WO2016147260A1 - 画像検索装置、及び画像を検索する方法 - Google Patents
画像検索装置、及び画像を検索する方法 Download PDFInfo
- Publication number
- WO2016147260A1 WO2016147260A1 PCT/JP2015/057516 JP2015057516W WO2016147260A1 WO 2016147260 A1 WO2016147260 A1 WO 2016147260A1 JP 2015057516 W JP2015057516 W JP 2015057516W WO 2016147260 A1 WO2016147260 A1 WO 2016147260A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- propagation
- query
- search
- feature
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2455—Query execution
Definitions
- the present invention relates to an image search device and an image search method.
- a large-scale, high-speed similar image search system has been put into practical use, and a technique has been proposed in which image features are extracted in advance and stored in a database to quickly search for images that look similar to the query image. ing.
- Patent Document 1 states that “an image search device extracts an image database, a search request reception unit that receives a query image, and a similar image that is similar to the query image from images registered in the image database.
- Search means and search result presenting means for arranging a similar image around the query image and presenting a search result in which the query image and the similar image are linked and displayed on the display means.
- search request accepting means When a new query image is accepted by the search request accepting means, a search result based on the new query image is added to the search result while maintaining a linked display of the search results presented in the past on the display means.
- To the display means (see summary).
- the search system described in Patent Document 1 updates the search result by interaction with the user. Specifically, the search system described in Patent Literature 1 repeats a search using a new query image as an image included in a search result obtained from a query image specified by the user. Thereby, if a user specifies a query image appropriately, there is a possibility that search results rich in variations regarding the target search target can be obtained.
- the present application includes a plurality of means for solving the above-described problems.
- a registered image one or more feature amounts including the first type of the registered image, and the registered image
- An image search device that searches for an image from an image database that holds propagation information indicating a link between feature quantities of the same type, and acquires the first image and the one or more types of feature quantities of the first image. And calculating the propagation intensity indicating the similarity between the image acquisition unit registered in the image database and the feature amount of each type of the first image and the feature amount of the type of the registered image.
- a link between the feature amount of the type of the first image and the feature amount of the type of the registered image having the maximum propagation strength between the feature amount of the type of the first image An attribute propagation part to be included in the information;
- a query input unit that obtains a first type of first feature quantity from the feature quantity of the registered image held in the image database and generates an input query including the first feature quantity, and the correspondence indicated by the propagation information Based on the first feature amount, the search is performed to follow the feature amount of the registered image, and the first type feature amount of the registered image having the feature amount on the route in the search is added.
- An image search device comprising: a query reinforcement unit that generates a query; and an image search unit that searches for an image from the image database using the input query and the additional query.
- Example 1 it is a block diagram which shows the structural example of an image search system.
- Example 1 it is a block diagram which shows the hardware structural example of an image search system. It is a figure which shows the structural example of the image database in Example 1.
- FIG. In Example 1, it is a figure explaining the example of the process which produces
- Example 1, it is a figure explaining the example of the process which produces
- it is a figure explaining the example of the process which produces
- 6 is a flowchart illustrating an example of attribute propagation information generation processing according to the first exemplary embodiment. In Example 1, it is a figure explaining the example of the image search using attribute propagation information.
- Example 10 is a flowchart illustrating an example of an image search process using attribute propagation information in the first embodiment. It is a figure which shows an example of the search screen in Example 1.
- FIG. In Example 1 it is a sequence diagram which shows an example of a process of the whole system.
- Example 2 it is a block diagram which shows the structural example of an image search system. It is a structural example of the annotation screen in Example 2.
- FIG. In Example 2, it is a flowchart which shows an example of an annotation process.
- Example 1 it is a figure explaining clustering and attribute propagation.
- the image search apparatus of the present embodiment extracts feature amounts of one or more predetermined attributes from a newly registered image, and extracts the feature amounts of each type of attribute and the same type of attribute features of each registered image. Compare the amount.
- the image search apparatus accumulates propagation information including links between feature amounts determined to have high similarity or feature amounts by the comparison.
- the link can be considered as a link between images via the same type of feature amount.
- a link (propagation) between feature quantities may be described as a link (propagation) between images.
- the image in this embodiment is a concept including a moving image and a still image.
- the attribute indicates a part constituting the image.
- each part such as an arm and a head of the person is an example of the attribute.
- the feature amount is a value indicating the feature of the image that each attribute has. A specific example of the feature amount will be described later.
- the feature amount of an attribute may be simply referred to as a feature amount.
- the propagation information includes, for example, information indicating the image of the propagation source and the image of the propagation destination, the type of feature amount, and the propagation intensity.
- the propagation intensity will be described later.
- a graph structure representing a link between images is formed.
- the graph is a concept including a tree.
- the image search apparatus can search for an image that is not directly similar to the query image by following the graph structure.
- the image search apparatus uses this propagation information at the time of search, for example, acquires a plurality of images linked to the query image of the given query from the database, and uses the obtained image as a query image as a query for the given query. Reinforce search queries by adding to The image search apparatus can improve the search accuracy and the coverage rate by performing similar image search using the reinforced query and integrating the search results.
- FIG. 1 is a block diagram illustrating a configuration example of an image search system 100 according to the first embodiment.
- the image search system 100 accumulates, as propagation information, a link with a registered image having a high-similarity feature type for one or more feature amounts of the input image, and makes a query using the propagation information.
- the search accuracy and the coverage rate are improved by performing a search using the reinforced query and the reinforced query.
- the image search system 100 includes, for example, an image storage device 101, an input device 102, a display device 103, and an image search device 104.
- the image storage device 101 is a storage medium for storing still image data or moving image data.
- the image storage device 101 is connected to a hard disk drive built in a computer or a network such as NAS (Network Attached Storage) or SAN (Storage Area Network). It is configured using a storage system.
- the image storage device 101 may be, for example, a cache memory that temporarily holds image data continuously input from the camera.
- the input device 102 is an input interface for transmitting user operations to the image search device 104, such as a mouse, a keyboard, and a touch device.
- the display device 103 is an output interface such as a liquid crystal display, and is used for displaying the recognition result of the image search device 104, interactive operation with the user, and the like.
- the image search device 104 performs registration processing that extracts information necessary for search from the images stored in the image storage device 101 and creates a database. Further, the image search device 104 performs a search process of searching for an image similar to the search query from the image database 108 using the search query designated by the user from the input device 102 and presenting information to the display device 103.
- the image search device 104 extracts feature quantities of one or more types of attributes from the image and registers them in the image database 108. In addition, the image search device 104 compares the feature amount of the registered image with the same type of feature amount of another registered image, and adds propagation information between features with high similarity or the same feature amount. Accumulate in the database.
- the image search device 104 includes an image input unit 105, a feature amount extraction unit 106, an attribute propagation unit 107, an image database 108, a propagation information storage unit 109, a query input unit 110, a query reinforcement unit 111, and an image search unit 112.
- the image input unit 105 accepts input of still image data or moving image data from the image storage device 101, and converts the input data into a data format used inside the image search device 104 according to the data format of the input data. Convert. For example, when the image input unit 105 receives input of moving image data, the image input unit 105 performs a moving image decoding process that decomposes the frame into still images (still image data format), for example.
