WO2020208955A1 - 情報処理装置、情報処理装置の制御方法及びプログラム - Google Patents

情報処理装置、情報処理装置の制御方法及びプログラム Download PDF

Info

Publication number
WO2020208955A1
WO2020208955A1 PCT/JP2020/007317 JP2020007317W WO2020208955A1 WO 2020208955 A1 WO2020208955 A1 WO 2020208955A1 JP 2020007317 W JP2020007317 W JP 2020007317W WO 2020208955 A1 WO2020208955 A1 WO 2020208955A1
Authority
WO
WIPO (PCT)
Prior art keywords
score
data
information
processing
result
Prior art date
Application number
PCT/JP2020/007317
Other languages
English (en)
French (fr)
Inventor
慧 ▲高▼山
優和 真継
敦史 野上
Original Assignee
キヤノン株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by キヤノン株式会社 filed Critical キヤノン株式会社
Publication of WO2020208955A1 publication Critical patent/WO2020208955A1/ja
Priority to US17/492,708 priority Critical patent/US20220019848A1/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/40Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53Querying
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/54Browsing; Visualisation therefor
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/778Active pattern-learning, e.g. online learning of image or video features
    • G06V10/7784Active pattern-learning, e.g. online learning of image or video features based on feedback from supervisors
    • G06V10/7788Active pattern-learning, e.g. online learning of image or video features based on feedback from supervisors the supervisor being a human, e.g. interactive learning with a human teacher
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning

