 for executing the material search method described in the first and second embodiments, a synthetic space display step S502, a correlation acquisition step S503, and a correlation analysis step S504 and expression term learning step S505 are included in this order. The material search method according to the present embodiment is a program for executing a composite space display step S502, a correlation acquisition step S503, a correlation analysis step S504, and an expression term learning step S505 after step S501. There may be. The program may be stored in a computer-readable recording medium.

  In the search step S501, the material search method described in the first or second embodiment is executed. For example, when the image of the material “room” shown in FIG. 11 is input, the material “chair” in harmony with the “material“ room ”shown in FIG. 12 is displayed as a search result.

  In the composite space display step S502, the material “room” input by the user of the material search method and the extracted “chair” are combined, and an image obtained by combining both materials is displayed on the output means 140 shown in FIG. FIG. 13 is an example of an image obtained by combining a plurality of materials. For example, the material “chair” is displayed in a pseudo manner in the material “room” using ARtool1Kit. In the composition of the material “room” and “chair”, the actual placement is based on the material physical quantity stored in association with the material “room” and the material physical quantity stored in association with the material “chair”. It is preferable to be displayed in a state close to that. For example, if the color tone and gloss of the material “chair” are adjusted according to the relative relationship between the position of the light source in the room and the location of the material “chair”, the harmony with the material “room” will be close to reality. Can be confirmed. In addition, if a visual marker is projected onto the camera when each material is acquired, a state close to real space can be reproduced on the computer. Further, it is possible to easily and accurately combine images between materials of different media information types.

  In the correlation acquisition step S503, it is acquired whether or not the result of combining a plurality of materials is harmonized. For example, two buttons are provided at the end of an image obtained by combining a plurality of materials. If one button is pressed, harmony is input, and if the other button is pressed, mismatch is input. If “harmonize” is input in this step, the process proceeds to step S504. On the other hand, when “disharmonic” is input in this step, the material search method according to the present embodiment is terminated.

  In the correlation analysis step S504, if it is determined that the harmony is obtained in the correlation obtaining step S503, an expression term expressing the harmony between the materials is obtained. For example, by comparing the material physical quantities of both the material “chair” and the material “room”, a material physical quantity common to both is extracted. For example, in the case where material physical quantities indicating the smoothness of a curve or a point having a calm color tone with less gloss are common, those physical quantities are extracted. Here, the material physical quantity to be extracted is not limited to one, and may be two or more.

  Further, in the correlation analysis step S504, an expression term expressing the extracted material physical quantity is acquired. The expression terms can be estimated by using, for example, principal component analysis and multiple regression analysis of material physical quantities. The expression terms “modern”, “formal”, and “intimacy” are acquired based on the gentleness of the curve and the physical quantity of the material that indicates a point with a calm color with less gloss. Here, the expression term to be acquired may already be included in the expression term of the material “chair” or the material “room”, or may be newly added. Further, the expression terms to be acquired may be one type for one material physical quantity or one type for a plurality of material physical quantities. By acquiring one type of expression terms for a plurality of material physical quantities, more accurate expression terms can be acquired. In addition, the expression terms are acquired by inputting by a person. The acquisition of the expression terms is performed by preparing a table in which the expression terms for the material physical quantity are determined in advance and referring to the table. In this way, the expression terms (adjective pairs) are automatically estimated from the feature quantities of different media information. Thereby, the correspondence of adjectives between different media information can be automatically constructed.

  In the expression term learning step S505, when an expression term different from the existing expression term is acquired, the acquired expression term is newly added. The material to which the expression term is added may be only the same media information type, or may be for a plurality of media information types. In this embodiment, for example, the expression terms “modern” are used for the expression terms of all materials of the media information type of the 3D information to which the material “chair” belongs and all materials of the media information type of the 2D information to which the material “room” belongs. Add “formal” and “familiarity”. According to this embodiment, an expression term can be learned each time the material search method according to this embodiment is used. If the learned expression terms are stored in the material information database, the user can search for a material closer to his / her request.

  Further, in the expression term learning step S505, it is preferable to perform a correlation analysis for each material to which the expression term is added. In this case, the correlation analysis result is added to the material information as a new expression term. By performing correlation analysis of new expression terms for all materials, the search accuracy can be further improved.