- the image input unit 105 may perform a partial region extraction process from the input image, for example, in accordance with an instruction from the user or the like as necessary. Specifically, for example, the image input unit 105 may extract a person area from each frame, and use the image of the extracted area internally as data. The image data processed by the image input unit 105 is sent to the feature amount extraction unit 106. Note that the image input unit 105 may also receive input of feature amounts and register the input feature amounts in the image database 108. At this time, the image search device 104 may not include the feature amount extraction unit 106.
- the feature amount extraction unit 106 extracts feature amounts of one or more types of attributes from each input image.
- the feature amount is a value that can be compared between images indicating the feature of the image, and is represented by, for example, a fixed-length vector.
- An image feature amount that is data obtained by digitizing appearance information such as the color and shape of an image is an example of an attribute feature amount.
- any image information can be used as a feature amount as long as the values can be compared between images.
- a moving object ID obtained by tracking a moving object between frames may be used as an attribute feature amount in addition to the image feature amount such as the shape and color of the person image.
- Image information including images and feature amounts is registered in the image database 108.
- the feature amount extraction unit 106 may perform data clustering processing for the purpose of high-speed search when registering image feature amounts in the image database 108, for example.
- the feature quantity extraction unit 106 generates a cluster composed of one or more registered data having a feature quantity with a high degree of similarity, for example, by clustering processing such as a k-means algorithm.
- the similarity is an index indicating the similarity between two feature quantities of the same type, and is obtained, for example, by substituting the distance between the two feature quantities into a predetermined decreasing function whose value range is [0, 1].
- the image database 108 records, for example, a cluster representative value (for example, an average vector of cluster members) and a cluster member ID.
- the image search device 104 compares, for example, the feature quantity of the search query with the representative value of the cluster at the time of the search, and the feature quantity between the search query and the cluster member only for a cluster having a high degree of similarity. By performing the comparison, the number of processing times can be reduced and the search can be performed at high speed.
- the attribute propagation unit 107 compares the feature quantity of each registered image with the same kind of feature quantity of the newly registered image, and for example, uses the link between the feature quantities with high propagation strength as the propagation information in the image database. 108.
- the propagation intensity is an index indicating the similarity or identity between two similar feature quantities. For example, a value given by a function in which a value when two feature values match is 1 and a value when the two feature values do not match is 0 is an example of propagation intensity. Therefore, even when the similarity cannot be defined between the feature quantities, the propagation intensity can be defined.
- the above-described similarity is an example of propagation intensity.
- the attribute propagation unit 107 When an image has a plurality of attributes, the attribute propagation unit 107 generates propagation information for the number of attributes, for example. Examples of attribute feature amounts are given below.
- the image feature value obtained by digitizing the appearance information of the image itself such as the color and shape is an example of the feature value.
- the attribute propagation unit 107 when the similarity between image feature amounts is equal to or greater than a predetermined threshold, the attribute propagation unit 107 generates propagation information between the images, and uses the similarity as the propagation strength.
- the object ID in the time-series image recognition result such as moving object tracking is an example of the feature amount.
- the same object in successive frames has the same ID.
- the attribute propagation unit 107 generates propagation information in the same object ID of each frame and sets the propagation intensity to 1.0.
- the information indicating the position in the fixed point observation image is an example of a feature amount.
- the attribute propagation unit 107 when the fixed-point observation image is divided by a predetermined grid, the attribute propagation unit 107 generates propagation information in cells at the same position between different times and sets the propagation intensity to 1.0.
- the tag specified by the user is an example of a feature amount.
- the attribute propagation unit 107 generates propagation information between an image to which a user-specified tag is assigned and a reference image used as a clue for tagging the image.
- the attribute propagation unit 107 calculates the similarity of another feature amount between the image and the reference image, and sets the calculated similarity as the propagation intensity.
- the attribute propagation unit 107 may generate propagation information between images having the same tag and set the propagation intensity to 1.0. Details of the propagation information generation process using tags will be described later.
- the search history is an example of a feature amount.
- the attribute propagation unit 107 refers to an operation log for repeated search, generates propagation information between the search query image and the search result image, and, for example, determines the similarity between the feature amounts used for the image search. Propagation intensity.
- FIG. 13 is a diagram for comparing structuring of feature amount spaces by attribute propagation and clustering. Clustering forms a feature quantity group centered on a representative vector, whereas attribute propagation forms a graph representing a link between feature quantities.
- the image search device 104 can obtain a link between images at distant locations in the feature amount space by using attribute propagation. A method for generating propagation information will be described later with reference to FIGS. 4A to 4C.
- the image database 108 holds image information obtained by the registration process described above.
- the image database 108 includes a propagation information storage unit 109 that stores propagation information.
- the propagation information accumulation unit 109 may be arranged outside the image database 108.
- the main memory of the client device may accumulate the propagation information.
- the image search device 104 can use the propagation information that is temporarily different for each user.
- the image database 108 stores feature amounts, and the image search unit 112 performs a similar image search using the feature amounts. Similar image search is a function that rearranges and outputs data in the order in which the feature amount is close to the query. For example, the image search unit 112 compares feature amounts using the Euclidean distance between vectors.
- the structure of the image database 108 will be described later in detail with reference to FIG. The operation of each unit in the registration process of the image search device 104 has been described above. Next, the operation of each unit in the search process of the image search apparatus 104 will be described.
- the query input unit 110 receives a query specified by the user via the input device 102.
- the query is an ID of registered data
- the feature amount is acquired from the image database 108.
- the feature amount is obtained from the image by the same processing as the feature amount extraction unit 106.
- the query reinforcement unit 111 uses the propagation information accumulated in the image database 108 to acquire registration data related to the input query and use it as an additional query.
- the image search unit 112 performs a similar image search process on the image database 108 using the input query and the additional query obtained by the query reinforcement unit 111.
- the image search unit 112 performs aggregation processing such as sorting the search results obtained from each query in order of similarity, and leaving only the low ranking results for the data with the same ID. Further, at this time, the result of the query added by the query reinforcement unit 111 may be weighted to the similarity as necessary. Query reinforcement and image retrieval will be described later with reference to FIG.
- the display device 103 presents the search result to the user by displaying the search result obtained by the above search process.
- FIG. 2 is a block diagram illustrating a hardware configuration example of the image search system 100 according to the present embodiment.
- the image search device 104 is realized by, for example, a general computer.
- the image search device 104 may include a processor 201 and a storage device 202 that are connected to each other.
- the storage device 202 is configured by any type of storage medium.
- the storage device 202 may be configured by a combination of a semiconductor memory and a hard disk drive.
- the functional units such as the image input unit 105, the feature amount extraction unit 106, the attribute propagation unit 107, the image database 108 and the propagation information storage unit 109, the query input unit 110, the query reinforcement unit 111, and the image search unit 112 include, for example, This is realized by the processor 201 executing the processing program 203 stored in the storage device 202. In other words, the processing executed by each functional unit described above is executed by the processor 201 based on the processing program 203.
- the data of the image database 108 is included in the storage device 202, for example.
- the image search device 104 further includes a network interface device (NIF) 204 connected to the processor 201.
- the image storage device 101 may be a NAS or a SAN connected to the image search device 104 via the network interface device 204, for example.
- the image storage device 101 may be included in the storage device 202.
- FIG. 3 is an explanatory diagram showing a configuration and data example of the image database 108 of the present embodiment.
- the information used by the system may be expressed in any data structure without depending on the data structure.
- FIG. 3 shows an example of a table format, for example, a data structure appropriately selected from a table, list, database, or queue can store information.