Definitions

  • the present invention relates to an information processing apparatus, a control method and a program of the information processing apparatus, and more particularly to a technique for appropriately determining parameters in image processing.
  • Patent Document 1 discloses a method of automatically determining the optimum parameter value by examining the similarity with the ideal data for each processing result.
  • Patent Document 2 a plurality of images acquired using different parameters are prepared, representative data is statistically selected from the images and displayed to the user, and the user is allowed to select the better one once. Discloses a method of determining the appropriate value of a parameter by trial of.
  • Patent Document 1 cannot be theoretically predicted, such as when ideal data cannot be prepared, or when the correct answer of the processing result cannot be determined objectively or objectively and must be subjective. Do not know the proper value of the parameter. Further, in the technique described in Patent Document 2, the burden is small because the user only needs to confirm the representative data, but the representative data does not always include the result of using the setting value of an appropriate parameter. ..
  • the present invention has been made in view of the above problems, and is for determining an appropriate value of a parameter while reducing the burden on the user in a situation where there are a plurality of data to be processed using the parameter. Providing technology.
  • the information processing device that achieves the above object is An image processing means for acquiring a plurality of images representing the results of executing the predetermined image processing on the target data using a plurality of candidate values of parameters related to the predetermined image processing.
  • a selection means for selecting a plurality of representative images from the plurality of images, and
  • a display control means for displaying the plurality of selected representative images on the display means, and
  • An operation means for accepting an operation input to the plurality of displayed representative images, and
  • a statistical processing means for determining at least one value from the plurality of candidate values based on the result of statistical processing of the information obtained by the received operation. It is characterized by having.
  • an appropriate value of the parameter can be determined while reducing the burden on the user.
  • the figure explaining the processing execution result and the example of the display screen The figure explaining the processing execution result and the example of the display screen
  • the figure explaining the processing example which determines an appropriate parameter The figure explaining the processing example which determines an appropriate parameter
  • the figure explaining the processing example which determines an appropriate parameter The figure which shows the functional structure of the information processing apparatus which concerns on one Embodiment A flowchart for explaining the flow of processing executed by the information processing apparatus according to the embodiment.
  • the figure explaining the processing example which determines an appropriate parameter The figure explaining the processing example which determines an appropriate parameter
  • the figure explaining the processing example which determines an appropriate parameter A flowchart for explaining the flow of processing executed by the information processing apparatus according to the embodiment.
  • the figure explaining the processing execution result and the example of the display screen The figure explaining the processing execution result and the example of the display screen
  • the figure explaining the processing execution result and the example of the display screen The figure explaining the processing example which determines an appropriate parameter
  • the figure explaining the processing example which determines an appropriate parameter The figure explaining the processing example which determines an appropriate parameter
  • the figure explaining the processing example which determines an appropriate parameter The figure which shows the functional structure of the information processing apparatus which concerns on one Embodiment
  • the figure explaining the processing example using the reference determination part The figure explaining the processing example using the reference determination part
  • the figure explaining the processing example using the reference determination part The figure explaining the processing example using the reference determination part
  • the information processing device inputs a plurality of candidate values for a plurality of data to be processed and parameters used for the processing, and outputs an appropriate value (also referred to as an appropriate parameter) of the parameter from among them.
  • the data to be processed using the parameters will be the data to be processed
  • the data obtained by executing the processing using the parameters will be the processing execution result
  • the values to be set in the parameters of interest as candidates for the appropriate parameters will be set. Described as a parameter candidate value.
  • the processing target data is an image obtained by shooting
  • the processing using parameters is a processing in which a model generated by machine learning is applied to the processing target data.
  • Machine learning is a technique related to a method of generating a model that automatically extracts specific information from data.
  • Machine learning using images is sometimes used as a method of constructing a model for detecting a region or contour of a specific object in an image.
  • the parameters at the time of learning include the number of learning times and the learning rate, and in the first embodiment, an example of applying a model having a different number of learning times to the data to be processed is shown.
  • a plurality of processing execution results are acquired by executing processing on a certain processing target data using a plurality of candidate values of parameters.
  • a plurality of representative data are selected from the processing execution results, the representative data is presented to the user, and suitable representative data is subjectively selected.
  • the information related to the user's selection will be referred to as the user selection information
  • the representative data selected by the user will be referred to as the user selection data.
  • a score is given to the parameter corresponding to the user selection data.
  • the relationship for example, similarity
  • the parameters of the processing execution result are similar. Is also given points.
  • the reason for scoring the processing execution results other than the user selection data is to consider the possibility that the appropriate processing execution result is not always selected as the representative data. After repeating the above processing while changing the data to be processed, the scores are totaled and the appropriate value of the parameter is determined.
  • the score of each candidate value of the parameter is determined based on the relationship between the user selection information and the processing execution result.
  • the relationship between the processing execution results may be used when selecting representative data. For example, the similarity between the processing execution results is examined as a relationship, and a combination of processing execution results having particularly low similarity is selected as representative data. If the relationship between the processing execution results is used when selecting the representative data, the processing execution results having a large difference in appearance can be easily displayed, and the user can easily make a judgment for the selection operation.
  • FIG. 1A is a hardware configuration diagram of the information processing device 100.
  • the information processing device 100 includes a CPU 1001, a ROM 1002, a RAM 1003, a storage device 1004, a bus 1005, an input device 1006, and a display device 1007.
  • the information processing device 100 is composed of a general-purpose personal computer or the like.
  • the CPU 1001 controls various operations of the information processing device 100 by the functional blocks described later.
  • the control content is instructed by a program on ROM 1002 or RAM 1003, which will be described later.
  • the CPU 1001 can also operate a plurality of computer programs in parallel.
  • the ROM 1002 stores a computer program and data in which the control procedure by the CPU 1001 is stored.
  • the RAM 1003 stores a control program for processing by the CPU 1001 and provides a work area for various data when the CPU 1001 executes various controls.
  • the function of the program code stored in the recording medium such as ROM 1002 or RAM 1003 is realized by being read and executed by the CPU 1001, but the type of the recording medium does not matter.
  • the storage device 1004 can store various data and the like.
  • the storage device 1004 includes a recording medium such as a hard disk, a floppy disk, an optical disk, a magnetic disk, a magneto-optical disk, a magnetic tape, or a non-volatile memory card, and a drive for driving the recording medium to record information.
  • the stored computer programs and data are called up on the RAM 1003 when necessary by instructions from a keyboard or the like or instructions from various computer programs.
  • Bus 1005 is a data bus or the like connected to each component, and is for realizing communication between each component and realizing information exchange at high speed.
  • the input device 1006 provides various input environments by the user.
  • a keyboard, mouse, or the like can be considered as providing various input operation environments, but a touch panel, stylus pen, or the like may also be used.
  • the display device 1007 is composed of an LCD display or the like, and displays to the user the status of various input operations and the calculation results corresponding thereto.
  • the configuration described above is an example, and is not limited to the configuration described above.
  • FIG. 1B is a diagram showing a functional configuration of the information processing apparatus 100.
  • the information processing device 100 is realized by executing software (program) acquired via a network or various information recording media by a computer composed of a CPU, a memory, a storage device, an input / output device, a bus, a display device, and the like. it can.
  • a general-purpose computer may be used, or hardware optimally designed for the software according to the present embodiment may be used.
  • the information processing device 100 includes a processing target data input unit 101, a parameter setting unit 102, a processing execution unit 103, a representative data selection unit 104, a display control unit 105, an operation unit 106, a statistical processing unit 107, and a user selection information holding unit 110. And a score information storage unit 111 is provided.
  • the processing target data input unit 101 has a function of receiving the processing target data.
  • the parameter setting unit 102 sets a plurality of candidate values for the parameters to be noted in performing the processing.
  • the processing execution unit 103 receives the processing target data and the candidate values of the parameters, executes the processing with different candidate values for the processing target data, and acquires the processing execution result.
  • the representative data selection unit 104 receives a plurality of processing execution results from the processing execution unit 103, and selects a plurality of representative data from the processing execution results.
  • the display control unit 105 receives the representative data from the representative data selection unit 104 and displays it on the display screen of the display device 1007 to prompt the user to perform the selection operation.
  • the operation unit 106 accepts an operation in which the user selects data from the representative data.
  • the operation unit 106 sends the user selection information based on the user's selection operation to the user selection information holding unit 110 and the statistical processing unit 107.
  • the user selection information holding unit 110 records each time the user selection information is received at any time. After repeatedly executing the processes of the representative data selection unit 104, the display control unit 105, and the operation unit 106 while changing the data to be processed, the statistical processing unit 107 performs statistical processing using the user selection information and the processing execution result. By doing so, the appropriate value of the parameter is determined.
  • the statistical processing unit 107 includes a score determination unit 108 and a totaling unit 109.
  • the score determination unit 108 determines the score for each candidate value of the parameter based on the relationship between the user selection information and the plurality of processing execution results.
  • the score determined here is stored in the score table by the score storage unit 111.
  • the totaling unit 109 aggregates the scores with reference to the score table of the score storage unit 111, and determines an appropriate value of the parameter.
  • the information of the user selection information holding unit 110 may be directly passed from the operation unit 106 to the statistical processing unit 107.
  • the information of the score information storage unit 111 may be directly passed from the score determination unit 108 to the totaling unit 109.
  • the processing target data input unit 101 reads the processing target data.
  • the parameter setting unit 102 sets a plurality of parameter candidate values.
  • the processing execution unit 103 executes processing for each candidate value for the processing target data, and acquires the processing execution result.
  • the statistical processing unit 107 investigates the similarity in the first embodiment as the relationship between the processing execution results.
  • the representative data selection unit 104 selects a combination of processing execution results having low similarity as representative data.
  • the display control unit 105 prompts the user to select desirable data from the representative data on the display screen by displaying the selected representative data.
  • the operation unit 106 accepts the user's selection operation.
  • the score determination unit 108 determines the score of each candidate value based on the relationship between the user selection information based on the selection operation by the operation unit 106 and the processing execution result. In S209, the score determination unit 108 determines whether or not the score determination for all the data to be processed has been completed. If it is not completed, the processes from S203 to S209 are repeated, and if it is completed, the process proceeds to S210.
  • the aggregation unit 109 aggregates all the processing target data using the scores determined by the score determination unit 108.
  • the totaling unit 109 determines an appropriate value of the parameter based on the totaling result and outputs it to the parameter setting unit 102.
  • the display control unit 105 displays an appropriate value of the parameter determined based on the aggregation result on the confirmation screen, and ends a series of processing.
  • each component of the information processing apparatus 100 ⁇ Processing content of each component> Subsequently, the processing contents of each component of the information processing apparatus 100 will be specifically described with reference to FIGS. 3A, 3B and 4A to 4D.
  • an image showing a person is used as data to be processed.
  • a model for extracting the outline of a person in the image is applied.
  • the parameter of interest in this embodiment is the number of learnings, which is one of the parameters related to model generation.
  • the processing target data input unit 101 has a function of inputting processing target data prepared by the user.
  • the user specifies the file of the data to be processed and the folder where the data to be processed is stored.
  • a large number of image files showing a person are designated and read.
  • the parameter setting unit 102 sets a plurality of candidate values for the parameter of interest.
  • a setting method prepare a file in which multiple candidate values are entered and specify the file. Alternatively, the user is prompted to input a candidate value on the display screen to set it.
  • the parameter of interest is the number of learnings
  • the candidate values are ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5.
  • prepare a learning model for each candidate value that is, in the present embodiment, five learned models having different learning times are prepared, such as a learning model in which the number of learnings is ⁇ 1 and a learning model in which the number of learnings is ⁇ 2.
  • the process execution unit 103 acquires a plurality of process execution results by executing a process using a set of candidate values for each process target data.
  • the score map is data of the same size as the data to be processed, and is an image in which the probability that each pixel seems to be the outline of a person is stored.
  • FIG. 3A shows an image of the score map as the processing execution result. Images 310, 320, 330, 340, and 350 are processing execution results obtained by executing processing on a certain processing target data image1, and correspond to candidate values ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5, respectively. And.
  • the representative data selection unit 104 selects a plurality of representative data from the plurality of processing execution results.
  • a selection method there is a method of selecting based on the relationship between the processing execution results.
  • the similarity between the processing execution results is examined as a relationship.
  • the similarity can be examined by calculating the degree of similarity between the processing execution results, and a combination of processing execution results having low similarity is selected and used as representative data.
  • the similarity of the processing execution results is calculated, for example, by obtaining the correlation of each score map. In the following, the information obtained by examining the relationship will be described as the relationship investigation result.
  • FIG. 4A shows an example of the relationship survey result in the processing target data image1.
  • the table shown in FIG. 4A stores the results of calculating the similarity for each combination of the processing execution results corresponding to the parameter candidate values ⁇ 1 to ⁇ 5.
  • the images 310 and 350 which are the combinations of the corresponding candidate values and are the processing execution results corresponding to ⁇ 1 and ⁇ 5, are selected as representative data.
  • the display control unit 105 creates a display image and outputs it to the screen of the display device 1007.
  • the display screen includes a work screen and a confirmation screen.
  • the confirmation screen will be described later, and the work screen will be described first.
  • the work screen is a screen during work to encourage the user to operate, and is basically a screen for displaying representative data and prompting the user to perform a selection operation.
  • Image 360 of FIG. 3B shows an example of a work screen.
  • the image 360 is the most basic example of the work screen, and the representative data, the image 310 and the image 350, are displayed.
  • the data to be processed may be displayed on the display screen.
  • an image in which representative data is superimposed on the processing target data may be created and displayed.
  • the processing execution result here, the score map
  • the processing execution result may be a transparent color. This improves visibility.
  • the display screen may display options (options related to the quality of the representative data) such as good, bad, and undecidable in case the user is uncertain about the selection. good.
  • options options related to the quality of the representative data
  • the option 371 is displayed in case the judgment is uncertain. In this way, by preparing options in case of doubt, the user does not have to select desirable data from the representative data each time. Therefore, when the representative data that is difficult to select is displayed, the user does not have to select it. Can reduce the time it takes to make a choice.
  • the operation unit 106 accepts an operation in which the user selects subjectively preferable data as user selection data from a plurality of representative data displayed on the display screen. Specifically, the selection operation is performed by holding the mouse pointer 363 over any of the representative data and clicking on the work screen 360. Further, when the options are displayed on the display screen in case the judgment is uncertain, the user can select those options on the operation unit 106. For example, on the work screen 370, when the user determines that both the images 310 and 350, which are the representative data, are good, the mouse pointer 363 is moved to the option 372 showing the characters that both are good and clicked. Perform the selection operation.
  • the user selection data acquired by the operation unit 106 and the option information prepared for the case where the user is uncertain about the selection are recorded in the user selection information holding unit 110 for each processing target data as user selection information.
  • the image 350 which is the representative data of the ⁇ 5
  • the image 350 will be described as being selected by the user in the operation unit 106.
  • the score determination unit 108 determines the score for each candidate value for each processing target data based on the relationship (here, similarity) between the user selection information and the processing execution result.
  • the method of determining the score will be described when the information of the user selection data is stored as the user selection information and the case where the information about the options is stored in case the user is uncertain about the selection.
  • the user selection data information is stored as the user selection information.
  • a score is given to the candidate value corresponding to the user selection data and the candidate value corresponding to the processing execution result having high similarity to the user selection data.