  In recent years, the Internet environment has spread to ordinary households, and a search system is indispensable for extracting necessary information from a vast amount of information on the Web. When you are looking for furniture on a shopping site, you can filter by text information such as keywords, price, designer name, rough category classification, furniture color, and so on. In a conventional search engine, a search cannot be performed unless the text data is associated with the text data in some way, for example, the text data is included in a file. The same applies to the image data search service, and a search is performed based on text information associated with an image. When a search is performed based on text information, a huge amount of search results are extracted if the text information to be selected is too general. At this time, it is not easy to search for furniture that suits your taste from among these furniture.

  Conventionally, text information associated with an image has been input by human judgment. When input by human judgment, the text information associated with the image lacks objectivity. For this reason, it is difficult for the text information input by the user of the search site and the text information associated with the image to match, and it is difficult for the user of the search site to extract furniture that suits his / her preference. On the other hand, since the search site requires manpower and time to input text information associated with images, it is difficult to index and it is difficult to build a search site that can be easily used by search site users. was.

  Also, conventionally, when coordinating rooms, products such as chairs (interior) are checked again in the real space (interior shop) to determine the harmony between the real space and the product, or on the search site Judgment was made by projecting the extracted products in real space in the head.

  However, if the material search method according to the present embodiment is used, if a computer acquires a photographic image of a room, connects the Internet, and sends the room image to a search site, furniture that is harmonized with the room can be obtained on a large scale. You can search the database and extract various products that match your sensibility. In particular, to examine the correlation between the adjectives of different media: the image (expression term) felt from 2D information such as the material “room” and the image (expression term) felt from 3D information (furniture data) such as the material “chair” Thus, it is possible to search for furniture that the user wants.

  Also, the side of building an image index for furniture manufacturers, individual designers, or search site providers, etc., can reduce the time required for image search, and can introduce products to buyers widely. . As for urban design, non-architectural experts can learn from this system and take care of their eyes so that they can express their opinions on the proposals proposed by developers and designers. Will be communicated to. As an application example, it is possible to support urban design in which street furniture is searched from cityscape images.

  Also, in the past, when choosing what kind of atmosphere the room wanted, the furniture that fits that atmosphere was selected, but it was sometimes unclear whether the selected furniture really matched the room. However, in the material search method according to the present embodiment, the 2D information material and the 3D information material of different media information types can be combined and displayed, so that it is easily and accurately determined whether or not they are in harmony. Can do. Furthermore, since the search service further has a learning function, it is possible to further learn the expression term depending on whether or not it is determined as “harmonious” and to provide a search service closer to the sensitivity of the user of the search service.

  It was investigated whether it is effective to associate the expression terms of the two-dimensional information material and the three-dimensional information material. First, in order to correlate expression terms, material physical quantities and expression terms for each material are required. For this reason, the material of 2D information and the material of 3D information were acquired in the computer. There were 36 types of materials of the acquired two-dimensional information. In addition, there were 34 types of materials for the acquired three-dimensional information.

  Next, the material physical quantity and expression terms of the two-dimensional information material and the three-dimensional information material to be evaluated in the evaluation experiment were acquired. In the acquisition of expression terms, adjective evaluation was performed on nine of the men and women in their 20s for each of the acquired materials. The expression term is, for example, an adjective pair composed of two opposing adjectives. The adjective evaluation was performed in 7 stages, with 1 as the adjective pair and 7 as the other. For example, if the adjective pair is “classy-vulgar”, the adjective evaluation evaluated “classy” as 1, “vulgar” as 7, and “neutral” as 4. In this example, 26 pairs of expression terms were acquired for each of the materials of the two-dimensional information. In addition, for each of the three-dimensional information materials, 23 pairs of expression terms were acquired.

  Next, about the combination sample which combined the raw material of two-dimensional information, and the raw material of three-dimensional information, nine men and women in their 20s became examinees, and the harmony evaluation was performed. The degree of harmony was evaluated in seven stages, with 1 for “harmony”, 4 for “neutral” and 7 for “disharmonic”. The material for which the harmony evaluation was performed was performed on a total of 59 types of combination samples determined to be 4 or less by the subject among combinations of 36 types of 2D information materials and 34 types of 3D information materials. It was.

  FIG. 14 is an example of a canonical load amount in the embodiment. The canonical load amount of the combination sample in which the absolute value of the canonical load amount is 0.4 or more is shown for each expression term of the material of the two-dimensional information and each of the expression terms of the material of the three-dimensional information. When the absolute value of the canonical load amount is 0.4 or more, it is classified as having a high correlation, and it can be grouped into three groups such as “modern, formal, closeness” from the adjectives included in each, It turns out that there is a correspondence between each other. Thus, it is possible to provide a search system that incorporates individual sensibilities by weighting adjective evaluation.