- the image database 108 includes, for example, an image table 300 that holds image information and a propagation information table 310 that holds links between images.
- the table configurations and the field configurations of the tables in FIG. 3 are merely examples, and for example, tables and fields may be added according to the application. Further, the table configuration may be changed as long as similar information is held. For example, the image table 300 and the propagation information table 310 may be combined into a single table.
- the image table 300 includes, for example, an image ID field 301, an image data field 302, and an attribute 1 feature amount field 303.
- the image table 300 includes a plurality of feature quantity fields.
- the image table 300 in FIG. 3 is an example in which feature amounts of two attributes are extracted, and includes an attribute 2 feature amount field 304.
- the image ID field 301 holds an identification number of each image data.
- the image data field 302 holds, for example, image data used when displaying a search result in binary.
- Each of the attribute 1 feature quantity field 303 and the attribute 2 feature quantity field 304 holds a corresponding type of feature quantity.
- the feature amount is given by, for example, fixed-length vector data. Further, the feature amount may be scalar data as long as it is possible to compare the images as in the feature amount held in the attribute 2 feature amount field 304, for example.
- the propagation information table 310 includes, for example, a propagation information ID field 311, an attribute ID field 312, a propagation source field 313, a propagation destination field 314, and a propagation intensity field 315.
- the propagation information ID field 311 holds an identification number for attribute propagation between images.
- the attribute ID field 312 holds the feature amount ID of the attribute to be propagated.
- the ID of the attribute feature amount may be managed by an application or a database including a table.
- the propagation source field 313 holds an image ID that is an attribute propagation source.
- the propagation destination field 314 holds an image ID that is an attribute propagation destination.
- the propagation strength field 315 holds a numerical value of propagation strength or reliability. For example, the propagation intensity increases as the distance between the propagation feature vector and the propagation feature vector decreases.
- the propagation information table 310 may further include, for example, a field for recording a time when propagation is generated.
- the image search apparatus 104 generates an additional query using the propagation information constructed at the time of registration, and performs a search using a search query obtained by adding the additional query to the input query. And the efficiency of image search and analysis of the image database 108 can be improved.
- 4A to 4C are diagrams illustrating a process in which propagation information is generated in the registration process.
- 4A to 4C show a process in which the state of the image database 108 changes from state 1 to state 3 due to the addition of an image.
- 4A to 4C show an example in which a similarity can be defined between two feature quantities, and the similarity is adopted as a propagation strength.
- an image having an image ID N N is a natural number
- FIG. 4A shows an example in which the state of the image database 108 is state 1.
- the state 1 indicates a state in which the image 2 is added to the image table 300 in which the image 1 has been registered.
- the attribute propagation unit 107 compares the same type of feature quantity, and records the propagation information in the propagation information table 310 if the similarity is equal to or greater than a predetermined threshold.
- the attribute propagation unit 107 does not propagate because the similarity is less than or equal to the threshold for the attribute 1 feature amount. Further, the attribute propagation feature 107 records the propagation information 411 because the feature amounts match with respect to the attribute 2 feature amount, that is, the similarity is equal to or greater than the threshold value. Specifically, the attribute propagation unit 107 adds 2 indicating the identifier of the attribute 2 feature amount in the attribute ID field 312 and 2 that is the image ID of the registered image in the propagation source field 313 in the propagation destination field 314. 1 is stored in the propagation intensity field 315, and the image ID 1 is stored.
- FIG. 4B shows an example in which the state of the image database 108 is state 2.
- a state 2 indicates a state where the image 3 is newly registered in the image table 300.
- the attribute propagation unit 107 performs feature amount comparison between the image 3 and the image 1 and feature amount comparison between the image 3 and the image 2.
- the attribute propagation unit 107 records the propagation information 421 regarding the attribute 1 feature amount.
- the propagation information 411 related to the attribute 2 feature is recorded between the image 1 and the image 2
- the propagation information 421 related to the attribute 1 feature is recorded between the image 1 and the image 3.
- One propagation information indicates that the image 3 and the image 2 are related.
- FIG. 4C shows an example in which the state of the image database 108 is state 3.
- State 3 indicates a state in which image 4, image 5, and image 6 are further registered.
- a plurality of graph structures connected by propagation information based on attribute feature amounts are formed.
- the image search device 104 reduces the number of feature comparisons using the clustering described above, or narrows down the images to be compared for feature comparison using bibliographic data such as image registration time. In this case, the registration processing speed can be improved.
- FIG. 5 illustrates an example of processing in which the image search device 104 according to the present exemplary embodiment extracts image feature amounts and propagation information between images from a moving image or a still image input from the image storage device 101 and registers them in the image database 108. It is a flowchart to show.
- the image input unit 105 acquires image data from the image storage device 101, converts the acquired image data into a format that can be used inside the system as necessary, and records the image data in the image table 300 (S501). For example, when the input of moving image data is received, the image input unit 105 performs a moving image decoding process for decomposing the moving image data into frames (still image data format). Further, the image input unit 105 may perform a partial region extraction process as necessary.
- the image search apparatus 104 repeats the processing from step S503 to step S507 for the feature amount of each type of attribute given as, for example, a system design item (S502).
- the feature amount extraction unit 106 calculates the type of feature amount from the input image (S503).
- the feature amount extraction unit 106 registers the feature amount obtained in step S503 in the image table 300 (S504). For example, when the number of records in the image table 300 is equal to or greater than a predetermined threshold, the feature amount extraction unit 106 may perform clustering processing using feature amounts as necessary.
- the attribute propagation unit 107 calculates a propagation strength between the feature amount obtained in step S503 and the same feature amount of each image registered in the image table 300, and whether or not the propagation strength is equal to or greater than a threshold value. Is determined (S505). If there is a registered image whose propagation intensity is greater than or equal to the threshold (S505: YES), the attribute propagation unit 107 executes step S507; otherwise (S505: NO), the attribute propagation unit 107 moves to step S508.
- the attribute propagation unit 107 records propagation information indicating the link between the feature amount of the registered image and the feature amount of the input image whose propagation intensity is a threshold value in the propagation information table 310 (S507). If there is a feature amount of an attribute for which the processing in steps S503 to S507 has not been performed, the process moves to step S502, and processing relating to the feature amount of another attribute is performed (S508). If the processing in steps S503 to S507 has been executed for all feature amounts of the attributes, the processing in FIG. 5 ends.
- FIG. 6 is a diagram illustrating an example of processing in which the image search apparatus 104 according to the present embodiment searches for an image registered in the image database 108 using a query designated by the user.
- the user inputs information for generating an input query for searching for a desired image from the image database 108.
- the input query includes image feature values held in the image table 300.
- the query input unit 110 receives input of an image ID of an image included in the image table 300 and information indicating the type of feature amount, and inputs a combination of the image ID and the feature amount acquired from the image table 300.
- An image corresponding to the image ID included in the input query is called a query image.
- the query input unit 110 may accept, for example, input of an image newly given from the outside and the type of feature amount. At this time, the query input unit 110 extracts the feature amount of the type from the given image, for example, by the same process as the feature amount extraction unit 106, and for example, extracts the feature amount extracted most from the image table 300. One registered image having a feature quantity with high similarity is identified. The query input unit 110 uses, for example, a combination of an image ID of the specified registered image and the feature amount of the registered image as an input query.