  • FIG. 4B shows an example of a score table when the user selects the image 350 (processing execution result of the candidate value ⁇ 5) as the user selection data from the representative data images 310 and 350.
  • Each row of the score table of FIG. 4B stores the score of each candidate value for each processing target data.
  • the score table since the candidate value corresponding to the user selection data is ⁇ 5, the dotted line region 420 showing the similarity between the processing execution results of ⁇ 5 and other candidate values in the relationship survey result of FIG. 4A is referred to, and the score table The information is stored at the dotted line 430.
  • the score is given to all the candidate values of the parameter, but the target to give the score may be only the candidate value of the processing execution result having a particularly high similarity to the user selection data. Specifically, among the processing execution results, a processing execution result having a higher degree of similarity to the user-selected data than the other representative data is selected, and a candidate value corresponding to the processing execution result is scored. Gives a positive value, and the score of other candidate values is 0 points.
  • the score is determined for all the candidate values including the non-representative data based on the selected contents. For example, if the user selects the option that all are good, all candidate values are given equal positive scores. If the user chooses the bad option, all candidate values are given equal negative scores. If the user selects the option of not being able to judge, 0 points are set for all candidate values. Further, when the user selects from these options, the data to be processed may be regarded as not subject to the score determination and may not be scored. In this way, by using the information on the options prepared for the case where the judgment is uncertain, the score can be determined more flexibly reflecting the intention of the user.
  • the aggregation unit 109 performs processing after the processing of the representative data selection unit 104, the display control unit 105, the operation unit 106, and the score determination unit 108 is repeatedly executed while changing the processing target data.
  • the totaling unit 109 aggregates the scores determined by the score determining unit 108, and determines an appropriate value from the candidate values.
  • the score is made into a histogram for each candidate value, and the candidate value having the highest score as a result of the aggregation is set as an appropriate value of the parameter.
  • the total score is calculated for each candidate value, and the aggregation result is obtained as shown in the dotted line area 440.
  • the candidate value ⁇ 4 is determined as an appropriate value.
  • the totaling unit 109 There are two types of timing for totaling in the totaling unit 109. The first is immediately after the score determination unit 108 determines the score for each data to be processed. This aggregation process is generated for the number of images of the data to be processed at the maximum, but it is also possible to aggregate a plurality of data to be processed at once without totaling each time the score is determined. The second is when the score has been determined for all the data to be processed, and this aggregation process occurs only once. In order to distinguish between the two, the result obtained by the former is described as the intermediate aggregation result, and the result obtained by the latter is described as the final aggregation result.
  • the display control unit 105 displays a confirmation screen.
  • the confirmation screen is a screen for allowing the user to confirm the result after the appropriate value is determined.
  • On the confirmation screen 380 of FIG. 3B in addition to the appropriate values of the parameters determined by the aggregation unit 109, information on the score, the processing execution result when the appropriate values of the parameters are used, and the coverage rate described later are displayed. .. An example of displaying the parameters finally determined as appropriate values is shown in the dotted line area 381 of the confirmation screen 380.
  • a histogram of the final aggregation result is shown in the dotted line area 383. It should be noted that the information on the score can be displayed not only on the confirmation screen but also on the work screen as needed for the calculation result on the way. By displaying the histogram, the user can grasp the range in which relatively good results can be obtained among all the candidate values of the parameter.
  • FIG. 384 and 385 An example of the processing execution result when an appropriate parameter value is used is shown in the dotted line areas 384 and 385. Each of these represents an example in which a high score was obtained and an example in which a low score was obtained when the candidate values were used.
  • the coverage rate is information on the ratio of the processing execution result desirable for the user when the processing is executed on all the processing target data using the parameter values determined by the aggregation unit 109. Specifically, the ratio of the number of data to be processed and the total number of data to be processed, which are scored above a predetermined value such as 0.5 points, is calculated from the parameters determined by the aggregation unit 109. An example of displaying the coverage rate is shown in the dotted line area 382. By displaying the coverage rate, the user can grasp how effective the finally determined appropriate parameter is for all the data to be processed.
  • the intermediate aggregation result calculated by the aggregation unit 109 can be used for the purpose of determining more appropriate parameters. A method of using the intermediate aggregation result in the parameter setting unit 102 and the display control unit 105 will be described below.
  • the parameter setting unit 102 can change the candidate value on the way based on the result of the calculation on the way. If there is a candidate value that is likely to be an appropriate value in the intermediate aggregation result, there is a possibility that the values before and after the candidate value have parameters that further suit the user's preference. On the contrary, if there is a candidate value that is unlikely to be an appropriate value in the intermediate aggregation result, it is considered that there is no need to look at the processing execution result of the candidate value any more. Therefore, if the total value of the scores of a specific candidate value is remarkable, the values before and after the candidate value are increased as the candidate value. On the other hand, if it is remarkable that the total score of a specific candidate value is low by examining the intermediate aggregation result, the value of the candidate value is deleted from the candidate values.
  • the process execution unit 103 executes processing using the modified candidate values.
  • the added candidate value may be a more appropriate value, it is difficult to finally obtain a high score because there is no score obtained so far at the time of addition even though the added value is added. Therefore, the score is set based on the total score of the close candidate values at that time as the intermediate aggregation result of the added candidate values. For example, as the intermediate aggregation result of the added candidate value, the same score as the intermediate aggregation result of the candidate value closest to the added candidate value is set. In this way, the parameter setting unit 102 can delete unnecessary candidate values by using the intermediate aggregation result, so that a series of processes can be made more efficient. Furthermore, since the candidate values can be limited around the values that are more preferable for the user, the possibility of finding more appropriate parameters increases.
  • the usage method in the display control unit 105 will be described.
  • the display control unit 105 performs the same display again.
  • the reason for displaying it again is to interpret that the user may have operated it by mistake and to reconfirm the user.
  • the timing of redisplaying may be immediately after the display related to the processing target data or after the display related to other processing target data.
  • the display control unit 105 can use the intermediate aggregation result to invalidate the score due to the user's erroneous operation, so that the reliability of the aggregation result can be improved. As a result, it can be expected that more appropriate parameters can be determined.
  • the relationship between the processing execution results is information about data similar to the representative data.
  • thumbnails of processing execution results having high similarity to the representative data are displayed around the representative data on the display screen.
  • FIG. 3B An example of displaying the relationship between the processing execution results is shown in the dotted line areas 391 and 392 of the image 390 in FIG. 3B.
  • the dotted line area 391 is displayed near the image 310 which is the representative data, and the processing execution result included in this area shows that the similarity with the image 310 which is the representative data is higher than that of the image 350 which is the representative data.
  • the dotted line region 372 is displayed near the image 350 which is the representative data, and the processing execution result included in this region has a higher degree of similarity to the image 350 which is the representative data than the image 310 which is the representative data. It is shown that.
  • the relationship when displaying the relationship of the processing execution result, the relationship may be changed by the operation unit 106.
  • the operation unit 106 performs an operation of changing the information related to the relationship and the score determination unit 108 changes the score according to the operation will be described.
  • the timing to change the information related to the relationship is when it is found that there is data that is not preferable for the user in the processing execution results displayed as data similar to the representative data that the user tried to select. At this time, by changing the information related to the relationship by the user's operation, the score is not given to the candidate value corresponding to the unfavorable processing execution result.
  • the user tried to select the representative data 310, but among the processing execution results displayed as data similar to the representative data 310, the processing execution result 330 is for the user. Not preferable.
  • the user uses the mouse pointer 363 to drag from the position of the image 330, which is the processing execution result, to the position of the dotted line area 392.
  • the candidate value corresponding to the image 330 which is the processing execution result can be prevented from being given a score.
  • the data to be processed is an image obtained by shooting, but the data is not limited to this, and may be vector data, moving image, three-dimensional data, or polyline data.
  • the image may be a grayscale image or a color image.
  • the processing target data may be data obtained by processing an image or an edge image obtained by performing edge extraction processing on the image.
  • the image processing includes an edge extraction process and a threshold value process.
  • the parameters of interest are the filter size and the filter type.
  • the processing using the parameter is the threshold value processing, the data to be processed becomes an edge image, and the parameter of interest becomes the threshold value.
  • a method of selecting the representative data by the representative data selection unit 104 a method of examining the similarity by calculating the similarity between the processing execution results and selecting the representative data based on the similarity is used.
  • the method of selecting representative data is not limited to this.
  • the representative data selection unit 104 may select representative data based on information that integrates the processing execution result and the processing target data. Specifically, a module that performs similarity determination of a new feature vector that integrates the feature vector of the processing target data and the processing execution result, and comprehensive similarity determination of the processing target data and the processing execution result (for example). Build a Deep Learning architecture). Representative data is selected by using such a module. Alternatively, a module that outputs representative data can be constructed and used by inputting a plurality of processing execution results and processing target data without explicitly evaluating and calculating the similarity. In this way, by using not only the processing execution result but also the information of the processing target data, it becomes possible to select the representative data with an emphasis on the important area on the processing target data. In addition, representative data may be randomly determined.
  • a method of setting a plurality of different candidate values by focusing on one parameter is shown, but two or more parameters can also be targeted.
  • the processing execution result is acquired for each candidate value, and the process of determining the score for each candidate value is repeated to determine an appropriate candidate value.
  • the process execution result is acquired for each candidate value pair, the process of determining the score for each candidate value pair is repeated, and finally an appropriate parameter pair is determined.
  • the process from the acquisition of the processing execution result of S203 to the determination of the score of S208 is repeated for each processing target data, but it can also be performed for all the processing target data at once. ..
  • the aggregation unit 109 determines the score of all the data to be processed and then obtains the final aggregation result.
  • the aggregation is performed every time the score is determined for the data to be processed. You may go at any time. By performing the aggregation at any time, the intermediate aggregation result can be used as described above, and as a result, the efficiency and reliability of the series of processing can be improved.
  • the aggregation unit 109 aggregates by making a histogram for each candidate value, and the candidate value having the highest score is set as an appropriate value.
  • the aggregation unit 109 will explain a method of calculating a candidate value located at the center of gravity in the parameter space using the score and setting the calculated candidate value as an appropriate value.
  • the processing target data is an edge image obtained by extracting an edge from an image showing a person.
  • the edge image is a grayscale image of the same size as the image, and the pixel value close to 1 is stored in the part with an edge and the pixel value close to 0 is stored in the part without an edge on the image before applying edge extraction. It is the two-dimensional data.
  • the process performed by the process execution unit 103 is a threshold value process, and the parameter of interest is a threshold value.
  • the configuration of the information processing apparatus according to the present embodiment is the same as the configuration described with reference to FIGS. 1A and 1B in the first embodiment, and the processing procedure is also the same as the flowchart of FIG. However, in the first embodiment, the similarity is calculated when the relationship is investigated, but the second embodiment is different in that the information about the group is requested.
  • the processing target data input unit 101 inputs the data group of the edge image obtained by performing edge extraction on the image group in which a person is captured as the processing target data.
  • Different threshold values are set for the parameter candidate values, which are ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5, respectively.
  • the processing execution unit 103 performs threshold processing on the processing target data using a set of candidate values.
  • Images 310, 320, 330, 340, and 350 in FIG. 3A are examples of processing execution results corresponding to the candidate values ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5.
  • the representative data selection unit 104 groups a plurality of processing execution results obtained by a plurality of candidate values, and selects one processing execution result from each group as representative data. Grouping is executed by clustering the processing execution results obtained by different candidate values for each processing target data. Clustering is a method of classifying a data group into a plurality of groups based on a data distribution.
  • FIG. 4C shows an example of the relationship survey result by grouping in one processing target data image1. It is assumed that g1 and g2 mean the group to which the processing execution result belongs. This table shows that ⁇ 1, ⁇ 2, and ⁇ 3 belong to group g1 and ⁇ 4 and ⁇ 5 are classified into group g2 by grouping.
  • a method of selecting based on the processing execution result There are two methods for selecting representative data from each group: a method of selecting based on the processing execution result and a method of selecting based on the candidate values of the parameters.
  • the processing execution result closest to the center of the group in the grouped space is used as the representative data.
  • the first method is to obtain the average value of the candidate values for each group and use the processing execution result of the candidate values close to the average value as the representative data.
  • the second method is a method in which when a value matching the average value does not exist in the candidate values, the average value is used as the parameter value and the processing execution result is reacquired and used as the representative data.
  • selecting representative data from each group it may be randomly selected.
  • the representative data selection unit 104 selects the processing execution results of the candidate values ⁇ 1 and ⁇ 5 as the representative data of the groups g1 and g2, respectively.
  • the score determination unit 108 gives a score to each candidate value.
  • the score is determined based on whether or not the processing execution result is the same as the group of user selection data.
  • a method of giving a score there is a method of giving an equal score to the group to which the user selection data belongs. For example, it is assumed that the processing execution result of ⁇ 5 is selected as the user selection data. Focusing on the processing target data image1 of the relationship survey result of FIG. 4C, the group g2 to which the user selection data belongs includes the candidate values ⁇ 4 and ⁇ 5. Therefore, in the score table shown in FIG. 4D, scores are evenly assigned to ⁇ 4 and ⁇ 5 as shown in the dotted line region 450.
  • the similarity between the user selection data and the processing execution result is examined, and only the candidate value of the corresponding group is assigned a degree of similarity as a score, and the other groups
  • the score of the candidate value may be 0 points.
  • the totaling unit 109 in the second embodiment examines the candidate value located at the center of gravity in the parameter space using the score and sets it as an appropriate value.
  • the total value of the scores acquired by each candidate value is obtained, the center of gravity in the parameter space is obtained using the total value and the candidate value, and the center of gravity is regarded as an appropriate value of the parameter.
  • an appropriate value is determined by obtaining the total score for each candidate value and then obtaining the center of gravity of the candidate value using the equation (1).
  • xi is the total score of each candidate value
  • ⁇ i is the candidate value
  • subscript i is a different value for each candidate value.
  • n is the number of candidate values.
  • the group of processing execution results is examined as a relationship, the score is determined based on the group, and the center of gravity is used at the time of aggregation. This can increase the possibility of finding a more appropriate candidate value.
  • the representative data is displayed only once for each data to be processed.
  • the representative data is displayed a plurality of times for each data to be processed. Specifically, when there are three or more representative data, considering that the result of selection by the user changes when the combination of representative data to be displayed is changed, the representative data is changed while changing the combination for each processing target data. indicate.
  • the third embodiment describes a method of analyzing a plurality of user selection information obtained for each processing target data, acquiring information regarding the ranking of candidate values, and determining a score based on the ranking.
  • an image showing a landscape or a person (for example, an image having a size of 640 pixcel ⁇ 480 pixcel) is treated as the processing target data for the purpose of determining a parameter that the user subjectively feels beautiful.
  • the process executed by the process execution unit 103 is a brightness conversion process, and the parameter of interest is a brightness conversion parameter.
  • FIG. 5 is a diagram showing a functional configuration of the information processing apparatus 500 according to the third embodiment.
  • the user selection information analysis unit 510 is added to the configuration of the information processing apparatus 100 described with reference to FIGS. 1A and 1B in the first embodiment.
  • the user selection information analysis unit 510 receives a plurality of user selection information for a certain processing target data from the user selection information holding unit 110, acquires the user selection information analysis result based on the user selection information, and obtains the user selection information analysis result. It is sent to the score determination unit 108.
  • the score determination unit 108 receives the user selection information analysis result from the user selection information analysis unit 510, the score determination unit 108 determines the score of each candidate value based on the relationship between the user selection information analysis result and the processing execution result.
  • the display control unit 105 selects one combination of data to be displayed on the display screen from the representative data.
  • the representative data selection unit 104 prompts the user to select preferable data by displaying the representative data.