  The material search method according to the present invention can be used as a search engine on the Internet, a multimedia-compatible database, and an entertainment device such as a karaoke device.

It is a figure which shows an example of the material search system which performs the material search method which concerns on 1st Embodiment. It is a 1st flowchart of the material search method which concerns on 1st Embodiment. It is an image which shows an example of a raw material, (a) is an example of two-dimensional information, (b) and (c) are examples of three-dimensional information. It is an example which extracts the material physical quantity of two-dimensional information by contrast analysis, (a) is a figure which shows two-dimensional information, (b) is an enlarged view of the section into which two-dimensional information was divided | segmented. 6 shows a second flowchart of the material search method according to the first embodiment. It is a figure which shows an example of a word pair. It is a figure which shows an example of the raw material search system which performs the raw material search method which concerns on 2nd Embodiment. It is a flowchart of the material search method which concerns on 2nd Embodiment. It is a figure which shows the selection result of the expression term by canonical correlation analysis. It is a flowchart which shows an example of the material search method which concerns on this embodiment. It is an example of a two-dimensional image of a material “room”. It is an example of a three-dimensional image of the material “chair”. It is an example of the image which synthesize | combined the several raw material. It is an example of the canonical load amount in the embodiment.

Explanation of symbols

100, 101 Material search system 110 Computing device 120 Material information database 130 Material acquisition means 140 Output means 150 Learning support means 160 Harmony information database 210 Two-dimensional information 212 Section S110 Material acquisition step S120 Material physical quantity extraction step S130 Expression term acquisition step S140, S141 Search step S150 Learning step

Claims (8)

A material search method for searching for a material belonging to a different media information type from a plurality of materials for each media information type, material information having a numerical value representing the material, and material information having an expression term representing the material,
A material acquisition step for acquiring search materials;
A material physical quantity extraction step for extracting the material physical quantity of the search material acquired in the material acquisition step;
The material information that belongs to the same media information type as the search material acquired in the material acquisition step and has the material physical quantity highly correlated with the material physical quantity extracted in the material physical quantity extraction step is overlapped from the material information. An expression term acquisition step of selecting by a regression analysis method and acquiring the expression term of the selected material;
A canonical correlation analysis method from among the material information belonging to a different media information type from the search material acquired in the material acquisition step, the expression terms having high harmony with the expression terms acquired in the expression term acquisition step A search step for obtaining the material corresponding to the selected expression term and obtaining a search result,
The material search method characterized by comprising in order. A plurality of materials for each media information type, material information having a numerical value representing the material, and material information having an expression term representing the material, selected by the canonical correlation analysis method, and belonging to different media information types A material search method for acquiring the material belonging to different media information types from the harmony information having the expression terms with high harmony between materials,
A material acquisition step for acquiring search materials;
A material physical quantity extraction step for extracting the material physical quantity of the search material acquired in the material acquisition step;
The material information that belongs to the same media information type as the search material acquired in the material acquisition step and has the material physical quantity highly correlated with the material physical quantity extracted in the material physical quantity extraction step is overlapped from the material information. An expression term acquisition step of selecting by a regression analysis method and acquiring the expression term of the selected material;
A search step for searching the expression terms having high harmony with the expression terms acquired in the expression term acquisition step from the harmony information, acquiring the material corresponding to the searched expression terms, and obtaining a search result;
The material search method characterized by comprising in order. Before the material acquisition step,
A learning step of acquiring the material, the material physical quantity, and the expression term corresponding to the material learned in advance and registering it as the material information;
The material search method according to claim 1, further comprising:   The material information includes, as the material, two-dimensional information in which the material physical quantity is represented by at least a shape and a color and three-dimensional information in which the material physical quantity is represented by at least a shape and a color. The material search method according to any one of 1 to 3.   A program for causing a computer to execute the material search method according to claim 1.   A computer-readable recording medium having recorded thereon a program for causing a computer to execute the material search method according to claim 1. A composite image display step of displaying an image obtained by combining the material acquired in the material acquisition step and the material acquired in the search step on an output unit;
A correlation acquisition step of acquiring harmony or incongruity of the image displayed in the composite image display step;
An expression term that expresses the extracted material physical quantity by extracting a common material physical quantity from the material physical quantity of the material obtained in the search step and the material obtained in the search step when harmony is obtained in the correlation obtaining step. A correlation analysis step to obtain
An expression term learning step of adding the expression term acquired in the correlation analysis step to the material information;
The material search method according to claim 1, further comprising: after the search step.   A program for causing a computer to execute the material search method according to claim 7.