- FIG. 6 shows an example in which the image ID of the input query 601 is 6, and the feature quantity type is attribute 1 feature quantity.
- the input query 601 may include a plurality of types of feature quantities.
- a plurality of images may be specified as the query image.
- the image search device 104 holds information indicating a link between registered images defined in the propagation information table 310, that is, information of a graph set 430 including one or more graphs.
- the query reinforcement unit 111 selects a graph including information of the query image specified by the input query 601 from the graph set 430, and follows the link using the feature amount specified by the input query 601 of the query image as a starting point. Search the selected graph.
- the query reinforcement unit 111 obtains the image 3 by following the link 602 from the attribute 1 feature amount of the image 6 specified by the input query 601.
- the query reinforcement unit 111 obtains a series of image sets including the image 6, the image 3, the image 1, and the image 2 that are on the route that follows the link in order from the feature amount of the obtained image.
- the query reinforcement unit 111 may perform attribute switching 603 in the same image to trace the feature amount link of the switching destination. By performing the attribute switching 603 in the same image by the query reinforcement unit 111, it is possible to improve the search coverage rate while maintaining the relationship between the feature amounts.
- the link is represented by a directed side, but the link may be an undirected side. That is, the image search apparatus 104 may perform a search according to the direction, or may perform a search ignoring the direction as shown in FIG.
- the query reinforcement unit 111 performs the above-described search using a search algorithm such as Dijkstra method.
- the query reinforcement unit 111 is a query obtained by adding, to the input query 601, an additional query that is a combination of the image IDs of the series of images obtained by the search described above and the same type of feature quantity as the input query in the image. Is a search query 604.
- the query reinforcement unit 111 performs a search using each feature quantity as a starting point, and generates a reinforcement query for each feature quantity.
- the image search unit 112 performs a similar image search for searching for an image having a similar feature amount from the image table 300 for each feature amount included in the search query 604.
- the image search unit 112 may search for images with similar types of feature amounts for each of the plurality of types of feature amounts, for example. An image having similar feature amounts obtained by combining the plurality of types of feature amounts may be searched.
- the image search unit 112 obtains a search result as a combination of the image ID and the similarity between the image ID and the query image ID that is the search source.
- the image search unit 112 rearranges all the obtained search results in the order of similarity and outputs them to the display device 103. At this time, for example, the image search unit 112 may output only search results with the same image ID that have a similarity greater than or equal to a predetermined threshold.
- the image search unit 112 may weight the similarity of the search results according to the propagation strength of the used query image from the input query image. Since the propagation intensity from the image 3 to the image 6 is 0.9, the image search unit 112, for example, multiplies the similarity of the search result with the image 3 as the query image by 0.9 as the weight.
- the image search unit 112 calculates, for example, the product of the propagation intensities of the plurality of propagation information with respect to the similarity of the search results obtained by tracing the plurality of propagation information from the image 6 that is the query image of the input query 601. Is multiplied by the weight.
- the image search unit 112 for example, the propagation strength between the feature amount of the image 6 that is the input query and the feature amount of the image of the search result with respect to the similarity of the search result obtained by performing the attribute switching.
- the calculated propagation intensity may be multiplied as the weight.
- FIG. 7 is a flowchart for explaining an example of processing in which the image search apparatus 104 according to the present embodiment searches for an image registered in the image database 108 using a query designated by the user. Hereinafter, each step of FIG. 7 will be described.
- the query input unit 110 receives information for generating an input query from the user, and generates an input query (S701).
- the query input unit 110 acquires the type of feature amount of the image having the image ID from the image table 300.
- the query input unit 110 receives external image data and the input of the type of feature amount, the query input unit 110 sets the image ID of the image having the feature amount similar to the feature amount of the type extracted from the image and the image ID.
- the feature amount of the type of the image it has is acquired from the image table 300.
- the query input unit 110 obtains, as an input query, a combination of the image ID of the query image and a predetermined feature amount of the query image.
- the query reinforcement unit 111 adds the input query obtained in step S701 to an empty query set (S702).
- the query reinforcement unit 111 executes the processing from step S704 to step S706 for each query included in the query set (S703).
- the query reinforcement unit 111 uses the query image as a propagation source and a propagation destination, and acquires propagation information having an attribute ID corresponding to the type of feature amount from the propagation information table 310 (S704).
- the query reinforcement unit 111 determines whether or not the acquired propagation information includes a propagation intensity that is greater than or equal to a threshold value (S705). If there is propagation information whose propagation intensity is greater than or equal to the threshold (S705: YES), step S706 is executed, otherwise (S705: NO), the process moves to step S707.
- the query reinforcement unit 111 uses, for example, a system predetermined value or a value given as a search parameter by the user as the threshold value. Note that the threshold value may be a different value for each attribute ID, for example.
- the query reinforcement unit 111 adds, to the query set, a query that uses an image whose propagation intensity is equal to or greater than a threshold as a query image (S706).
- a query image S706
- the query reinforcement unit 111 excludes, from the query set, images whose propagation time is within a predetermined time from the current time. May be.
- step S704 to step S706 If the processing of step S704 to step S706 has been executed for all the queries in the query set, the process moves to step S708, and if there is a query for which the processing of step S704 to step S706 has not been executed, the process returns to step S704. Processing is performed on the query (S707).
- the image search unit 112 executes the processing from step S709 to step S710 for all the queries included in the query set (S708). Note that, for example, for a query (a query with high similarity) in which the propagation intensity is equal to or greater than a predetermined threshold (the threshold is a value greater than the threshold in step S705), the image search unit 112 includes only a predetermined number of randomly selected queries. And the processing of step S708 to step S709 may be skipped for the remaining queries. There is a high possibility that similar image search results by two queries having extremely high propagation intensities are very similar to each other. Therefore, for example, for two queries having a propagation intensity higher than the threshold, the image search unit 112 can reduce the search time while suppressing a decrease in the search coverage rate by performing a search process for only one query. it can.
- the image search unit 112 performs an image search using the query and acquires a similar image from the image database 108. For example, the image search unit 112 obtains, as a search result, a similar image ID and a combination of similarities between the feature amount of the query image and the feature amount of the similar image (S709).
- the image search unit 112 assigns a weight according to the propagation strength between the input query and the corresponding query to the similar image obtained in step S709, and adds it to the set of search results (S710). For example, the image search unit 112 may switch whether or not to execute the process of step S710 according to a search parameter specified by the user.
- step S712 If the processes in steps S709 to S710 have been executed for all the queries included in the query set, the process moves to step S712. If there is a query for which the processes in steps S709 to S710 have not been executed, the process proceeds to step S712. Returning to S709, the query is processed (S711).
- the image search unit 112 rearranges the search results in descending order of the degree of similarity (or the degree of similarity after weighting if the weight is assigned to the degree of similarity), displays the result on the display device 103, and ends the processing (S712). .
- the image search unit 112 may collect only a predetermined number of search results in descending order of similarity.
- FIG. 8 is a diagram illustrating a configuration example of an operation screen for performing an image search using the image search apparatus 104 of the present embodiment.
- the operation screen is presented to the user on the display device 103, for example.
- the user operates the cursor 800 displayed on the screen using the input device 102 to give a processing instruction to the image search device 104.
- the operation screen includes, for example, a query image display area 801, a detailed option display button 802, an additional query display area 803, a propagation information display area 804, a search button 805, and a search result display area 806.
- Information displayed in the query image display area 801 is output to the display device 103 by the query input unit 110, for example.