  • the operation unit 106 accepts the user's selection operation.
  • the user selection information holding unit 110 records the user selection information.
  • the display control unit 105 determines whether or not all possible combinations of representative data are selected and displayed. If not all combinations are displayed, the process returns to the step of selecting combinations from the representative data of S601. On the other hand, when all the combinations have been displayed, the process proceeds to S604.
  • the user selection information analysis unit 510 analyzes the user selection information.
  • the score determination unit 108 determines the score of each candidate value for one processing target data based on the relationship between the user selection information analysis result and the processing execution result.
  • the score determination unit 108 determines whether or not the score of each candidate value has been determined for all the processing target data. If it has not been determined, the process returns to S203, and the process execution unit 103 applies the process to the next data to be processed and acquires the process execution result. On the other hand, if it has been decided, the process proceeds to S210.
  • the totaling unit 109 totals the points.
  • the tabulation unit 109 determines an appropriate value of the parameter.
  • the display control unit 105 displays a confirmation screen for the determined parameter.
  • the processing target data input unit 101 acquires a plurality of images as processing target data.
  • the parameter setting unit 102 sets a plurality of setting values of parameters related to brightness.
  • ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5 are set as candidate values.
  • the processing execution unit 103 performs brightness conversion on the processing target data image1 using the candidate values of ⁇ 1 to ⁇ 5, and acquires a plurality of processing execution results.
  • the process execution unit 103 groups the process execution results by clustering in the same manner as in the second embodiment in order to investigate the relationship between the process execution results, and as a result, the relationship investigation result shown in FIG. 7A is obtained.
  • the table of FIG. 7A represents a group to which each candidate value belongs, the processing execution results of ⁇ 1 and ⁇ 2 belong to group g1, the processing execution results of ⁇ 3 and ⁇ 4 belong to group g2, and the processing execution results of ⁇ 5 belong to groups. It shows that it belongs to g3.
  • the representative data selection unit 104 randomly selects representative data from each group based on the relationship survey results.
  • the processing execution results of ⁇ 1, ⁇ 3, and ⁇ 5 are selected.
  • the display control unit 105 changes and displays the combination to be displayed among the processing execution results of the candidate values ⁇ 1, ⁇ 3, and ⁇ 5.
  • the user who confirms the display performs a user selection data selection operation using the operation unit 106.
  • Each row of the table in FIG. 7B shows the result selected by the user when the combination of the representative parameters to be displayed is changed. 1 is stored in the candidate value of the user selection data, 0 is stored in the other candidate values, and NULL is stored when not displayed.
  • FIG. 7B shows that ⁇ 1, ⁇ 1, and ⁇ 3 were selected when the combinations of ⁇ 1 and ⁇ 3, ⁇ 1 and ⁇ 5, and ⁇ 3 and ⁇ 5 were displayed, respectively.
  • the user selection information analysis unit 510 refers to the user selection information related to the processing target data image1 and aggregates each parameter candidate value to obtain information on the ranking of the candidate values as the user selection information analysis result.
  • An example of information regarding the ranking in the third embodiment is shown in FIG. 7C.
  • ⁇ 1 was selected twice, ⁇ 3 was once, and ⁇ 5 was selected 0 times. Therefore, ⁇ 1, ⁇ 3, and ⁇ 5 were selected by 2 respectively. Points, 1 point, and 0 points are stored.
  • the score determination unit 108 determines the score based on the relationship between the processing execution results and the information regarding the ranking obtained as the user selection information analysis result.
  • the score table shown in FIG. 7D is created based on FIGS. 7A and 7C.
  • the scores of ⁇ 1, ⁇ 3, and ⁇ 5 are different according to the information regarding the ranking in FIG. 7C. Further, based on the relationship survey result of FIG. 7A, the same score is given to the candidate values of the same group.
  • the score determination unit 108 does not give a score to the data to be processed. You can also do it.
  • the above processing is repeatedly executed for all the data to be processed, and the aggregation unit 109 aggregates and determines an appropriate value of the parameter.
  • the user selection information analysis unit 510 examines the information regarding the ranking, and the score determination unit 108 shows a method of reflecting the information.
  • the present invention when displaying, the combination of representative data is changed and continuously displayed for one processing target data, but the present invention is not limited to this.
  • the display of the representative data obtained from the other processing target data may be inserted.
  • the score is determined by using the user selection information analysis result acquired by the user selection information analysis unit 510.
  • the information regarding the ranking is used in the third embodiment, whereas the information regarding the reliability described later is used in the fourth embodiment.
  • the data handled in the fourth embodiment is a large wall image of the infrastructure structure.
  • it is an object of the present invention to determine an appropriate parameter for acquiring the result of detecting a deformed portion such as a floating portion on a wall surface.
  • the processing contents will be described first.
  • the data is the same size as the wall surface image, and the value close to 1 for the deformed pixel, so Acquires a score map in which a value close to 0 is stored in a non-pixel.
  • polyline data is created using the score map, and the polyline data is used as the processing target data.
  • Polyline data is data composed of location information. As a method of creating a polyline, it can be created by binarizing the score map, performing a thinning process, and then further performing a vectorizing process.
  • An example of polyline data in the fourth embodiment is shown in 910 of FIG. 9A.
  • the line segments 911, 912, 913, 914, and 915 represent the polyline data, and indicate the position of the portion that seems to be deformed.
  • the polyline data may be represented by a line segment connecting adjacent points, or may be represented by a curve with mathematical information in addition to the position information.
  • the polyline that may be erroneously detected is removed from the polyline data.
  • extremely short polylines may be erroneously detected as holes or scratches on the wall surface rather than cracks or floats, so they should be deleted.
  • the parameter of interest is the length of the polyline used as the basis for removal.
  • the configuration of the information processing device according to the fourth embodiment is the same as that of the information processing device 500 described with reference to FIG. 5 in the third embodiment, the description thereof will be omitted.
  • the processing execution unit 103 confirms whether or not the processing from S203 to S602 has been performed for all the processing target data. If it is not carried out, the process returns to S203, and if it is carried out, the process proceeds to S802. In S802, the process execution unit 103 confirms whether or not the processes from S203 to S801 have been performed for all users. If it has not been executed, the process returns to S203 and the processes from S203 to S801 are executed for the users who have not yet executed it. On the other hand, if it can be carried out, the process proceeds to S604.
  • the user selection information analysis unit 510 analyzes the user selection information.
  • the score determination unit 108 assigns a score to each candidate value of the parameter based on the relationship between the processing execution result and the user selection information analysis result.
  • the score determination unit 108 determines whether or not the score determination for all the data to be processed has been completed. If it is not finished, the process returns to S604. On the other hand, if it is finished, the process proceeds to S210. In S210, the aggregation unit 109 aggregates all the processing target data using the scores determined by the score determination unit 108.
  • the aggregation unit 109 determines an appropriate value of the parameter based on the aggregation result and outputs it to the parameter setting unit 102.
  • the display control unit 105 displays an appropriate value of the parameter determined based on the aggregation result on the confirmation screen, and ends a series of processing.
  • the processing target data input unit 101 inputs the processing target data. For example, by applying a model for detecting deformation and image processing such as thinning or polylining, polyline data in a two-dimensional space of the same size as the original wall image is acquired from a large wall image. Here, a plurality of small areas having a size of about 500 pixcel ⁇ 500 pixcel are cut out from the space of the large wall surface image and used as the data to be processed.
  • a model for detecting deformation and image processing such as thinning or polylining
  • the parameter setting unit 102 sets the parameter of interest to the length of the reference polyline for removing the polyline data, and the candidate values are ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5.
  • the processing execution unit 103 receives the processing target data and the candidate value, and performs a process of removing a short polyline from the polyline data.
  • An example of the processing execution result is shown in FIG. 9B.
  • Images 920, 930, 940, 950, and 960 represent processing execution results corresponding to the candidate values ⁇ 1, ⁇ 2, ⁇ 3, ⁇ 4, and ⁇ 5. It shows the processing execution result when the length of the reference polyline to be removed is gradually increased from the image 920 to the image 960.
  • the representative data selection unit 104 selects the representative data.
  • the processing execution results are grouped as in the second embodiment as the relationship between the processing execution results, and as a result, the information shown in FIG. 10A is obtained.
  • the table of FIG. 10A represents the group to which each candidate value belongs, and it is shown that ⁇ 1, ⁇ 2, and ⁇ 3 belong to group g1, and ⁇ 4 and ⁇ 5 belong to group g2.
  • the processing execution results of the candidate values ⁇ 1 and ⁇ 5 are randomly selected as representative data from each group.
  • the display control unit 105 displays the processing execution results of the candidate values ⁇ 1 and ⁇ 5 for a plurality of users, and the operation unit 106 executes an operation of selecting user selection data.
  • the timing of display and operation does not need to be the same among users.
  • the user selection information holding unit 110 stores the user selection information.
  • FIG. 10B shows an example of user selection information regarding the processing target data image1.
  • Each row of the table represents the result of selection when the representative data is displayed to different users, and 1 is stored in the representative data selected by each user and 0 is stored in the representative data not selected.
  • users 1 and 2 select ⁇ 1, and user 3 selects ⁇ 5.
  • the user selection information analysis unit 510 acquires information on the reliability of the processing target data image1 as the user selection information analysis result.
  • the information on reliability is a score for each candidate value considering the degree of familiarity of each user with the selection operation. Specifically, the selection results of familiar users are given a high weight, and the selection results of unfamiliar users are given a low weight for aggregation.
  • As a method of determining an accustomed user and an unfamiliar user user selection information by a plurality of users is acquired in advance using different data groups to be processed, and the selection contents are outliers compared to other users. There is a way to consider many users as unfamiliar users. In the fourth embodiment, it is assumed that three users are targeted, users 1 and 2 are accustomed to it, and user 3 is not accustomed to it.
  • FIG. 10C shows an example in which the weights of the users 1 and 2 who are accustomed to the user 1 and the weights of the users 3 who are not accustomed to the user 3 are set to 0.5 points, and the information on the reliability is obtained.
  • ⁇ 1 is 2 points because two familiar users selected it
  • ⁇ 5 is 0.5 points because one unfamiliar user selected it.
  • the score determination unit 108 determines the score based on the relationship between the processing execution results and the information on reliability.
  • the score table shown in FIG. 10D is created based on FIGS. 10A and 10C.
  • the score of ⁇ 1 is 2 points
  • the score of ⁇ 5 is 0.5 points.
  • the candidate value ⁇ 2 of the same group is given the same score as ⁇ 1.
  • the same scores as ⁇ 5 are given to the candidate values ⁇ 3 and ⁇ 4 in the same group.
  • the above processing is repeatedly executed for all the data to be processed, and the aggregation unit 109 aggregates the data to determine an appropriate value of the parameter.
  • the display control unit 105 displays the coverage range on the confirmation screen.
  • the coverage range is information that can be displayed when handling a large image, and is a range of an image area in which a processing execution result preferable to the user is obtained by a certain candidate value.
  • the score table focusing on the score obtained in each processing target data when the candidate value determined by the aggregation unit 109 as an appropriate value is used, the processing in which the score equal to or higher than the predetermined value is obtained. Extract the target data.
  • FIG. 9C shows an example of the confirmation screen 970.
  • the dotted line region 971 displays the parameters determined as appropriate values.
  • the right side of the display image 970 is an example of the coverage range, and the area 973 in the large image 972 represents the coverage range according to the determined parameters.
  • the user selection information analysis unit 510 acquires the information regarding the reliability, and the score determination unit 108 reflects it in the score, so that a more appropriate value of the parameter can be determined.
  • time information from the display of the representative data to the user performing the selection operation can also be used.
  • the user selection information it is necessary to record the time from displaying the representative data to acquiring the user selection information. If the time from display to the user's selection is long, it is interpreted that the user has trouble determining which is preferable compared to the short time, and the score determination unit 108 gives a low score. To do. In this way, the quality of the score can be improved by using the time information.
  • the display control unit 105 displays the data to be processed on the display device 1007 to prompt the user to perform a selection operation.
  • the representative data may be very similar to each other, and it may be difficult to visually determine which is preferable.
  • the display control unit 105 statistically determines the criterion for displaying the representative data will be described using the relationship survey result reflecting the user operation.
  • an image of a person is handled, and the data to be processed is an edge image with edge extraction for the purpose of extracting the outline of the person.
  • the content of the process in the process execution unit 103 is a threshold value process, and the parameter of interest is a threshold value.
  • FIG. 11 is a diagram showing a functional configuration of the information processing apparatus 1100 according to the fifth embodiment.
  • the reference determination unit 1101 is added to the configuration of the information processing apparatus 100 described with reference to FIGS. 1A and 1B in the first embodiment.
  • the reference determination unit 1101 determines the standard of the data to be presented to the user by using the user selection information held in the user selection information holding unit 110. After receiving the representative data, the display control unit 105 determines whether or not to display the representative data on the display device 1007 based on the reference determined by the reference determination unit 1101, and the representative data to be displayed according to the determination result. Show only.
  • Image 1210 is an example of a display screen, and a display method in a case where similarity is examined as a relationship will be described.
  • Images 1215, 1216, 1217, 1218, and 1219 shown in the display image 1210 are processing execution results in which the processing execution unit 103 applies processing to a certain processing target data using different candidate values.
  • the dotted line areas 1211, 1212, and 1213 display the results of grouping the processing execution results based on the relationships.
  • the processing execution results 1215, 1216, and 1217 included in the dotted line region 1211 mean that they belong to the group g1, and the processing execution results 1218 and 1219 included in the dotted line region 1213 mean that they belong to the group g2.
  • the dotted line area 1212 is an area prepared for the processing execution result that could not be classified when grouped.
  • the user looks at the classification on the display screen, and if the classification is different from the visual impression, the user changes the affiliation of the processing execution result using the operation unit 106. For example, if the user wants to change the affiliation of the process execution result 1217 from the group g1 to unclassifiable, the user can not classify the affiliation from g1 by holding the mouse pointer 1214 over the process execution result 1217 and dragging in the direction of the arrow 1215. Change to.
  • the affiliation of all the processing execution results can be changed to unclassifiable by dragging in the same way.
  • an option indicating that classification is not possible may be prepared on the display screen so that the user can select the option.
  • FIG. 12B shows an example of the relationship reflecting the user operation.
  • the table of FIG. 12B shows the classification of the processing execution results of the candidate values for each processing target data.
  • the groups belonging to the candidate values ⁇ 1, ⁇ 2, ⁇ 4, and ⁇ 5 are stored, and the processing execution result of ⁇ 3 stores the NULL instead of the group because the user changed the affiliation to be unclassifiable. There is.
  • the reference determination unit 1101 statistically determines the standard of the representative data displayed by the display control unit 105 based on the relationship investigation result, the representative data, and the user selection information.
  • the reference determination unit 1101 first examines the similarity using the processing execution results confirmed to belong to different groups as a result of the user's visual inspection.
  • the similarity survey is performed with reference to FIG. 12B, which shows the relationship survey results for each data to be processed.
  • FIG. 12B shows the relationship survey results for each data to be processed.
  • the process execution results belonging to different groups are selected one by one, and the process of calculating the distance between the process execution results is repeatedly executed for all combinations.
  • the minimum value of the distance for each data to be processed is recorded in the reference analysis table shown in FIG. 12C.
  • the distance may be obtained by calculating the sum of squares of the differences between the pixels whose positions correspond to each other in the two processing execution results.
  • the distance is not calculated and the data is moved to the next processing target data.
  • the processing target data image1 of FIG. 12B since the group information g1 and g2 are stored, the distances of the processing execution results of different groups are calculated while changing the combination, and the reference analysis table shown in FIG. 12C is used. Stores the minimum distance.
  • the processing content after the standard is determined is the same as the processing described with reference to FIG.
  • the display control unit 105 calculates the distance between the representative data newly received from the representative data selection unit 104. Then, the distance is compared with the reference value determined by the reference determination unit 1101, and if the distance is equal to or greater than the reference value, it is determined that the data is different for the user and displayed on the display screen. On the contrary, if the distance is less than the reference value, the representative data is not displayed.
  • the representative data to be displayed can be limited, so that the display and selection can be performed efficiently.
  • the reference may be determined while repeating the processing of the representative data selection unit 104, the display control unit 105, and the operation unit 106 while changing the processing target data.
  • options are displayed on the display screen in case it is difficult to judge that all are good, all are bad, and cannot be judged. Then, except when they are selected, only when the representative data is selected, the distance between the representative data is calculated and stored in the reference analysis table. By obtaining the average value of the calculated distances of the processed data that has been displayed and using it as a reference, the reference can be updated while repeating the display. This can reduce the user's work compared to the method of collecting information in advance to determine the criteria.
  • the present invention supplies a program that realizes one or more functions of the above-described embodiment to a system or device via a network or storage medium, and one or more processors in the computer of the system or device reads and executes the program. It can also be realized by the processing to be performed. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.
  • a circuit for example, ASIC

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Human Computer Interaction (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

情報処理装置であって、所定の画像処理に係るパラメータの複数の候補値を用い、対象データに対して前記所定の画像処理を実行した結果を表す複数の画像を取得する画像処理手段と、前記複数の画像から、複数の代表画像を選択する選択手段と、選択された前記複数の代表画像を表示手段に表示させる表示制御手段と、表示された前記複数の代表画像に対して入力される操作を受け付ける操作手段と、受け付けられた前記操作によって得られた情報の統計処理の結果に基づいて、前記複数の候補値から、少なくとも1つの値を決定する統計処理手段とを備える。

Description

情報処理装置、情報処理装置の制御方法及びプログラム
 本発明は、情報処理装置、情報処理装置の制御方法及びプログラムに関し、特に、画像処理におけるパラメータを適切に決定するための技術に関する。
 画像処理において処理のパラメータを適切に決定することは重要な課題である。最良の処理の結果を選ぶ際に、理想的なデータや正解データを用意できる場合には、理想的なデータと最も近い処理の結果を調べれば良い。
 特許文献1は、処理の結果ごとに理想的なデータとの類似性を調べることで、自動的にパラメータ最適値を決める方法を開示している。
 特許文献2は、異なるパラメータを用いて取得した複数の画像を用意しておき、画像の中から代表のデータを統計的に選んだ上でユーザに表示し、良い方をユーザに選ばせ、一度の試行によりパラメータの適切な値を決定する方法を開示している。
特許第5368873号公報 特開2015-198300号公報
 しかしながら、特許文献1に記載の技術では、理想的なデータを用意できない場合や、処理結果の正解が一義的、客観的に定まらず主観によらざるを得ない場合など、理論的に予測できない場合には、パラメータの適切な値が分からない。また、特許文献2に記載の技術では、ユーザは代表データのみを確認すれば良いため負担が小さいものの、代表データに必ずしも適切なパラメータの設定値を用いた結果が含まれているとは限らない。
 このように、従来の技術では、パラメータを用いた処理を行う対象のデータが複数存在する状況において、ユーザの負担を軽減しながらパラメータの適切な値を決定することが難しいという課題がある。
 本発明は、上記の課題に鑑みてなされたものであり、パラメータを用いた処理を行う対象のデータが複数存在する状況において、ユーザの負担を軽減しながらパラメータの適切な値を決定するための技術を提供する。
 上記の目的を達成する本発明に係る情報処理装置は、
 所定の画像処理に係るパラメータの複数の候補値を用い、対象データに対して前記所定の画像処理を実行した結果を表す複数の画像を取得する画像処理手段と、
 前記複数の画像から、複数の代表画像を選択する選択手段と、
 選択された前記複数の代表画像を表示手段に表示させる表示制御手段と、
 表示された前記複数の代表画像に対して入力される操作を受け付ける操作手段と、
 受け付けられた前記操作によって得られた情報の統計処理の結果に基づいて、前記複数の候補値から、少なくとも1つの値を決定する統計処理手段と、
 を備えることを特徴とする。
 本発明によれば、パラメータを用いた処理を行う対象のデータが複数存在する状況において、ユーザの負担を軽減しながら、パラメータの適切な値を決定することができる。
 本発明のその他の特徴及び利点は、添付図面を参照とした以下の説明により明らかになるであろう。なお、添付図面においては、同じ若しくは同様の構成には、同じ参照番号を付す。
 添付図面は明細書に含まれ、その一部を構成し、本発明の実施の形態を示し、その記述と共に本発明の原理を説明するために用いられる。
一実施形態に係る情報処理装置のハードウェア構成例を示す図 一実施形態に係る情報処理装置の機能構成例を示す図 一実施形態に係る情報処理装置が実行する処理の流れを説明するフローチャート 処理実行結果と表示画面の例を説明する図 処理実行結果と表示画面の例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 一実施形態に係る情報処理装置の機能構成を示す図 一実施形態に係る情報処理装置が実行する処理の流れを説明するフローチャート 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 一実施形態に係る情報処理装置が実行する処理の流れを説明するフローチャート 処理実行結果と表示画面の例を説明する図 処理実行結果と表示画面の例を説明する図 処理実行結果と表示画面の例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 適切なパラメータを決定する処理例を説明する図 一実施形態に係る情報処理装置の機能構成を示す図 基準決定部を用いた処理例を説明する図 基準決定部を用いた処理例を説明する図 基準決定部を用いた処理例を説明する図
 以下、添付図面を参照して実施形態を詳しく説明する。なお、以下の実施形態は特許請求の範囲に係る発明を限定するものではない。実施形態には複数の特徴が記載されているが、これらの複数の特徴の全てが発明に必須のものとは限らず、また、複数の特徴は任意に組み合わせられてもよい。さらに、添付図面においては、同一若しくは同様の構成に同一の参照番号を付し、重複した説明は省略する。
 (実施形態1)
 本実施形態では、理想的なデータや正解データを用意できない場合において、パラメータを用いた処理を行う対象のデータが複数存在する状況を想定して、人が見るべきデータ量を減らしながら、パラメータに設定される適切な値を決定する。
 情報処理装置では、処理対象である複数のデータと、その処理に使用するパラメータについて複数の候補値を入力し、その中からパラメータの適切な値(適切なパラメータ、ともいう)を出力する。以降では、パラメータを用いた処理を行う対象となるデータを処理対象データ、パラメータを用いた処理を実行して得られるデータを処理実行結果、適切なパラメータの候補として注目するパラメータに設定する値をパラメータの候補値と記載する。
 実施形態1では、処理対象データは、撮影により得られる画像とし、パラメータを用いた処理は、機械学習により生成したモデルを処理対象データに適用する処理とする。機械学習は、データから特定の情報を自動で抽出するモデルを生成する方法に関する技術である。画像を用いた機械学習は、画像内の特定の対象物の領域や輪郭を検知するためのモデルを構築する方法として用いられることがある。学習時のパラメータには、学習回数や学習率があり、実施形態1では、処理対象データに学習回数が異なるモデルを適用する例を示す。
 <概要>
 本実施形態に係る情報処理装置の概要を説明する。まず、ある1つの処理対象データに対して、パラメータの複数の候補値を用いて処理を実行することで、複数の処理実行結果を取得する。次に、それらの処理実行結果から複数の代表データを選択し、代表データをユーザに提示し、好適な代表データを主観的に選択させる。以降では、ユーザの選択に関する情報をユーザ選択情報とし、特にユーザに選択された代表データをユーザ選択データと記載する。ユーザによる選択操作の後、ユーザ選択データに対応するパラメータに得点を付与する。このとき、処理実行結果とユーザ選択データとの関係性(例えば類似性)を調べ、その結果、処理実行結果のうちユーザ選択データと似ているデータがあれば、似ている処理実行結果のパラメータにも得点を付与する。ユーザ選択データ以外の処理実行結果にも得点をつけるのは、適切な処理実行結果が必ずしも代表データとして選ばれない可能性を考慮するためである。以上の処理を、処理対象データを変更しながら繰り返し実行した後、得点の集計を行い、パラメータの適切な値を決定する。
 このように、本実施形態では、ユーザ選択情報と、処理実行結果との関係性に基づいて、パラメータの各候補値の得点を決定する。
 また、処理実行結果の関係性については代表データの選択の際に利用しても良い。例えば、関係性として処理実行結果間の類似性を調べ、特に類似性の低い処理実行結果の組み合わせを代表データとして選択する。代表データの選択の際に処理実行結果の関係性を利用すると、見た目上差異の大きい処理実行結果が表示されやすくなり、ユーザにとって選択操作のための判断が容易となる。
 実施形態1では、処理実行結果の関係性として類似性を調べ、類似性に基づいて代表データの選択および得点の決定を行う方法を説明する。
 <情報処理装置のハードウェア構成>
 図1Aは、情報処理装置100のハードウェア構成図である情報処理装置100は、CPU1001、ROM1002、RAM1003、記憶装置1004、バス1005、入力装置1006、及び表示装置1007を備えている。情報処理装置100は汎用のパーソナルコンピュータなどにより構成される。
 CPU1001は、情報処理装置100の後述の機能ブロックによる各種動作を制御する。その制御内容は、後述するROM1002やRAM1003上のプログラムによって指示される。また、CPU1001は、複数の計算機プログラムを並列に動作させることもできる。ROM1002は、CPU1001による制御の手順を記憶させた計算機プログラムやデータを格納している。RAM1003は、CPU1001が処理するための制御プログラムを格納するとともに、CPU1001が各種制御を実行する際の様々なデータの作業領域を提供する。ROM1002やRAM1003などの記録媒体に格納されたプログラムコードの機能は、CPU1001が読み出して実行することによって実現されるが、記録媒体の種類は問われない。
 記憶装置1004は、さまざまなデータ等を記憶することができる。記憶装置1004は、ハードディスクやフロッピーディスク、光ディスク、磁気ディスク、光磁気ディスク、磁気テープ、不揮発性のメモリカード等の記録媒体と、当該記録媒体を駆動して情報を記録するドライブとを有する。保管された計算機プログラムやデータはキーボード等の指示や、各種計算機プログラムの指示により、必要な時にRAM1003上に呼び出される。
 バス1005は、各構成要素と接続されているデータバスなどであり、各構成要素間の通信を実現し、情報のやり取りを高速に実現するためのものである。入力装置1006は、ユーザによる各種入力環境を提供する。各種入力操作環境を提供するものとして、キーボードやマウス等が考えられるが、タッチパネル、スタイラスペン等であっても良い。
 表示装置1007は、LCDディスプレイなどで構成され、各種入力操作の状態やそれに応じた計算結果などをユーザに対して表示する。なお、以上述べてきた構成は一例であり、説明した構成に限定されるものでない。
 <情報処理装置の機能構成>
 図1Bは、情報処理装置100の機能構成を示す図である。情報処理装置100は、ネットワークまたは各種情報記録媒体を介して取得したソフトウェア(プログラム)を、CPU、メモリ、ストレージデバイス、入出力装置、バス、表示装置などにより構成される計算機によって実行することで実現できる。なお、計算機については、汎用の計算機を用いても良いし、本実施形態に係るソフトウェアに最適に設計されたハードウェアを用いても良い。
 情報処理装置100は、処理対象データ入力部101、パラメータ設定部102、処理実行部103、代表データ選択部104、表示制御部105、操作部106、統計処理部107、ユーザ選択情報保持部110、及び得点情報格納部111を備えている。
 処理対象データ入力部101は、処理対象データを受け付ける機能を持つ。
 パラメータ設定部102は、処理を行う上で注目するパラメータについて候補値を複数設定する。
 処理実行部103は、処理対象データとパラメータの候補値とを受け取り、処理対象データに対して異なる候補値により処理を実行して処理実行結果を取得する。
 代表データ選択部104は、処理実行部103から複数の処理実行結果を受け取り、処理実行結果の中から複数の代表データを選択する。
 表示制御部105は、代表データ選択部104から代表データを受け取り、表示装置1007の表示画面に表示してユーザに選択の操作を促す。
 操作部106は、ユーザが代表データの中からデータを選択する操作を受け付ける。操作部106は、ユーザの選択操作に基づくユーザ選択情報を、ユーザ選択情報保持部110と統計処理部107とへ送る。
 ユーザ選択情報保持部110は、ユーザ選択情報を受け取るたびに随時記録する。処理対象データを変更しながら、代表データ選択部104、表示制御部105、操作部106の処理を繰り返し実行した後、統計処理部107は、ユーザ選択情報と処理実行結果とを用いて統計処理を行うことでパラメータの適切な値を決定する。統計処理部107は、得点決定部108と集計部109とを備えている。
 得点決定部108は、ユーザ選択情報と複数の処理実行結果との関係性に基づいてパラメータの候補値ごとに得点を決定する。ここで決定する得点は、得点格納部111で得点テーブルに格納する。集計部109は、得点格納部111の得点テーブルを参照して得点の集計を行い、パラメータの適切な値を決定する。
 なお、図1Bの構成には記憶装置としてユーザ選択情報保持部110や得点情報格納部111を記載しているが、これらは構成上必須ではない。ユーザ選択情報保持部110の情報は操作部106から統計処理部107へ直接渡しても良い。得点情報格納部111の情報は得点決定部108から集計部109へ直接渡しても良い。
 <処理フロー>
 図2を用いて、実施形態1に係る情報処理装置100が実施する処理の手順を説明する。S201において、処理対象データ入力部101は、処理対象データを読み込む。S202において、パラメータ設定部102は、パラメータの候補値を複数設定する。S203において、処理実行部103は、処理対象データに対して候補値ごとに処理を実行し、処理実行結果を取得する。
 S204において、統計処理部107は、処理実行結果間の関係性として、実施形態1では類似性を調査する。S205において、代表データ選択部104は、類似性の低い処理実行結果の組み合わせを代表データとして選択する。S206において、表示制御部105は、選択された代表データを表示することにより、ユーザに表示画面上の代表データの中から望ましいデータを選択するように促す。S207において、操作部106は、ユーザの選択操作を受け付ける。
 S208において、得点決定部108は、操作部106による選択操作に基づくユーザ選択情報と、処理実行結果との関係性に基づいて、各候補値の得点を決定する。S209において、得点決定部108は、全処理対象データについて得点の決定を終えたか否かを判定する。終えていない場合は、S203からS209までの処理を繰り返し、終えている場合はS210へ進む。
 S210において、集計部109は、得点決定部108により決定された得点を用いて全処理対象データについて集計を行う。S211において、集計部109は、集計結果に基づいてパラメータの適切な値を決定し、パラメータ設定部102へ出力する。S212において、表示制御部105は、集計結果に基づいて決定されたパラメータの適切な値を確認画面に表示し、一連の処理を終了する。
 以上の処理により、ユーザの負担を軽減しながら、全処理対象データに対して、適切なパラメータを決定することができる。
 <各構成要素の処理内容>
 続いて、図3A、図3B及び図4A~図4Dを用いて、情報処理装置100の各構成要素の処理内容を具体的に説明する。実施形態1では、人の写っている画像を処理対象データとする。処理実行部103で処理対象データに対して実行する処理として、画像内の人の輪郭抽出を行うためのモデルを適用する。本実施形態で注目するパラメータは、モデル生成に関するパラメータの1つである、学習回数とする。
 まず、処理対象データ入力部101について説明する。処理対象データ入力部101は、ユーザが用意した処理対象データを入力する機能を持つ。入力のため、ユーザは処理対象データのファイルを指定したり、処理対象データが格納されているフォルダを指定したりする。実施形態1では、人の写った大量の画像ファイルを指定して読み込む。
 次に、パラメータ設定部102について説明する。パラメータ設定部102は、注目するパラメータについて候補値を複数設定する。設定方法として、複数の候補値を記入したファイルを用意してファイルを指定する。または、表示画面上でユーザに候補値の入力を促して設定させる。実施形態1では、注目するパラメータは学習回数であり、候補値はα1、α2、α3、α4、α5とする。設定後は、候補値ごとに学習モデルを用意する。つまり、本実施形態では、学習回数をα1として学習した学習モデル、学習回数をα2として学習した学習モデル、というように、学習回数の異なる5つの学習済みモデルを用意する。
 次に、処理実行部103について説明する。処理実行部103は、処理対象データごとに一通りの候補値を用いた処理を実行することにより、複数の処理実行結果を取得する。実施形態1では、処理対象データに対し、異なる学習回数により生成されたモデルを適用することで、人の輪郭抽出を行い、処理実行結果としてスコアマップを取得する。スコアマップは、処理対象データと同じサイズのデータで、各画素に人の輪郭らしいかという確率を格納している画像である。図3Aに、処理実行結果としてスコアマップのイメージを示す。画像310、320、330、340、350は、ある処理対象データimage1に対して処理を実行することで得られる処理実行結果であり、それぞれ候補値α1、α2、α3、α4、α5に対応するものとする。
 次に、代表データ選択部104について説明する。代表データ選択部104は、複数の処理実行結果の中から複数の代表データを選択する。選択方法には、処理実行結果間の関係性に基づいて選択する方法がある。実施形態1では、関係性として処理実行結果間の類似性を調べる。類似性は、処理実行結果間の類似度を算出することで調べることができ、類似性の低い処理実行結果の組み合わせを選択して、代表データとする。処理実行結果の類似度は、例えば、それぞれのスコアマップの相関を求める等により算出する。以降では、関係性を調べることで得られる情報を関係性調査結果と記載する。
 ここで、図4Aは、処理対象データimage1における関係性調査結果の例を示す。図4Aに示す表は、パラメータの候補値α1からα5に対応する処理実行結果の組み合わせごとに、類似度を算出した結果を格納している。ここでは点線領域410に示す、最も類似度が低い組み合わせに着目し、該当する候補値の組み合わせである、α1とα5に対応する処理実行結果である画像310と画像350を代表データとして選択する。
 次に、表示制御部105について説明する。表示制御部105は、表示画像を作成して表示装置1007の画面に出力する。表示画面には、作業画面と確認画面とがある。確認画面については後で説明するものとし、まずは作業画面について説明する。
 作業画面は、ユーザに操作を促すための作業中の画面であり、基本的には代表データを表示し、ユーザに選択の操作を促すための画面である。図3Bの画像360に作業画面の例を示す。画像360は作業画面の最も基本的な例であり、代表データである画像310、画像350が表示されている。表示画面上には、代表データの他に、処理対象データを表示しても良い。処理対象データと代表データとを同じ表示画面に表示することにより、ユーザは処理の適用前後の画像を確認できるようになるため、選択の判断が容易になる。
 また、表示画面上には、処理対象データに代表データを重畳した画像を作成して表示しても良い。重畳した画像を表示することで、処理対象データと代表データとの位置の対応が、ユーザにとってより分かりやすくなる。重畳する際、処理実行結果(ここではスコアマップ)を透過色にしても良い。これにより、視認性が向上する。
 また、表示画面には、代表データの他にも、ユーザが選択に迷う場合に備えて、いずれも良好、いずれも不良、判断不可、という選択肢(代表データの良否に関する選択肢)を表示しても良い。以降では、このような選択肢を、判断に迷うケースに備えた選択肢と記載する。図3Bの画像370に具体例を示す。画像370は作業画面であり、代表データである画像310、350の他に、判断に迷うケースに備えた選択肢371を表示している。このように、判断に迷うケースに備えた選択肢を用意することで、ユーザは代表データの中から望ましいデータを毎回選ばなくても良くなるため、選択が難しい代表データが表示されたときに、ユーザが選択に悩む時間を削減することができる。
 続いて、操作部106について説明する。操作部106は、ユーザが表示画面に表示されている複数の代表データから主観的に好ましいデータをユーザ選択データとして選択する操作を受け付ける。具体的には、作業画面360にてマウスポインタ363をいずれかの代表データにかざしてクリックすることにより選択の操作を行う。また、表示画面上に判断に迷うケースに備えた選択肢を表示している場合は、ユーザは操作部106にてそれらの選択肢を選択することができる。例えば作業画面370において、ユーザが代表データである画像310、350のいずれも良好であると判断した場合、マウスポインタ363を、いずれも良好という文字を示した選択肢372に移動してクリックすることで選択の操作を行う。
 このように操作部106で取得する、ユーザ選択データや、ユーザが選択した判断に迷うケースに備えた選択肢の情報は、ユーザ選択情報として処理対象データごとにユーザ選択情報保持部110に記録する。以降、操作部106でα5の方の代表データである画像350がユーザにより選択されたものとして説明を行う。
 次に、得点決定部108について説明する。得点決定部108では、ユーザ選択情報と、処理実行結果との関係性(ここでは類似性)に基づいて、各処理対象データについて候補値ごとに得点を決定する。得点の決定方法について、ユーザ選択情報としてユーザ選択データの情報が格納されている場合と、ユーザが選択した判断に迷うケースに備えた選択肢に関する情報が格納されている場合についてそれぞれ説明する。
 まず、1つ目の、ユーザ選択情報としてユーザ選択データの情報が格納されている場合について説明する。この場合、ユーザ選択データに対応する候補値と、ユーザ選択データと類似性の高い処理実行結果に対応する候補値とに得点を付与する。
 得点の付け方として、各候補値の処理実行結果とユーザ選択データとの類似度を得点とする方法がある。類似度を得点とすることにより、ユーザ選択データとの類似性が高い程、より高い得点を付けることになる。具体例として、代表データである画像310、350の中からユーザが画像350(候補値α5の処理実行結果)をユーザ選択データとして選択した場合の得点テーブルの例を、図4Bに示す。図4Bの得点テーブルの各行には処理対象データごとの各候補値の得点を格納している。ここではユーザ選択データに対応する候補値がα5であるため、図4Aの関係性調査結果における、α5と他の候補値の処理実行結果の類似度を示す点線領域420を参照し、得点テーブルの点線430の箇所にその情報を格納する。