- Information displayed in the additional query display area 803 and the propagation information display area 804 is output to the display device 103 by the query reinforcement unit 111, for example.
- the information displayed in the search result display area 806 is output to the display device 103 by the image search unit 112, for example.
- the operation screen may display a dialog for selecting a registered image or may include an interface for inputting an external image.
- the image designated by the user is displayed in the query image display area 801.
- the query reinforcement unit 111 generates an additional query using the propagation information of the query image specified by the user.
- the additional query display area 803 displays, for example, information on the generated additional query, for example, a query image of the additional query, a feature amount, and the like.
- the user can determine whether the additional query is appropriate. For example, when the user checks the detailed option display button 802, the additional query display area 803 may display information on the additional query.
- the propagation information display area 804 displays propagation information whose query image is a propagation source or a propagation destination, for example. By displaying the propagation information in the propagation information display area 804, it is possible to intuitively tell the user how the additional query was obtained.
- the user determines that the additional query displayed in the additional query display area 803 is not appropriate as the additional query, for example, the user can operate the additional query display area 803 and the propagation information display area 804 to exclude the query.
- an operation screen may be configured.
- the image search unit 112 When the user clicks the search button 805, the image search unit 112 performs a similar image search using the input query and the search query on the image database 108.
- the search results are rearranged in the order of similarity, for example, and the search results having the same image ID are collected and displayed in the search result display area 806.
- FIG. 9 is a sequence diagram illustrating an example of processing of the image search system 100 of the present embodiment.
- FIG. 9 specifically shows a processing sequence among the user 900, the image storage device 101, the computer 901, and the image database 108 in the image registration and image search processing of the image search system 100 described above.
- the computer 901 is a computer that implements the image search apparatus 104.
- the user 900 transmits a request and an instruction to the computer 901 via the input device 102 and inputs data, and receives a processing result from the computer 901 via the display device 103.
- S910 indicates a registration process
- S920 indicates a search process.
- the registration process S910 includes the processes shown in steps S911 to S916.
- the computer 901 issues an image data acquisition request to the image storage device 101 (S912), and acquires image data from the image storage device 101 (S913).
- the processes in steps S914 to S916 described below correspond to a series of registration processes described in FIG.
- the computer 901 extracts a feature amount from the acquired image, and registers the feature amount and image data in the image database 108 (S914).
- the computer 901 compares the acquired feature amount of the image with the same kind of feature amount of the image registered in the image database 108, and determines the link between the feature amounts having high propagation strength and the propagation strength at the link. Is recorded in the image database 108 (S915).
- the computer 901 notifies the user 900 of the completion of registration (S916).
- the search process S920 includes the processes shown in steps S921 to S928 and corresponds to the series of search processes described in FIG.
- the computer 901 reads query image data from the image database 108 (S922). Further, when an image is given from the outside in the search request, the computer 901 extracts a feature amount of the image, for example, selects an image having a high similarity in the feature amount from the image database 108, and selects the selected image.
- the query image is described in FIG.
- the computer 901 searches the image database 108 using the propagation information, and adds the image obtained by the search to the query set (S923).
- the added query is presented to the user 900 (S924), and the user 900 corrects the query as necessary and sends a search execution request to the computer 901 (S925).
- the computer 901 executes a similar image search using each image feature amount of the query set (S926).
- the computer 901 rearranges the obtained search results in order of similarity, aggregates the search results having the same image ID (S927), and presents the search results to the user 900 (S928).
- the image search apparatus 104 compares one or more types of feature amounts of the input image with the feature amounts of the registered images, generates propagation information indicating a link between images with high similarity, Record in the image database 108.
- the image search device 104 can increase the search coverage rate while ensuring search accuracy by performing a similar image search using a search query obtained by reinforcing an input query based on propagation information. .
- the image search apparatus 104 generates propagation information at the time of image registration.
- the image search apparatus 104 according to the present embodiment updates the propagation information after image registration or adds propagation information for a new attribute. You can do it.
- FIG. 10 is a block diagram illustrating a configuration example of the image search system according to the present embodiment.
- the image search system 100 assigns attributes and updates propagation information using image search.
- the image search system 100 according to the present embodiment includes a propagation information update unit 1001 in addition to the configuration of the image search system 100 according to the first embodiment.
- the propagation information update unit 1001 assigns attributes and propagation information to an image included in a search result obtained in the search process by the image search unit 112.
- the propagation information update unit 1001 gives propagation information between the query image and the search result image obtained by the image search unit 112, for example, in accordance with a user instruction.
- FIG. 11 is an example of a screen configuration for performing annotation on an image using the propagation information addition function.
- Annotation is a task of adding tags such as words and sentences that describe an image to the image, and is performed for the purpose of classifying and searching images by tags, analyzing a database, and the like. Further, if a sufficient number of image and tag pairs are obtained, an image classifier that recognizes an unknown image can be created by using machine learning.
- the annotation support screen in FIG. 11 is a screen for propagating a tag by human judgment when correct tagging cannot be expected only by automatic determination of image similarity.
- the annotation screen includes a tag input area 1101, a search button 1102, a reference image display area 1103, a propagation information display area 1104, a tagging candidate display area 1105, and a tagging button 1106.
- Information displayed in the reference image display area 1103 and the propagation information display area 1104 is output to the display device 103 by the query reinforcement unit 111, for example.
- the information displayed in the tagging candidate display area 1105 is output to the display device 103 by the image search unit 112, for example.
- the query reinforcement unit 111 acquires an image having the input tag as a reference image.
- a reference image display area 1103 displays a reference image.
- the image search unit 112 performs a similar image search using a predetermined feature amount other than the input tag using each reference image as a query, and acquires a similar image to which no input tag is assigned from the image database 108.
- the image search unit 112 rearranges search results including combinations of similar images and similarities, for example, in the order of similarity. Since the search result is similar to an image with a tag attached, it can be considered as a candidate image to which the same tag should be assigned.
- the tagging candidate display area 1105 displays the search result.
- the query reinforcement unit 111 may acquire propagation information between the reference image and the candidate image with reference to the propagation information table 310.
- the propagation information display area 1104 displays the propagation information acquired by the query reinforcement unit 111.
- the propagation information display area 1104 visualizes the propagation information, so that, for example, it becomes easy for the user to find a reference image that causes an erroneous candidate presentation.
- the candidate images c, g, and h have a propagation relationship with the reference image 4, and there is a possibility that an incorrect tag is assigned to the reference image 4 itself.
- the annotation screen may highlight such an image.
- the user checks the image displayed in the tagging candidate display area 1105 and selects the image to which the input tag should be added.
- the selection operation may be performed by dragging the mouse or by clicking a check box.
- the image search device 104 writes an input tag as an attribute feature amount for the selected image, and adds propagation information to the reference image.
- FIG. 12 is a flowchart showing an example of annotation processing. Hereinafter, each step of FIG. 12 will be described.
- the query input unit 110 acquires a tag input by the user (S1201).