 なお、上記の例ではパラメータの候補値の全てに得点を付与しているが、得点を付与する対象はユーザ選択データと特に類似性の高い処理実行結果の候補値のみとしても良い。具体的には、処理実行結果のうち、ユーザ選択データとの類似度がもう一方の代表データとの類似度に比べて高い処理実行結果を選び、その処理実行結果に対応する候補値の得点には正の値を付与して、その他の候補値の得点は0点とする。
 図4Aの関係性調査結果を用いて例を示す。代表データに対応するパラメータの候補値α1とα5における、各候補値との類似度として、点線領域470、420に注目する。本実施形態1では、α5の処理実行結果がユーザ選択データであるため、候補値の中から、α1よりα5との類似度が高いものを調べる。該当する候補値がα4のみであることから、α4、α5に正の得点を付与する。このように、ユーザ選択データと類似性の高い処理実行結果の候補値のみに得点を付与することにより、ユーザの意図に沿わない得点付与を行うリスクを軽減できる。
 次に、2つ目の、ユーザ選択情報として、判断に迷うケースに備えた選択肢(代表データの良否に関する選択肢)に関する情報が格納されている場合について説明する。この場合、その選択内容に基づいて代表データでないものも含め、全ての候補値について得点を決定する。例えば、ユーザがいずれも良好という選択肢を選択した場合は、全ての候補値に等しく正の得点を付ける。ユーザがいずれも不良という選択肢を選択した場合は、全ての候補値に等しく負の得点を付ける。ユーザが判断不可という選択肢を選択した場合は、全ての候補値に0点を設定する。また、ユーザがこれらの選択肢から選択したときには、その処理対象データは得点決定の対象外とみなして得点をつけないようにしても良い。このように、判断に迷うケースに備えた選択肢に関する情報を用いることで、ユーザの意志をより柔軟に反映した得点決定を行うことができる。
 次に、集計部109について説明する。集計部109は、代表データ選択部104、表示制御部105、操作部106、および得点決定部108の処理が、処理対象データを変更しながら繰り返し実行された後に処理を行う。集計部109は、得点決定部108で決定した得点の集計を行い、候補値から適切な値を決定する。実施形態1では、集計は、候補値ごとに得点をヒストグラム化し、集計の結果、最高点を得た候補値をパラメータの適切な値とする。
 具体的には、図4Bの得点テーブルにおいて、候補値ごとに得点の合計を算出し、点線領域440に示すように集計結果を得る。この例では候補値α4の集計結果が最高値であることから、候補値α4を適切な値として決定する。
 なお、集計部109で集計を行うタイミングは2種類ある。1つ目は、得点決定部108で各処理対象データについて得点を決定した直後である。この集計処理は最大で処理対象データの画像枚数分だけ発生するが、得点を決定する度に毎回集計せずに、複数の処理対象データについてまとめて集計することもできる。2つ目は、全処理対象データについて得点の決定を終えた時点であり、この集計処理は一度だけ発生する。両者の区別のため、前者で得る結果を途中集計結果、後者で得る結果を最終集計結果と記載するものとする。
 集計部109の処理の後は、表示制御部105が確認画面を表示する。確認画面は、適切な値が決定された後、ユーザに結果を確認させるための画面である。図3Bの確認画面380には、集計部109で決定されたパラメータの適切な値の他に、得点に関する情報、パラメータの適切な値を用いたときの処理実行結果、後述のカバー率を表示する。最終的に適切な値として決定したパラメータの表示の例を、確認画面380の点線領域381に示す。
 得点に関する情報の表示の例として、点線領域383に最終集計結果のヒストグラムを示す。なお、得点に関する情報は、確認画面のみならず、随時途中集計結果を求めて作業画面に表示することもできる。ヒストグラムを表示すると、ユーザは、パラメータの全候補値の中で、比較的良好な結果が得られる範囲を把握できる。
 適切なパラメータの値を用いたときの処理実行結果の例を点線領域384、385に示す。これらはそれぞれ、その候補値を用いたときに、高得点を取得した例と低得点を取得した例を表している。このように、決定した候補値による様々な処理実行結果を表示することで、ユーザにその候補値を採用することの最終確認をさせることができる。
 カバー率は、集計部109で決定されたパラメータの値を用いて全処理対象データに処理を実行したときに、ユーザにとって望ましい処理実行結果の占める割合に関する情報である。具体的には、集計部109で決定したパラメータにより例えば0.5点などの所定値以上の得点を得た処理対象データ数と、全処理対象データ数との比を算出する。点線領域382に、カバー率の表示の例を示す。カバー率を表示することで、ユーザは、最終的に決定した適切なパラメータが、全処理対象データに対してどの程度有効なものであるかを把握できる。
 以上に述べた各構成要素の処理によって、パラメータの適切な値を決定することができる。
 <途中集計結果の利用>
 集計部109で算出される途中集計結果は、より適切なパラメータを決定することを目的として利用することができる。以下に、途中集計結果を、パラメータ設定部102、表示制御部105で利用する方法を説明する。
 まず、パラメータ設定部102における利用方法を述べる。パラメータ設定部102は、途中集計結果に基づいて、候補値を途中で変更することができる。途中集計結果において適切な値となる可能性の高い候補値があれば、その候補値の前後の値に、さらにユーザの好みに合うパラメータがある可能性がある。反対に、途中集計結果において適切な値となる可能性の低い候補値があれば、その候補値の処理実行結果は、それ以上は見る必要がないと考えられる。このため、途中集計結果を調べ、ある特定の候補値の得点の合計値が高いことが顕著であれば、その候補値の前後の値を候補値として増やす。一方で、途中集計結果を調べ、ある特定の候補値の得点の合計値が低いことが顕著であれば、その候補値の値を候補値の中から削除する。
 候補値の追加や削除を行った後は、処理実行部103が、修正した候補値を用いて処理を実行する。ただし、追加した候補値は、より適切な値である可能性があるため追加したにも関わらず、追加した時点ではそれまでの獲得得点がないため、最終的に高い得点を得ることが難しい。このため、追加した候補値の途中集計結果として、近い候補値のその時点における合計点に基づいて得点を設定する。例えば、追加した候補値の途中集計結果として、追加した候補値と最も近い候補値の途中集計結果と同じ得点を設定する。このように、パラメータ設定部102が途中集計結果を利用することにより、不要な候補値を削除できるため、一連の処理を効率化することができる。さらに、ユーザにとってより好ましい値を中心に候補値を限定できるため、より適切なパラメータを見つけられる可能性が高まる。
 次に、表示制御部105における利用方法を述べる。表示制御部105は、途中集計結果において得点の低い(例えば所定値以下の)候補値に対応する代表データが、ユーザにより選択された場合、同じ表示を再度行う。再度表示するのは、ユーザが誤って操作した可能性があると解釈し、ユーザに再確認させるためである。再表示を行うタイミングは、その処理対象データに関する表示の直後であっても、他の処理対象データに関する表示の後であっても良い。
 ここで、ユーザ選択データが1回目と2回目とで同じである場合、その処理対象データにおけるユーザの操作は誤操作ではなかったと解釈し、その処理対象データについて、通常通り得点を決定する。一方、ユーザ選択データが1回目と2回目とで異なる場合は、1回目の操作は誤操作であったと解釈する。そして、その処理対象データについては得点を無効とするか、2回目の操作により選択されたユーザ選択データを用いて得点を決定する。このように、表示制御部105が途中集計結果を利用することにより、ユーザの誤操作による得点を無効にすることができるため、集計結果の信頼性を向上させることができる。結果として、より適切なパラメータを決定できるようになることが期待できる。
 このように、途中集計結果を利用することで、一連の処理の効率や信頼性を向上できる。
 <関係性に関する情報の操作>
 次に、表示制御部105が、作業画面に、処理実行結果の関係性に関する情報を表示する方法について説明する。
 処理実行結果の関係性は、代表データと似ているデータに関する情報である。表示する方法としては、表示画面上の代表データの周辺に、代表データとの類似性が高い処理実行結果のサムネイルを表示する。
 処理実行結果の関係性の表示例を、図3Bの画像390の点線領域391、392に示す。点線領域391は代表データである画像310の近くに表示されており、この領域に含まれる処理実行結果は、代表データである画像350より代表データである画像310との類似度が高いことを示している。また、点線領域372は、代表データである画像350の近くに表示されており、この領域に含まれる処理実行結果は、代表データである画像310より代表データである画像350との類似度が高いことを示している。このように、処理実行結果間の関係性に関する情報を表示することによって、代表データの選択により、どの処理実行結果に対応する候補値に高い得点が付与されるかをユーザが容易に認識できるようになる。
 また、処理実行結果の関係性を表示する場合に、操作部106により関係性を変更できるようにしても良い。以下、操作部106により関係性に関する情報の変更の操作を行い、当該操作に応じて、得点決定部108が得点を変更する方法を説明する。
 関係性に関する情報の変更を行うタイミングは、ユーザが選択しようとした代表データに類似したデータとして表示される処理実行結果の中に、ユーザにとって好ましくないデータがあることが分かった時である。このとき、ユーザの操作により関係性に関する情報を変更することで、好ましくない処理実行結果に対応する候補値には得点が付与されないようにする。
 具体的には、図3Bの作業画面390において、ユーザが代表データ310を選択しようとしたが、代表データ310と類似したデータとして表示されている処理実行結果のうち、処理実行結果330がユーザにとって好ましくないとする。このとき、ユーザはマウスポインタ363を用いて、処理実行結果である画像330の位置から点線領域392の位置までドラッグをする。この操作により、処理実行結果である画像310がユーザ選択データとして選ばれても、処理実行結果である画像330に対応する候補値には、得点が付与されないようにすることができる。
 なお、本実施形態では、処理対象データは撮影により得られる画像としたが、これに限らず、ベクタデータ、動画、3次元データ、ポリラインデータであっても良い。画像はグレースケール画像であっても良いし、カラー画像であっても良い。また、処理対象データは、画像を加工したデータや、画像にエッジ抽出処理を行って得られるエッジ画像であっても良い。
 また、本実施形態では、パラメータを用いた処理として、機械学習により生成したモデルを処理対象データに適用する処理を説明したが、処理の内容はこれに限らず、一般的な画像処理であっても良い。具体的には、画像処理は、具体的にはエッジ抽出処理や閾値処理がある。パラメータを用いた処理がエッジ抽出処理である場合は、注目するパラメータはフィルタサイズやフィルタの種類となる。パラメータを用いた処理が閾値処理である場合は、処理対象データはエッジ画像となり、注目するパラメータは閾値となる。
 また、本実施形態では、代表データ選択部104が代表データを選択する方法として、処理実行結果間の類似度を算出することにより類似性を調べ、類似度に基づいて代表データを選択する方法を説明したが、代表データの選択方法はこれに限定されない。
 例えば、代表データ選択部104は、処理実行結果と処理対象データとを統合した情報に基づいて代表データを選択しても良い。具体的には、処理対象データの特徴ベクトルと処理実行結果とを統合した新たな特徴ベクトル等の類似性判定や、処理対象データと処理実行結果とに関して総合的な類似性判定を行うモジュール(例えばDeep Learningのアーキテクチャ)を構築する。このようなモジュールを使用することにより、代表データを選択する。または、類似性を明示的に評価、算出せずに、複数の処理実行結果や処理対象データを入力すると、代表データを出力するモジュールを構築して使用することもできる。このように、処理実行結果のみならず処理対象データの情報も利用することにより、処理対象データ上の重要な領域に重点をおいて代表データを選択できるようになる。その他にも、代表データはランダムに決定しても良い。
 なお、本実施形態では、1つのパラメータに着目し、異なる複数の候補値を設定する方法を示したが、2つ以上のパラメータを対象とすることもできる。パラメータが1つの場合は、候補値ごとに処理実行結果を取得し、候補値ごとに得点を決定する処理を繰り返して、適切な候補値を決定した。一方、パラメータが2つある場合は、候補値のペアごとに処理実行結果を取得し、候補値のペアごとに得点を決定する処理を繰り返して、最終的に適切なパラメータのペアを決定するようにする。
 本実施形態では、図2のフローチャートにおいて、S204で調べた処理実行結果の関係性の情報を、S205の代表データ選択と、S208の得点の決定において使用する方法を示した。しかし、関係性調査結果の使用は、S208の得点の決定には必須であるが、S205の代表データ選択には必須ではないため、S204の処理は、S203からS208までの間であれば、どの段階で実行しても良い。
 また、本実施形態では、S203の処理実行結果の取得からS208の得点の決定までの工程を処理対象データごとに繰り返す方法を示したが、全処理対象データに対して一度に実施することもできる。なお、本実施形態では、S210で、集計部109が、全処理対象データの得点を決定してから最終的な集計結果を求めているが、集計は、処理対象データに得点を決定するたびに随時行っても良い。集計を随時行うことにより、上述したように途中集計結果を利用することができるようになり、結果として一連の処理の効率面や信頼性を向上できる。
 (実施形態2)
 実施形態1では、処理実行結果の関係性を類似性に基づいて調査し、ユーザ選択情報と類似性とに基づいて得点を決定する方法を説明した。これに対し、実施形態2では、処理実行結果の関係性を、処理実行結果を、データの分布によりグループに分けすることにより調べ、グループに基づいて得点を決定する方法を説明する。
 また、実施形態1では、集計部109は、候補値ごとにヒストグラム化することにより集計を行い、最高点を得た候補値を適切な値とした。これに対し、実施形態2では、集計部109は、得点を用いてパラメータ空間上における重心に位置する候補値を算出し、算出した候補値を適切な値とする方法を説明する。
 実施形態2では、処理対象データは、人の写っている画像に対してエッジ抽出することで得られるエッジ画像とする。エッジ画像は、画像と同じサイズのグレースケール画像であり、エッジ抽出を適用する前の画像上でエッジのある部分には1に近い画素値、エッジのない部分には0に近い画素値を格納した2次元データである。処理実行部103が行う処理は、閾値処理とし、注目するパラメータは、閾値とする。
 なお、本実施形態に係る情報処理装置の構成は実施形態1で図1A及び図1Bを参照して説明した構成と同様であり、処理の手順も図2のフローチャートと同様である。ただし、実施形態1では、関係性を調査する際に類似度を算出したが、実施形態2ではグループに関する情報を求める点で異なっている。
 <各構成要素の処理内容>
 実施形態2の各構成要素の処理内容について説明する。処理対象データ入力部101は、人の写った画像群に対してエッジ抽出を行って得られるエッジ画像のデータ群を処理対象データとして入力する。パラメータの候補値には異なる閾値を設定し、それぞれα1、α2、α3、α4、α5とする。
 処理実行部103は、処理対象データに対して、一通りの候補値を用いて閾値処理をする。図3Aの画像310、320、330、340、350は、候補値α1、α2、α3、α4、α5に対応する処理実行結果の例である。
 代表データ選択部104は、複数の候補値により得られた複数の処理実行結果をグループ化し、各グループから1つの処理実行結果を代表データとして選択する。グループ化は、処理対象データごとに、異なる候補値により得られた処理実行結果をクラスタリングすることにより実行する。クラスタリングとは、データ群をデータ分布に基づき複数のグループに分類する方法である。
 1つの処理対象データimage1におけるグループ化による関係性調査結果の例を図4Cに示す。g1、g2は処理実行結果が所属するグループを意味するものとする。この表は、グループ化によりα1、α2、α3はグループg1に属し、α4、α5はグループg2に分類されていることを示している。
 各グループから代表データを選択する方法には、処理実行結果に基づき選択する方法、パラメータの候補値に基づき選択する方法がある。処理実行結果に基づき選択する方法では、グループ分けした空間上においてグループの中心に最も近い処理実行結果を代表データとする。
 候補値に基づき選択する方法は2つある。1つ目は、グループごとに候補値の平均値を求め、平均値に近い候補値の処理実行結果を代表データとする方法である。2つ目は、平均値と一致する値が候補値の中に存在しないとき、その平均値をパラメータの値として処理実行結果を取得し直して代表データとする方法である。その他、各グループから代表データを選択するときには、ランダムに選択しても良い。実施形態2では、代表データ選択部104が、グループg1、g2の代表データとして、それぞれ候補値α1、α5の処理実行結果を選択したものとする。
 表示制御部105が代表データを表示して、操作部106によるユーザ操作を行った後、得点決定部108が各候補値に得点を付与する。実施形態2では、処理実行結果がユーザ選択データのグループと同じであるかどうかに基づき得点を決定する。得点の付与方法としては、ユーザ選択データの属するグループには均等の得点を付ける方法がある。例えば、ユーザ選択データとしてα5の処理実行結果が選ばれたとする。図4Cの関係性調査結果の処理対象データimage1に着目すると、ユーザ選択データの属するグループg2には候補値α4、α5が含まれている。このため、図4Dに示す得点テーブルにおいて、点線領域450に示すようにα4、α5に得点を均等に付与する。
 その他の得点の付与方法として、ユーザ選択データの属するグループの中で、ユーザ選択データと処理実行結果との類似性を調べ、該当グループの候補値のみ得点として類似度を付与し、他のグループの候補値の得点は0点としても良い。
 次に、集計部109について説明する。実施形態2における集計部109は、得点を用いてパラメータ空間上における重心に位置する候補値を調べて適切な値とする。各候補値が取得した得点の合計値を求めて、合計値と候補値を用いてパラメータ空間上での重心を求め、重心をパラメータの適切な値とみなす。具体的には、図4Dの点線領域460に示すように、候補値ごとに合計得点を求め、その後、式(1)を用いて候補値の重心を求めることにより、適切な値を決定する。
Figure JPOXMLDOC01-appb-M000001
 ただし、xiは各候補値の合計得点、αiは候補値とし、添え字iは候補値ごとに異なる値をとるものとする。また、nは候補値の数とする。このように、重心を求める処理は、パラメータ設定部102が設定する設定値に適切なパラメータが必ずしも含まれているとは限らないことを考慮した処理である。重心から適切な値を決めることで、より適切な候補値を見つけることができる可能性がある。
 以上のように、本実施形態では、関係性として処理実行結果のグループを調べ、グループに基づいて得点を決定し、集計の際に重心を利用する。これにより、より適切な候補値を見つけられる可能性を高めることができる。
 (実施形態3)
 実施形態1及び実施形態2では、処理対象データごとに代表データを一度だけ表示していた。これに対し、実施形態3では、処理対象データごとに代表データの表示を複数回行う。具体的には、代表データが3つ以上ある場合に、表示する代表データの組み合わせを変更するとユーザによる選択の結果が変化することを考慮し、処理対象データごとに組み合わせを変更しながら代表データを表示する。実施形態3では、特に、処理対象データごとに得られる複数のユーザ選択情報を分析して、候補値の順位に関する情報を取得し、順位に基づいて得点の決定を行う方法を説明する。
 実施形態3では、ユーザが主観的に美しいと感じるパラメータを決定することを目的として、処理対象データには景色や人物の写った画像(例えば640pixcel×480pixcelのサイズの画像)を扱う。処理実行部103が実施する処理は、明るさ変換処理とし、注目するパラメータは、明るさ変換パラメータとする。
 <情報処理装置の機能構成>
 図5は、実施形態3に係る情報処理装置500の機能構成を示す図である。実施形態3では、実施形態1で図1A及び図1Bを参照して説明した情報処理装置100の構成に、ユーザ選択情報分析部510が追加されている。
 ユーザ選択情報分析部510は、ある処理対象データに対する複数のユーザ選択情報をユーザ選択情報保持部110から受け取り、ユーザ選択情報に基づいてユーザ選択情報分析結果を取得して、ユーザ選択情報分析結果を得点決定部108に送る。得点決定部108は、ユーザ選択情報分析部510からユーザ選択情報分析結果を受け取ると、ユーザ選択情報分析結果と処理実行結果との関係性に基づいて各候補値の得点を決定する。
 <処理フロー>
 図6を用いて、実施形態1に係る情報処理装置500が実施する処理の手順を説明する。S201で処理対象データ入力部101が処理対象データを読み込んでから、S204で統計処理部107が処理実行結果の関係性を調査するまでは、実施形態1及び2と同様である。S205において、代表データ選択部104は、実施形態1、2では代表データを2つ選択していたのに対し、実施形態3では代表データを3つ以上選択する。
 続いて、S601では、表示制御部105は、代表データから表示画面に表示するデータの組み合わせを1つ選ぶ。S206において、代表データ選択部104は、代表データを表示することによりユーザに好ましいデータを選択するよう促す。S207において、操作部106は、ユーザの選択操作を受け付ける。S602において、ユーザ選択情報保持部110は、ユーザ選択情報を記録する。
 S603において、表示制御部105は、代表データの組み合わせとしてあり得るものを全て選択して表示したか否かを判定する。全組み合わせを表示していない場合は、S601の代表データから組み合わせを選択する工程に戻る。一方、全組み合わせを表示し終えている場合は、S604へ進む。
 S604において、ユーザ選択情報分析部510は、ユーザ選択情報を分析する。S605において、得点決定部108は、ユーザ選択情報分析結果と処理実行結果との間の関係性に基づいて、1つの処理対象データについて、各候補値の得点を決定する。
 続いて、S209において、得点決定部108は、全処理対象データについて各候補値の得点を決定済みであるか否かを判定する。決定済みでない場合は、S203に戻り、処理実行部103が次の処理対象データに処理を適用して処理実行結果を取得する。一方、決定済みである場合は、S210へ進む。
 S210において、集計部109は、得点の集計を行う。S211において、集計部109は、パラメータの適切な値を決定する。S212において、表示制御部105が、決定されたパラメータについての確認画面を表示する。
 <各構成要素の処理内容>
 実施形態3における各構成要素の処理内容について説明する。まず、処理対象データ入力部101は、複数の画像を処理対象データとして取得する。次に、パラメータ設定部102は、明るさに関するパラメータの設定値を複数設定する。ここでは、候補値として、α1、α2、α3、α4、α5を設定したとする。次に、処理実行部103は、処理対象データimage1に対してα1からα5の候補値を用いて明るさ変換を行い、複数の処理実行結果を取得する。
 続いて、処理実行部103は、処理実行結果間の関係性の調査のため、実施形態2と同様にクラスタリングにより処理実行結果のグループ化を行い、結果として図7Aに示す関係性調査結果が得られたとする。図7Aの表は、各候補値の属すグループを表しており、α1、α2の処理実行結果はグループg1に属し、α3、α4の処理実行結果はグループg2に属し、α5の処理実行結果はグループg3に属していることを示している。
 次に、代表データ選択部104が、関係性の調査結果に基づいて、各グループから代表データをランダムに選択する。ここでは、α1、α3、α5の処理実行結果が選択されたものとする。
 表示制御部105は、候補値α1、α3、α5の処理実行結果のうち、表示する組み合わせを変更して表示する。表示を確認したユーザが、操作部106を用いてユーザ選択データの選択操作を行う。
 図7Bの表の各行は、表示する代表パラメータの組み合わせを変更したときにユーザが選択した結果を表している。ユーザ選択データの候補値には1、それ以外の候補値には0、表示しなかった場合はNULLを格納する。図7Bは、α1とα3、α1とα5、α3とα5の組み合わせをそれぞれ表示したときに、それぞれα1、α1、α3が選ばれたことを示している。
 ユーザ選択情報分析部510は、処理対象データimage1に関するユーザ選択情報を参照してパラメータの候補値ごとに集計を行うことで、ユーザ選択情報分析結果として、候補値の順位に関する情報を求める。実施形態3における順位に関する情報の例を図7Cに示す。この例では、代表データの組み合わせを変更した3回の試行のうち、α1が2回、α3が1回、α5が0回ユーザから選択されていることから、α1、α3、α5に、それぞれ2点、1点、0点が格納されている。
 得点決定部108は、処理実行結果の関係性と、ユーザ選択情報分析結果として得られた順位に関する情報とに基づいて、得点の決定を行う。実施形態3では、図7A及び図7Cに基づいて、図7Dに示す得点テーブルを作成する。まず、図7Cの順位に関する情報に従って、α1、α3、α5の得点に差をつける。さらに、図7Aの関係性調査結果に基づいて、同じグループの候補値には同じ得点を付与する。
 なお、この例では、図7Cのように順位に関する情報において候補値間で差があるが、差がない場合には、得点決定部108は、その処理対象データについては得点を付与しないようにすることもできる。
 このように、順位に関する情報を用いることで、表示する代表データの組み合わせを変更することによりユーザによる選択の結果が変化することを考慮した得点付与を行うことができる。
 以上の処理を全処理対象データに対して繰り返し実行し、集計部109が集計を行ってパラメータの適切な値を決定する。
 以上説明したように、本実施形態では、ユーザ選択情報分析部510が順位に関する情報を調べて、得点決定部108で反映する方法を示した。
 なお、本実施形態では、図6のフローチャートにおいて、表示の際に、1つの処理対象データについて、代表データの組み合わせを変更して続けて表示しているが、これに限定されない。ある処理対象データから得られた代表データの組み合わせを複数回変更して表示する際に、他の処理対象データから得られた代表データの表示を挟んでも良い。
 (実施形態4)
 実施形態1から3では、1人のユーザを対象としていたが、実施形態4では、複数のユーザを対象とする例を説明する。実施形態4では、実施形態3と同様に、ユーザ選択情報分析部510が取得するユーザ選択情報分析結果を利用して得点を決定する。ユーザ選択情報分析結果として、実施形態3では順位に関する情報を利用したのに対し、実施形態4では後述の信頼性に関する情報を利用する。信頼性に関する情報を用いることで、ユーザの慣れの度合い(熟練度)の得点への影響を小さくすることが可能となる。
 実施形態4で扱うデータは、インフラ構造物の大きな壁面画像とする。実施形態4では、壁面上の浮きなどの変状部分を検知した結果を取得する上で、適切なパラメータを決定することを目的とするものとする。
 実施形態4では、処理対象データを取得するための処理を事前に行うことを前提とするため、まずその処理内容について説明する。まず、変状部分を検知するためのモデルを生成しておき、そのモデルを壁面画像に適用することで、壁面画像と同じサイズのデータであって、変状らしい画素に1に近い値、そうでない画素に0に近い値が格納されたスコアマップを取得する。その後、スコアマップを用いてポリラインデータを作成し、ポリラインデータを処理対象データとする。
 ポリラインデータとは、位置情報から構成されるデータである。ポリラインの作成方法としては、スコアマップに対して2値化して細線化処理を施した後、さらにベクタ化処理を実施することで作成することができる。実施形態4における、ポリラインデータの例を図9Aの910に示す。線分911、912、913、914、915はポリラインデータを表しており、変状らしい部分の位置を示している。この例のように、ポリラインデータは、隣接する点を繋ぐ線分で表現されても良いし、位置情報の他に数式情報を持たせておき、曲線で表現されても良い。
 実施形態4では、処理実行部103における処理として、ポリラインデータのうち誤検知の可能性のあるポリラインを除去する。具体的には、極端に短いポリラインは、ひび割れや浮きではなく、壁面上の穴や傷を誤検知している可能性があると考えられるため、削除する。注目するパラメータは、除去の基準とするポリラインの長さとする。
 実施形態4に係る情報処理装置の構成は、実施形態3で図5を参照して説明した情報処理装置500と同様であるため、説明は省略する。
 <処理フロー>
 まず、図6を用いて、複数のユーザを対象とする実施形態4に係る情報処理装置500が実施する処理の手順を説明する。S201で処理対象データ入力部101が処理対象データを読み込む工程から、S207で操作部106がユーザの選択操作を受け付ける工程までの処理は、実施形態1と同様である。S207の処理の後、S602へ進む。S602において、ユーザ選択情報保持部110は、ユーザ選択情報を記録し、S801へ進む。
 S801において、処理実行部103は、全処理対象データについてS203からS602までの処理を実施したか否かを確認する。実施していない場合は、S203に戻り、実施している場合は、S802へ進む。S802において、処理実行部103は、全てのユーザに対してS203からS801までの処理を実施したか否かを確認する。実施していない場合は、S203に戻り、まだ実施していないユーザを対象に、S203からS801までの処理を実行する。一方、実施できている場合は、S604へ進む。
 S604において、ユーザ選択情報分析部510は、ユーザ選択情報を分析する。S605において、得点決定部108は、処理実行結果の関係性とユーザ選択情報分析結果とに基づいて、パラメータの各候補値に得点を付与する。
 そして、S209において、得点決定部108は、全処理対象データについて得点の決定を終えたか否かを判定する。終えていない場合は、S604に戻る。一方、終えている場合は、S210へ進む。S210において、集計部109は、得点決定部108により決定された得点を用いて全処理対象データについて集計を行う。
 S211において、集計部109は、集計結果に基づいてパラメータの適切な値を決定し、パラメータ設定部102へ出力する。S212において、表示制御部105は、集計結果に基づいて決定されたパラメータの適切な値を確認画面に表示し、一連の処理を終了する。
 <各構成要素の処理内容>
 実施形態4の各構成要素の処理内容について、図9A~図9C及び図10A~図10Dを用いて説明する。まず、処理対象データ入力部101が処理対象データを入力する。例えば、変状を検知するためのモデルの適用や、細線化やポリライン化などの画像処理によって、大きな壁面画像から元の壁面画像と同じサイズの2次元空間にあるポリラインデータを取得する。ここでは、大きな壁面画像の空間から、500pixcel×500pixcel程度のサイズの小領域を複数切り出し、処理対象データとする。
 パラメータ設定部102は、注目するパラメータを、ポリラインデータを除去する基準のポリラインの長さとし、候補値はα1、α2、α3、α4、α5とする。
 処理実行部103は、処理対象データと候補値とを受け取り、ポリラインデータのうち短いポリラインを除去する処理を行う。処理実行結果の例を図9Bに示す。画像920、930、940、950、960は、候補値α1、α2、α3、α4、α5に対応する処理実行結果を表してる。画像920から画像960に向かうにつれて、除去する基準のポリラインの長さを徐々に大きくしたときの処理実行結果を表している。
 その後、代表データ選択部104は、代表データを選択する。実施形態4では、処理実行結果の関係性として、実施形態2と同様に、処理実行結果のグループ化を行い、結果として、図10Aに示す情報が得られたとする。図10Aの表は、各候補値の属するグループを表しており、α1、α2、α3はグループg1に属し、α4、α5はグループg2に属していることを示している。ここでは、各グループからランダムに、代表データとして、候補値α1とα5の処理実行結果が選択されたとする。
 次に、複数のユーザに対して、表示制御部105が、候補値α1とα5の処理実行結果を表示し、操作部106によりユーザ選択データを選択させる操作を実施する。表示と操作のタイミングは、ユーザ間でそろえる必要はない。ユーザから操作を受けた後、ユーザ選択情報保持部110がユーザ選択情報を格納する。図10Bに、処理対象データimage1に関するユーザ選択情報の例を示す。表の各行は、異なるユーザに代表データを表示したときの選択の結果を表しており、各ユーザが選んだ代表データには1、選ばなかった代表データには0を格納している。この例では、ユーザ1、2はα1の方を選んでおり、ユーザ3はα5の方を選んでいる。
 次に、ユーザ選択情報分析部510は、ユーザ選択情報分析結果として、処理対象データimage1に関する信頼性に関する情報を取得する。信頼性に関する情報は、各ユーザの選択操作への慣れの度合いを考慮した候補値ごとの得点である。具体的には、慣れているユーザの選択結果には高い重みを付け、慣れていないユーザの選択結果には低い重みを付けて集計することにより求める。慣れているユーザと慣れていないユーザの決定方法として、予め別の処理対象データ群を用いて複数のユーザによるユーザ選択情報を取得しておき、他のユーザと比べて選択の内容に外れ値の多いユーザを慣れていないユーザとみなす方法がある。実施形態4では、3名のユーザを対象とし、ユーザ1、2は慣れているものとし、ユーザ3は慣れていないものとする。
 以下に、各候補値の信頼性に関する情報を算出するための式を示す。
Figure JPOXMLDOC01-appb-M000002
 ただし、mはユーザの数、xjはユーザが選択したかどうかに関する情報で0か1をとるものとする。wjはユーザごとの慣れの度合いを考慮した重みとし、添え字jはユーザごとに異なる値をとる。ここでは、慣れているユーザ1、2の重みを1、慣れていないユーザ3の重みを0.5点とし、信頼性に関する情報を求めた例を図10Cに示す。図10Cにおいて、α1は慣れているユーザ2名が選択したことから2点、α5は慣れていないユーザ1名が選択したことから0.5点となっている。信頼性に関する情報を用いて得点の決定をすることで、ユーザの慣れの度合いの得点への影響を軽減することができる。
 得点決定部108は、処理実行結果の関係性と、信頼性に関する情報とに基づいて得点を決定する。ここでは、図10A及び図10Cに基づいて、図10Dに示す得点テーブルを作成する。まず、図10Cの信頼性に関する情報に基づいてα1の得点は2点、α5の得点は0.5点とする。さらに、図10Aの関係性調査結果に基づき同じグループの候補値α2にはα1と同じ得点を付与する。同様に、同じグループの候補値α3、α4にはα5と同じ得点を付与する。
 以上の処理を全処理対象データに繰り返し実行し、集計部109が集計することで、パラメータの適切な値を決定する。
 パラメータの適切な値が決まった後は、表示制御部105が確認画面にカバー範囲を表示する。カバー範囲とは、大きな画像を扱うときに表示できる情報であり、ある候補値によりユーザにとって好ましい処理実行結果が得られている画像領域の範囲である。具体的には、得点テーブルのうち、集計部109が適切な値として決定した候補値を用いたときに各処理対象データで得られた得点に着目し、所定値以上の得点が得られた処理対象データを抽出する。
 そして、全体の大きな画像に対して抽出した処理対象データの占める範囲を描写する。図9Cに、確認画面970の例を示す。点線領域971は、適切な値として決定したパラメータを表示している。表示画像970の右側はカバー範囲の例であり、大きな画像972の中の領域973は、決定されたパラメータによるカバー範囲を表している。カバー範囲を表示することにより、決定されたパラメータが大きな画像の中で、どの範囲に適していているかを、ユーザは容易に確認できるようになる。
 このように、実施形態4では、ユーザ選択情報分析部510が信頼性に関する情報を取得して、得点決定部108が得点に反映することで、パラメータのより適切な値を決定することができる。
 なお、上述の実施形態3、4では、ユーザ選択情報分析部510で調べたユーザ選択情報分析結果として、順位に関する情報、信頼性に関する情報をそれぞれ利用する例を説明した。一方、ユーザ選択情報分析結果として、その他にも、代表データを表示してからユーザが選択操作を行うまでの時間情報を利用することができる。この場合、ユーザ選択情報として、代表データを表示してからユーザ選択情報を取得するまでの時間を記録する必要がある。表示してからユーザが選択を行うまでの時間が長い場合は、短い場合に比べ、ユーザがどちらの方が好ましいか判断に悩んだためであると解釈し、得点決定部108が低い得点を付与する。このように、時間情報を利用することにより、得点の質を高めることができる。
 (実施形態5)
 実施形態1から4までは、全処理対象データに関して、表示制御部105が表示装置1007上に表示を行って、ユーザに選択操作を促していた。しかし、代表データの組み合わせによっては、代表データ同士が良く似ていて、どちらが好ましいかを目視では判断しにくい場合がある。このようなデータを表示するときは、いずれも良好などの判断に迷うケースに備えた選択肢(代表データの良否に関する選択肢)を用意して選択させれば良い。しかし、人が判断できないようなデータは重要でないため、毎回表示するのはユーザにとって煩わしい可能性がある。このため、処理対象データから得る代表データを全て表示するのではなく、人が判断に迷うようなデータは表示しないことが望ましい。
 実施形態5では、ユーザ操作を反映した関係性調査結果を用いて、表示制御部105が代表データを表示する基準を、統計的に決定する例を説明する。本実施形態では、実施形態1、2と同様に、人の写った画像を扱うものとし、人の輪郭を抽出することを目的として、処理対象データは、エッジ抽出したエッジ画像とする。また、処理実行部103における処理の内容は閾値処理とし、注目するパラメータは閾値とする。
 <情報処理装置の機能構成>
 図11は、実施形態5に係る情報処理装置1100の機能構成を示す図である。実施形態3では、実施形態1で図1A及び図1Bを参照して説明した情報処理装置100の構成に、基準決定部1101が追加されている。
 基準決定部1101は、ユーザ選択情報保持部110に保持されているユーザ選択情報を利用して、ユーザに提示するデータの基準を決定する。基準決定部1101により決定された基準に基づいて、表示制御部105は、代表データを受け取った後、表示装置1007に表示する否かを判定し、その判定結果に応じて、表示すべき代表データのみを表示する。
 <各構成要素の処理内容>
 基準決定部1101が基準を決定するタイミングは、実施形態1の図2に示したような一連の処理を行う前である。まず、表示制御部105が出力する表示内容について、図12Aの画像1210を用いて説明する。画像1210は表示画面の例であり、関係性として類似性を調べるケースにおける表示方法を説明する。表示画像1210に示す画像1215、1216、1217、1218、1219は、ある処理対象データに対して、異なる候補値を用いて処理実行部103が処理を適用した処理実行結果である。
 点線領域1211、1212、1213は、処理実行結果を関係性に基づいてグループ化した結果を表示している。点線領域1211に含まれる処理実行結果1215、1216、1217はグループg1に属することを意味しており、点線領域1213に含まれる処理実行結果1218、1219はグループg2に属することを意味している。点線領域1212は、グループ化したとき分類できなかった処理実行結果のために用意してある領域である。
 ユーザは、表示画面の分類を見て、分類が目視上の印象と異なる場合は、操作部106を用いて処理実行結果の所属を変更する。例えば、ユーザが処理実行結果1217の所属をグループg1から分類不可に変更したい場合、ユーザは処理実行結果1217にマウスポインタ1214をかざし、矢印1215の方向にドラッグすることで、所属をg1から分類不可に変更する。
 ユーザが表示画面の分類を見て、いずれの処理実行結果も似ていて差異がないと判断した場合は、同様にドラッグ操作により全ての処理実行結果の所属を分類不可に変更することもできる。または、表示画面に分類不可などの旨を示す選択肢を用意し、ユーザに選択させても良い。
 ユーザ操作を反映した関係性の例を図12Bに示す。図12Bの表は、処理対象データごとの候補値の処理実行結果の分類を表している。処理対象データimage1において、候補値α1、α2、α4、α5には所属するグループを格納し、α3の処理実行結果はユーザが所属を分類不可に変更したことから、グループではなくNULLを格納している。
 次に、基準決定部1101について説明する。基準決定部1101は、関係性調査結果と、代表データと、ユーザ選択情報とに基づいて、表示制御部105が表示する代表データの基準を統計的に決定する。
 基準決定部1101は、まず、ユーザの目視の結果、異なるグループに属することが確認された処理実行結果を用いて、類似性を調べる。類似性の調査は、処理対象データごとに関係性調査結果を示す図12Bを参照しながら行う。処理実行結果がグループ分けされている場合、異なるグループに属する処理実行結果を1つずつ選び、処理実行結果間の距離を算出する処理を、全組み合わせに対して繰り返し実施する。そして、処理対象データごとに距離の最小値を、図12Cに示す基準分析テーブルに記録する。距離は、2つの処理実行結果において位置が対応する画素の差分の二乗和を算出することで求めれば良い。関係性調査結果において、全候補値にNULLしか設定されていない場合、距離の算出は行わず、次の処理対象データに移る。図12Bの処理対象データimage1においては、グループの情報g1、g2が格納されていることから、異なるグループの処理実行結果の距離を、組み合わせを変更しながら算出し、図12Cに示す基準分析テーブルに距離の最小値を格納する。
 処理対象データを変更しながら距離の最小値の記録を繰り返し、ユーザに表示する基準を決定する上で十分な量のデータを収集できたら、基準を統計的に決定する。具体的には、距離が求められている処理対象データの情報のみ抽出して、距離の平均値を基準とする。
 基準が決まった後の処理内容は、図2を参照して説明した処理と同様である。表示制御部105は、代表データ選択部104から新たに受け取った代表データ間の距離を算出する。そして、その距離と、基準決定部1101により決定された基準値とを比較して、距離が基準値以上であれば、ユーザにとって差異のあるデータであると判定して表示画面に表示する。反対に、距離が基準値未満であれば、その代表データは表示しない。
 このように、基準決定部1101により表示する代表データの基準を決定することで、表示する代表データを限定することができるため、効率的に表示と選択を行うことができる。
 なお、上記の説明では、図12Aの画像1210のような表示により、基準を決定するために事前に情報を収集したが、必ずしも事前に行う必要はない。上述の各実施形態で示した通り、処理対象データを変更しながら代表データ選択部104、表示制御部105、操作部106の処理を繰り返す中で基準を決定しても良い。
 具体的には、表示画面に、代表データの他に、いずれも良好、いずれも不良、判断不可、という判断に迷うケースに備えた選択肢(代表データの良否に関する選択肢)を表示する。そして、それらが選択された場合を除き、代表データが選択された場合のみ、代表データ間の距離を算出して基準分析テーブルに格納していく。表示を終えた処理対象データについて算出した距離の平均値を求めて基準とすることで、表示を繰り返しながら基準を更新していくことができる。これにより、基準を決定するために事前に情報収集する方法と比較すると、よりユーザの作業を少なくすることができる。
 (その他の実施形態)
 本発明は、上述の実施形態の1以上の機能を実現するプログラムを、ネットワーク又は記憶媒体を介してシステム又は装置に供給し、そのシステム又は装置のコンピュータにおける1つ以上のプロセッサーがプログラムを読出し実行する処理でも実現可能である。また、1以上の機能を実現する回路(例えば、ASIC)によっても実現可能である。
 発明は上記実施形態に制限されるものではなく、発明の精神及び範囲から離脱することなく、様々な変更及び変形が可能である。従って、発明の範囲を公にするために請求項を添付する。
 本願は、2019年4月9日提出の日本国特許出願特願2019-074269を基礎として優先権を主張するものであり、その記載内容の全てを、ここに援用する。