- the query reinforcement unit 111 acquires a predetermined number of images with input tags from the image database 108, and the predetermined number of queries including a combination of the image ID of the acquired image and a predetermined type of feature amount other than the input tag. Is a query set (S1202). At this time, the query reinforcement unit 111 may select query images to be included in the query set using the propagation information. For example, since the similarity between certain feature amounts of propagation information is high, there is a high possibility that an image having the feature amount for tagging is redundant as a reference image. Therefore, for example, for the adjacent images on the propagation information graph structure, the query reinforcement unit 111 selects, for example, only one of the images as a query image. Thereby, the query reinforcement
- the image search unit 112 performs a similar image search using the predetermined type of feature amount of the reference image obtained in step S1202 (S1203). For example, the image search unit 112 rearranges the search results in descending order of similarity, and collects the search results having the same image ID. In addition, the image search unit 112 may collect and output search results similar to each other, thereby allowing the image search unit 112 to reduce the number of search results to be displayed while ensuring variations in the search results. it can.
- the image search unit 112 displays an image included in the search result on the display device 103 as a candidate image for tagging (S1204).
- the user selects an image to be tagged from the candidate images and informs the system.
- the display device 103 may highlight a reference image that is a propagation source or a propagation destination, and at this time, the user can easily confirm the propagation information.
- the image search apparatus 104 may issue an alert.
- the propagation information update unit 1001 gives a tag to the image selected by the user (S1205). Also, the propagation information update unit 1001 adds propagation information between the tag assigned to each reference image and the newly assigned tag. That is, the propagation information update unit 1001, for example, in the propagation information table 310, the ID indicating the input tag in the attribute ID field 312, the image ID of the query image in the propagation source field 313, and the selected image in the propagation destination field 314. Is stored as a value (for example, 1.0) designated by the user for the propagation intensity.
- the image search apparatus 104 can update the propagation information after image registration.
- the image search apparatus 104 according to the present embodiment can search tagging candidate images in annotation work with high accuracy and exhaustiveness.
- this invention is not limited to the above-mentioned Example, Various modifications are included.
- the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described.
- a part of the configuration of a certain embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of a certain embodiment.
- each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit.
- Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor.
- Information such as programs, tables, and files that realize each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
- control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Processing Or Creating Images (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
本実施形態の画像検索装置は、新規に登録する画像から所定の1種類以上の属性の特徴量を抽出し、抽出した各種類の属性の特徴量と、登録済み画像それぞれの同種の属性の特徴量と、を比較する。画像検索装置は、当該比較によって、類似性が高い又は特徴量が同一であると判定した特徴量間のリンクを含む伝搬情報、を蓄積する。当該リンクは、同一種類の特徴量を介した、画像間のリンクと考えることもできる。本実施形態において、特に区別する必要が無い場合においては、特徴量間のリンク(伝搬)を、画像間のリンク(伝搬)として記載することもある。なお、本実施形態における画像とは、動画及び静止画を含む概念である。
図1は、実施例1の画像検索システム100の構成例を示すブロック図である。画像検索システム100は、入力画像の1種類以上の特徴量に対して、類似性の高い当該種類の特徴量を有する登録済み画像とのリンクを伝搬情報として蓄積し、伝搬情報を用いてクエリを補強し、補強したクエリによる検索を行うことで検索精度及び網羅率を向上させる。
画像検索装置104は、画像記憶装置101に蓄積された画像から検索に必要な情報を抽出しデータベース化する、登録処理を行う。また、画像検索装置104は、ユーザが入力装置102から指定した検索クエリを用いて、画像データベース108から検索クエリに類似する画像を検索し、表示装置103に情報提示する、検索処理を行う。
Claims (14)
- 登録済み画像と、前記登録済み画像の第1種類を含む1種類以上の特徴量と、前記登録済み画像の同種の特徴量間のリンクを示す伝搬情報と、を保持する画像データベースから画像を検索する、画像検索装置であって、
第1画像及び前記第1画像の前記1種類以上の特徴量を取得し、前記画像データベースに登録する画像取得部と、
前記第1画像の各種類の特徴量について、前記登録済み画像の当該種類の特徴量それぞれと、の間の類似同一性を示す伝搬強度を算出し、前記第1画像の当該種類の特徴量と、前記第1画像の当該種類の特徴量との間の伝搬強度が最大である登録済み画像の当該種類の特徴量と、のリンクを、前記伝搬情報に含める属性伝搬部と、
前記画像データベースが保持する前記登録済み画像の特徴量から、第1種類の第1特徴量を取得し、前記第1特徴量を含む入力クエリを生成するクエリ入力部と、
前記伝搬情報が示す対応に基づいて、前記第1特徴量を出発点とした、前記登録済み画像の特徴量をたどる探索を行い、前記探索における経路上の特徴量を有する登録済み画像の第1種類の特徴量を含む追加クエリを生成する、クエリ補強部と、
前記入力クエリ及び前記追加クエリを使用して、前記画像データベースから画像を検索する、画像検索部と、を含む画像検索装置。 - 請求項1に記載の画像検索装置であって、
前記クエリ補強部は、前記探索において、同じ登録済み画像の異なる種類の特徴量をたどる、画像検索装置。 - 請求項1に記載の画像検索装置であって、
前記画像データベースが保持する伝搬情報は、前記登録済み画像の同種の特徴量間のリンクそれぞれにおける伝搬強度を含み、