Claims (20)

  1.  情報処理装置であって、
     所定の画像処理に係るパラメータの複数の候補値を用い、対象データに対して前記所定の画像処理を実行した結果を表す複数の画像を取得する画像処理手段と、
     前記複数の画像から、複数の代表画像を選択する選択手段と、
     選択された前記複数の代表画像を表示手段に表示させる表示制御手段と、
     表示された前記複数の代表画像に対して入力される操作を受け付ける操作手段と、
     受け付けられた前記操作によって得られた情報の統計処理の結果に基づいて、前記複数の候補値から、少なくとも1つの値を決定する統計処理手段と、
     を備えることを特徴とする情報処理装置。
  2.  前記統計処理手段は、
      前記操作によって得られた情報と、前記複数の画像とに基づいて、前記複数の候補値の各々について得点を決定する得点決定手段と、
      前記得点の集計結果に基づいて、前記複数の候補値から前記少なくとも1つの値を決定する集計手段と、
     を備えることを特徴とする請求項1に記載の情報処理装置。
  3.  前記得点決定手段は、前記操作によって得られた情報と、前記複数の候補値を用いた前記所定の画像処理の結果との関係性に基づいて、前記複数の候補値の各々について得点を決定することを特徴とする請求項2に記載の情報処理装置。
  4.  前記得点決定手段は、前記操作手段により操作が受け付けられた代表画像と、前記複数の候補値を用いた前記所定の画像処理の結果を表す前記複数の画像との類似度を算出し、前記類似度に基づいて前記得点を決定することを特徴とする請求項3に記載の情報処理装置。
  5.  前記選択手段は、前記複数の候補値を用いた前記所定の画像処理の結果のデータ分布に基づいて前記複数の画像をグループ化することにより前記関係性を取得し、
     前記得点決定手段は、各画像が、前記操作により選択された代表画像が属するグループと同一かどうかに基づいて前記得点を決定することを特徴とする請求項3に記載の情報処理装置。
  6.  前記集計手段は、前記候補値ごとにヒストグラム化することにより前記得点の集計を行うことを特徴とする請求項2乃至5の何れか1項に記載の情報処理装置。
  7.  前記表示制御手段は、全ての対象データについて前記得点が決定されていない場合に前記集計手段により集計された途中集計結果を示す情報を前記表示手段にさらに表示させることを特徴とする請求項2乃至6の何れか1項に記載の情報処理装置。
  8.  全ての対象データについて前記得点が決定されていない場合に前記集計手段により集計された途中集計結果に基づいて、前記候補値を変更する変更手段をさらに備えることを特徴とする請求項2乃至7の何れか1項に記載の情報処理装置。
  9.  前記表示制御手段は、全ての対象データについて前記得点が決定されていない場合に前記集計手段により集計された途中集計結果において、前記得点が所定値以下である候補値に対応する代表画像が前記操作により選択された場合、前記複数の代表画像を前記表示手段に再表示させることを特徴とする請求項2乃至7の何れか1項に記載の情報処理装置。
  10.  前記表示制御手段は、前記集計手段により決定された候補値と、前記得点に関する情報と、前記集計手段により決定された値に基づいて所定値以上の得点が決定された対象データ数の全対象データに対する比を示すカバー率との少なくとも1つを前記表示手段に表示させることを特徴とする請求項2乃至9の何れか1項に記載の情報処理装置。
  11.  前記表示制御手段は、各対象データが1つの画像を構成する小領域のデータである場合に、前記集計手段により決定された候補値に基づいて所定値以上の得点が決定された対象データの、前記画像におけるカバー範囲を前記表示手段に表示させることを特徴とする請求項2乃至9の何れか1項に記載の情報処理装置。
  12.  前記表示制御手段は、前記複数の代表画像とともに、代表画像の良否に関する選択肢を前記表示手段に表示させ、
     前記得点決定手段は、前記選択肢の選択結果と、前記関係性とに基づいて前記得点を決定することを特徴とする請求項3に記載の情報処理装置。
  13.  各対象データについて前記操作によって得られた情報に基づいて情報分析結果を取得する分析手段をさらに備え、
     前記得点決定手段は、前記情報分析結果と、前記関係性とに基づいて前記得点を決定することを特徴とする請求項3に記載の情報処理装置。
  14.  前記分析手段は、前記複数の代表画像が表示されてからユーザが前記操作を行うまでの時間の情報に基づいて、前記情報分析結果を取得することを特徴とする請求項13に記載の情報処理装置。
  15.  前記分析手段は、各対象データについて得られるユーザ選択情報を分析して、前記複数の候補値の順位に関する情報を、前記情報分析結果として取得することを特徴とする請求項13に記載の情報処理装置。
  16.  前記分析手段は、1以上のユーザによる前記操作の結果に基づいて判定される各ユーザの熟練度に基づいて、各対象データに対する信頼性に関する情報を、前記情報分析結果として取得することを特徴とする請求項13に記載の情報処理装置。
  17.  情報処理装置であって、
     構造物を撮影した画像から前記構造物に生じたひび割れを検知した結果を、画素ごとのひび割れらしさに基づくポリラインデータとして取得する取得手段と、
     前記ポリラインデータの少なくとも一部に対して、ポリラインの長さの違いであるパラメータの複数の候補値を用い、候補値より短いポリラインを削除した結果を表す複数の画像を取得する処理手段と、
     前記複数の画像の一部の画像を表示手段に表示させる表示制御手段と、
     表示された前記画像に対して入力される操作を受け付ける操作手段と、
     前記表示手段に表示された画像が異なる複数の状態で受け付けられた前記操作によって得られた情報に基づいて、前記複数の候補値のうち少なくとも1の値を決定する決定手段と、
     を有することを特徴とする情報処理装置。
  18.  情報処理装置の制御方法であって、
     所定の画像処理に係るパラメータの複数の候補値を用い、対象データに対して前記所定の画像処理を実行した結果を表す複数の画像を取得する画像処理工程と、
     前記複数の画像から、複数の代表画像を選択する選択工程と、
     選択された前記複数の代表画像を表示手段に表示させる表示制御工程と、
     表示された前記複数の代表画像に対して入力される操作を受け付ける操作工程と、
     受け付けられた前記操作によって得られた情報の統計処理の結果に基づいて、前記複数の候補値から、少なくとも1つの値を決定する統計処理工程と、
     を有することを特徴とする情報処理装置の制御方法。
  19.  情報処理装置の制御方法であって、
     構造物を撮影した画像から前記構造物に生じたひび割れを検知した結果を、画素ごとのひび割れらしさに基づくポリラインデータとして取得する取得工程と、
     前記ポリラインデータの少なくとも一部に対して、ポリラインの長さの違いであるパラメータの複数の候補値を用い、候補値より短いポリラインを削除した結果を表す複数の画像を取得する処理工程と、
     前記複数の画像の一部の画像を表示手段に表示させる表示制御工程と、
     表示された前記画像に対して入力される操作を受け付ける操作工程と、
     前記表示手段に表示された画像が異なる複数の状態で受け付けられた前記操作によって得られた情報に基づいて、前記複数の候補値のうち少なくとも1の値を決定する決定工程と、
     を有することを特徴とする情報処理装置の制御方法。
  20.  コンピュータを、請求項1乃至17の何れか1項に記載の情報処理装置として機能させるためのプログラムを格納したコンピュータ読み取り可能な記憶媒体。
PCT/JP2020/007317 2019-04-09 2020-02-25 情報処理装置、情報処理装置の制御方法及びプログラム WO2020208955A1 (ja)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/492,708 US20220019848A1 (en) 2019-04-09 2021-10-04 Information processing apparatus, control method for information processing apparatus, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2019-074269 2019-04-09
JP2019074269A JP7418158B2 (ja) 2019-04-09 2019-04-09 情報処理装置、情報処理方法及びプログラム

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/492,708 Continuation US20220019848A1 (en) 2019-04-09 2021-10-04 Information processing apparatus, control method for information processing apparatus, and storage medium

Publications (1)

Publication Number Publication Date
WO2020208955A1 true WO2020208955A1 (ja) 2020-10-15

Family

ID=72752002

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/007317 WO2020208955A1 (ja) 2019-04-09 2020-02-25 情報処理装置、情報処理装置の制御方法及びプログラム

Country Status (3)

Country Link
US (1) US20220019848A1 (ja)
JP (1) JP7418158B2 (ja)
WO (1) WO2020208955A1 (ja)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344550A (zh) * 2021-06-30 2021-09-03 西安力传智能技术有限公司 一种流程处理方法、装置、设备及存储介质

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015198300A (ja) * 2014-03-31 2015-11-09 キヤノン株式会社 情報処理装置、撮像装置、画像管理システム

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5705711B2 (ja) * 2011-12-02 2015-04-22 大成建設株式会社 ひび割れ検出方法
SG11201709275YA (en) * 2015-05-26 2017-12-28 Mitsubishi Electric Corp Detection apparatus and detection method
JP6723909B2 (ja) * 2016-12-09 2020-07-15 キヤノン株式会社 画像処理方法、画像処理装置、及びプログラム

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015198300A (ja) * 2014-03-31 2015-11-09 キヤノン株式会社 情報処理装置、撮像装置、画像管理システム

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
J. MARKS ET AL.: "Design Galleries: A General Approach to Setting Parameters for Computer Graphics and Animation", SIGGRAPH' 97 PROCEEDINGS OF THE 24TH ANNUAL CONFERENCE ON COMPUTER GRAPHICS AND INTERACTIVE TECHNIQUES, ASSOCIATION FOR COMPUTING MACHINERY, 1 October 1997 (1997-10-01), pages 389 - 400, XP000979805, ISBN: 878-0-89791-896-1 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113344550A (zh) * 2021-06-30 2021-09-03 西安力传智能技术有限公司 一种流程处理方法、装置、设备及存储介质
CN113344550B (zh) * 2021-06-30 2023-11-28 西安力传智能技术有限公司 一种流程处理方法、装置、设备及存储介质

Also Published As

Publication number Publication date
US20220019848A1 (en) 2022-01-20
JP7418158B2 (ja) 2024-01-19
JP2020173554A (ja) 2020-10-22

Similar Documents

Publication Publication Date Title
CN109791693B (zh) 用于提供可视化全切片图像分析的数字病理学系统及相关工作流程
WO2020224403A1 (zh) 分类任务模型的训练方法、装置、设备及存储介质
Komura et al. Machine learning methods for histopathological image analysis
US11176415B2 (en) Assisted image annotation
KR101346539B1 (ko) 얼굴들을 상관시킴으로써 디지털 이미지들을 구조화하기
JP4037869B2 (ja) 画像解析支援方法、画像解析支援プログラムおよび画像解析支援装置
JP5283088B2 (ja) 画像検索装置および同画像検索装置に適用される画像検索用コンピュータプログラム
TWI254891B (en) Face image detection method, face image detection system, and face image detection program
CN105917354A (zh) 用于图像处理的空间金字塔池化网络
EP1583023B1 (en) Model of documents and method for automatically classifying a document
CN108762740A (zh) 页面数据的生成方法、装置及电子设备
JP2018206252A (ja) 画像処理システム、評価モデル構築方法、画像処理方法及びプログラム
CN111916206A (zh) 一种基于级联的ct影像辅助诊断系统
WO2020208955A1 (ja) 情報処理装置、情報処理装置の制御方法及びプログラム
EP3837664A1 (en) Generating annotation data of tissue images
US20140029813A1 (en) Method and system to digitize pathology specimens in a stepwise fashion for review
CN116468690B (zh) 基于深度学习的浸润性非粘液性肺腺癌的亚型分析系统
JP2021018531A (ja) 機械学習モデルの運用装置とその作動方法および作動プログラム、並びに機械学習モデルの学習装置とその作動方法および作動プログラム
CN115115947A (zh) 遥感图像检测方法、装置、电子设备以及存储介质
JP7370759B2 (ja) 画像処理装置、画像処理方法およびプログラム
JP2004038744A (ja) 画像認識アルゴリズム生成方法,画像認識アルゴリズム生成装置,画像認識アルゴリズム生成プログラムおよび画像認識アルゴリズム生成プログラムを記録した記録媒体
Lu et al. An interactive interpretability system for breast cancer screening with deep learning
JP7336268B2 (ja) 情報処理装置、情報処理方法及びプログラム
JP5083060B2 (ja) 画像表示装置、画像表示方法及びプログラム
JPH10222688A (ja) 画像処理方法

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: 20788584

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20788584

Country of ref document: EP

Kind code of ref document: A1