前記属性伝搬部は、前記伝搬情報に含めたリンクにおける伝搬強度を、前記伝搬情報に含め、
前記画像検索部は、
前記入力クエリ及び前記追加クエリの特徴量それぞれと、前記登録済み画像の前記第1種類の特徴量それぞれと、の間の類似性を示す類似度を算出し、
前記算出した類似度それぞれに、前記探索における経路上において、当該類似度の算出に用いられた前記入力クエリ及び前記追加クエリの特徴量と、前記第1特徴量と、の間に存在する前記伝搬情報が示す伝搬強度、に基づく重みを付与し、
前記重みが付与された類似度に基づいて、前記画像データベースから画像を検索する、画像検索装置。 - 請求項1に記載の画像検索装置であって、
前記画像データベースが保持する伝搬情報は、前記登録済み画像の同種の特徴量間のリンクそれぞれにおける伝搬強度を含み、
前記属性伝搬部は、前記伝搬情報に含めたリンクにおける伝搬強度を、前記伝搬情報に含め、
前記画像検索部は、前記入力クエリ又は前記追加クエリに含まれる特徴量の組み合わせであって、前記探索における経路上において、前記伝搬情報が示す伝搬強度が閾値以上である特徴量の組み合わせ、が存在する場合、前記組み合わせに含まれる特徴量の一方を使用して、前記画像データベースから画像を検索する、画像検索装置。 - 請求項1に記載の画像検索装置であって、
前記クエリ補強部は、
前記追加クエリの特徴量を有する登録済み画像それぞれを前記画像データベースから取得し、
前記取得した登録済み画像それぞれと、前記探索における経路を示す情報と、を出力する、画像検索装置。 - 請求項1に記載の画像検索装置であって、
前記1種類以上の特徴量は、画像に第1タグが付与されているか否かを示す第2種類の特徴量と、前記第2種類と異なる第3種類の特徴量と、を含み、
前記クエリ補強部は、前記画像データベースから前記第1タグが付与されている登録済み画像を特定し、前記特定した登録済み画像それぞれの前記第3種類の特徴量を取得し、
前記画像検索部は、前記取得した第3種類の特徴量それぞれを含むクエリを用いて、前記画像データベースからタグ付け候補画像を検索し、
前記画像検索装置は、
前記タグ付け候補画像に含まれる第2画像に前記第1タグを付与し、前記第2画像に付与された第1タグと、前記特定した登録済み画像に付与されている第1タグそれぞれと、のリンクを前記伝搬情報に含める、伝搬情報更新部を含む、画像検索装置。 - 請求項6に記載の画像検索装置であって、
前記クエリ補強部は、
前記特定した登録済み画像それぞれと、前記タグ付け候補画像の前記第3種類の特徴量と、を前記画像データベースから取得し、
前記伝搬情報を参照して、前記取得した第3種類の特徴量それぞれと、前記タグ付け候補画像の前記第3種類の特徴量それぞれと、のリンクを取得し、
前記特定した登録済み画像それぞれと、前記取得したリンクを示す情報と、を出力し、
前記画像検索部は、前記タグ付け候補画像を出力する、画像検索装置。 - 画像検索装置が、登録済み画像と、前記登録済み画像の第1種類を含む1種類以上の特徴量と、前記登録済み画像の同種の特徴量間のリンクを示す伝搬情報と、を保持する画像データベースから、画像を検索する方法、であって、
前記方法は、
前記画像装置が、
第1画像及び前記第1画像の前記1種類以上の特徴量を取得し、前記画像データベースに登録する手順と、
前記第1画像の各種類の特徴量について、前記登録済み画像の当該種類の特徴量それぞれと、の間の類似同一性を示す伝搬強度を算出し、前記第1画像の当該種類の特徴量と、前記第1画像の当該種類の特徴量との間の伝搬強度が最大である登録済み画像の当該種類の特徴量と、のリンクを、前記伝搬情報に含める手順と、
前記画像データベースが保持する前記登録済み画像の特徴量から第1種類の第1特徴量を取得し、前記第1特徴量を含む入力クエリを生成する手順と、
前記伝搬情報が示す対応に基づいて、前記第1特徴量を出発点とした、前記登録済み画像の特徴量をたどる探索を行い、前記探索における経路上の特徴量を有する登録済み画像の第1種類の特徴量を含む追加クエリを生成する手順と、
前記入力クエリ及び前記追加クエリを使用して、前記画像データベースから画像を検索する手順と、を含む方法。 - 請求項8に記載の方法であって、
前記画像検索装置が、
前記追加クエリを生成する手順において、前記探索において、同じ登録済み画像の異なる種類の特徴量をたどる、方法。 - 請求項8に記載の方法であって、
前記画像データベースが保持する伝搬情報は、前記登録済み画像の同種の特徴量間のリンクそれぞれにおける伝搬強度を含み、
前記方法は、
前記画像検索装置が、
前記含める手順において、前記伝搬情報に含めたリンクにおける伝搬強度を、前記伝搬情報に含め、
前記検索する手順において、
前記入力クエリ及び前記追加クエリの特徴量それぞれと、前記登録済み画像の前記第1種類の特徴量それぞれと、の間の類似性を示す類似度を算出し、
前記算出した類似度それぞれに、前記探索における経路上において、当該類似度の算出に用いられた前記入力クエリ及び前記追加クエリの特徴量と、前記第1特徴量と、の間に存在する前記伝搬情報が示す伝搬強度、に基づく重みを付与し、
前記重みが付与された類似度に基づいて、前記画像データベースから画像を検索する、方法。 - 請求項8に記載の方法であって、
前記画像データベースが保持する伝搬情報は、前記登録済み画像の同種の特徴量間のリンクそれぞれにおける伝搬強度を含み、
前記方法は、
前記画像検索装置が、
前記含める手順において、前記伝搬情報に含めたリンクにおける伝搬強度を、前記伝搬情報に含め、
前記検索する手順において、前記入力クエリ又は前記追加クエリに含まれる特徴量の組み合わせであって、前記探索における経路上において、前記伝搬情報が示す伝搬強度が閾値以上である特徴量の組み合わせ、が存在する場合、前記組み合わせに含まれる特徴量の一方を使用して、前記画像データベースから画像を検索する、方法。 - 請求項8に記載の方法であって、
前記画像検索装置が、
前記追加クエリの特徴量を有する登録済み画像それぞれを前記画像データベースから取得する手順と、
前記取得した登録済み画像それぞれと、前記探索における経路を示す情報と、を出力する手順と、をさらに含む方法。 - 請求項8に記載の方法であって、
前記1種類以上の特徴量は、画像に第1タグが付与されているか否かを示す第2種類の特徴量と、前記第2種類と異なる第3種類の特徴量と、を含み、
前記方法は、
前記画像検索装置が、
前記画像データベースから前記第1タグが付与されている登録済み画像を特定し、前記特定した登録済み画像それぞれの前記第3種類の特徴量を取得する手順と、
前記取得した第3種類の特徴量それぞれを含むクエリを用いて、前記画像データベースからタグ付け候補画像を検索する手順と、
前記タグ付け候補画像に含まれる第2画像に前記第1タグを付与し、前記第2画像に付与された第1タグと、前記特定した登録済み画像に付与されている第1タグそれぞれと、のリンクを前記伝搬情報に含める、手順と、をさらに含む方法。 - 請求項13に記載の方法であって、
前記画像検索装置が、
前記特定した登録済み画像それぞれと、前記タグ付け候補画像の前記第3種類の特徴量と、を前記画像データベースから取得する手順と、
前記伝搬情報を参照して、前記取得した第3種類の特徴量それぞれと、前記タグ付け候補画像の前記第3種類の特徴量それぞれと、のリンクを取得する手順と、
前記特定した登録済み画像それぞれと、前記取得したリンクを示す情報と、を出力する手順と、
前記画像検索部は、前記タグ付け候補画像を出力する手順と、をさらに含む方法。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201580067867.1A CN107004025B (zh) | 2015-03-13 | 2015-03-13 | 图像检索装置及检索图像的方法 |
US15/556,994 US10657171B2 (en) | 2015-03-13 | 2015-03-13 | Image search device and method for searching image |
JP2017505775A JP6371465B2 (ja) | 2015-03-13 | 2015-03-13 | 画像検索装置、及び画像を検索する方法 |
PCT/JP2015/057516 WO2016147260A1 (ja) | 2015-03-13 | 2015-03-13 | 画像検索装置、及び画像を検索する方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2015/057516 WO2016147260A1 (ja) | 2015-03-13 | 2015-03-13 | 画像検索装置、及び画像を検索する方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2016147260A1 true WO2016147260A1 (ja) | 2016-09-22 |
Family
ID=56918489
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2015/057516 WO2016147260A1 (ja) | 2015-03-13 | 2015-03-13 | 画像検索装置、及び画像を検索する方法 |
Country Status (4)
Country | Link |
---|---|
US (1) | US10657171B2 (ja) |
JP (1) | JP6371465B2 (ja) |
CN (1) | CN107004025B (ja) |
WO (1) | WO2016147260A1 (ja) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018159095A1 (ja) * | 2017-02-28 | 2018-09-07 | 株式会社日立製作所 | 画像検索装置及び画像検索方法 |
WO2023203849A1 (ja) * | 2022-04-21 | 2023-10-26 | 株式会社日立製作所 | 空間可視化システム及び空間可視化方法 |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106354885A (zh) * | 2016-10-14 | 2017-01-25 | 百度在线网络技术(北京)有限公司 | 搜索方法和装置 |
CN112352230B (zh) * | 2018-06-28 | 2021-08-27 | 三菱电机株式会社 | 检索装置、检索方法及机器学习装置 |
JP7018001B2 (ja) | 2018-09-20 | 2022-02-09 | 株式会社日立製作所 | 情報処理システム、情報処理システムを制御する方法及びプログラム |
WO2020201866A1 (ja) * | 2019-03-29 | 2020-10-08 | 株式会社半導体エネルギー研究所 | 画像検索システム、及び画像検索方法 |
CN111125410A (zh) * | 2019-12-19 | 2020-05-08 | 湖北南楚网络传媒有限公司 | 一种海量图形图像智能识别检索系统 |
JP7227934B2 (ja) * | 2020-02-20 | 2023-02-22 | 株式会社日立製作所 | 車両運用支援装置、及び車両運用支援方法 |
CN113609335B (zh) * | 2021-08-12 | 2023-02-03 | 北京滴普科技有限公司 | 一种目标对象查找方法、系统、电子设备及存储介质 |
CN114119989B (zh) * | 2021-11-29 | 2023-08-11 | 北京百度网讯科技有限公司 | 图像特征提取模型的训练方法、装置及电子设备 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080069480A1 (en) * | 2006-09-14 | 2008-03-20 | Parham Aarabi | Method, system and computer program for interactive spatial link-based image searching, sorting and/or displaying |
JP2008129942A (ja) * | 2006-11-22 | 2008-06-05 | Hitachi Ltd | コンテンツ検索装置 |
JP2009245304A (ja) * | 2008-03-31 | 2009-10-22 | Fujitsu Ltd | 画像間対応付けプログラム、画像間対応付け装置および画像間対応付け方法 |
WO2012101697A1 (ja) * | 2011-01-26 | 2012-08-02 | パナソニック株式会社 | 画像管理装置、画像管理方法、プログラム、記録媒体、集積回路 |
JP2014032656A (ja) * | 2012-08-03 | 2014-02-20 | Fuji Xerox Co Ltd | コンテンツ・リンクを生成する方法、装置およびプログラム |
Family Cites Families (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3505393B2 (ja) * | 1998-06-19 | 2004-03-08 | 日本電信電話株式会社 | 類似オブジェクト検索方法、装置、および類似オブジェクト検索プログラムを記録した記録媒体 |
JP2003316819A (ja) * | 2002-04-22 | 2003-11-07 | Shinkichi Himeno | オブジェクト分類検索装置及びそれを実行するためのプログラム |
US20040015777A1 (en) * | 2002-07-22 | 2004-01-22 | International Business Machines Corporation | System and method for sorting embedded content in Web pages |
JP2007323319A (ja) * | 2006-05-31 | 2007-12-13 | Nippon Telegr & Teleph Corp <Ntt> | 類似検索処理方法及び装置及びプログラム |
JP4989308B2 (ja) * | 2007-05-16 | 2012-08-01 | キヤノン株式会社 | 画像処理装置及び画像検索方法 |
JP5199939B2 (ja) | 2009-04-15 | 2013-05-15 | ヤフー株式会社 | 画像検索装置、画像検索方法及びプログラム |
JP2012208553A (ja) * | 2011-03-29 | 2012-10-25 | Sony Corp | 画像処理装置、および画像処理方法、並びにプログラム |
CN102136007B (zh) * | 2011-03-31 | 2013-07-10 | 石家庄铁道大学 | 基于小世界特性的工程信息组织方法 |
JP6075294B2 (ja) * | 2011-12-12 | 2017-02-08 | 日本電気株式会社 | 画像処理システム及び画像処理方法 |
JP5955199B2 (ja) * | 2011-12-20 | 2016-07-20 | 富士フイルム株式会社 | 画像処理装置および画像処理方法、並びに、画像処理プログラム |
SG11201407749TA (en) * | 2012-05-24 | 2014-12-30 | Hitachi Ltd | Image analysis device, image analysis system, and image analysis method |
CN103995831B (zh) * | 2014-04-18 | 2017-04-12 | 新浪网技术(中国)有限公司 | 基于物品间相似度的物品处理方法、系统和装置 |
-
2015
- 2015-03-13 CN CN201580067867.1A patent/CN107004025B/zh active Active
- 2015-03-13 WO PCT/JP2015/057516 patent/WO2016147260A1/ja active Application Filing
- 2015-03-13 US US15/556,994 patent/US10657171B2/en active Active
- 2015-03-13 JP JP2017505775A patent/JP6371465B2/ja active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080069480A1 (en) * | 2006-09-14 | 2008-03-20 | Parham Aarabi | Method, system and computer program for interactive spatial link-based image searching, sorting and/or displaying |
JP2008129942A (ja) * | 2006-11-22 | 2008-06-05 | Hitachi Ltd | コンテンツ検索装置 |
JP2009245304A (ja) * | 2008-03-31 | 2009-10-22 | Fujitsu Ltd | 画像間対応付けプログラム、画像間対応付け装置および画像間対応付け方法 |
WO2012101697A1 (ja) * | 2011-01-26 | 2012-08-02 | パナソニック株式会社 | 画像管理装置、画像管理方法、プログラム、記録媒体、集積回路 |
JP2014032656A (ja) * | 2012-08-03 | 2014-02-20 | Fuji Xerox Co Ltd | コンテンツ・リンクを生成する方法、装置およびプログラム |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018159095A1 (ja) * | 2017-02-28 | 2018-09-07 | 株式会社日立製作所 | 画像検索装置及び画像検索方法 |
WO2023203849A1 (ja) * | 2022-04-21 | 2023-10-26 | 株式会社日立製作所 | 空間可視化システム及び空間可視化方法 |
Also Published As
Publication number | Publication date |
---|---|
JPWO2016147260A1 (ja) | 2017-09-28 |
CN107004025A (zh) | 2017-08-01 |
JP6371465B2 (ja) | 2018-08-08 |
US10657171B2 (en) | 2020-05-19 |
US20180060356A1 (en) | 2018-03-01 |
CN107004025B (zh) | 2020-07-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6371465B2 (ja) | 画像検索装置、及び画像を検索する方法 | |
JP6811645B2 (ja) | 画像検索装置及び画像検索方法 | |
US11243993B2 (en) | Document relationship analysis system | |
JP6190887B2 (ja) | 画像検索システムおよび情報記録媒体 | |
US20190251471A1 (en) | Machine learning device | |
CN106095738B (zh) | 推荐表单片段 | |
JP2008102594A (ja) | コンテンツの検索方法及び検索装置 | |
JP6615225B2 (ja) | 画像処理装置、画像検索インタフェース表示装置、及び画像検索インタフェースの表示方法 | |
US10346450B2 (en) | Automatic datacenter state summarization | |
US11080173B2 (en) | Boundary search test support device and boundary search test support method | |
JP5515117B2 (ja) | データ処理装置 | |
US9990444B2 (en) | Apparatus and method for supporting visualization of connection relationship | |
JP2019148859A (ja) | フローダイアグラムを用いたモデル開発環境におけるデザインパターンの発見を支援する装置および方法 | |
JP5954742B2 (ja) | 文書を検索する装置及び方法 | |
JP4189251B2 (ja) | キーワード解析方法及びそれに使用するプログラム | |
JP2016081132A (ja) | 分析プログラム、分析方法及び分析装置 | |
US20220342879A1 (en) | Data searching system, device, method and program | |
JP2024033123A (ja) | 文書分析装置、及び文書分析用プログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15885358 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2017505775 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15556994 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15885358 Country of ref document: EP Kind code of ref document: A1 |