Computer system and data generation method for use in learning of model for executing image segmentation
A computer system facilitates efficient annotation for image segmentation by iteratively performing segment division and integration, enabling users to generate high-quality training data without extensive image processing knowledge.
Patent Information
- Application Number
- JP2023073617
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-27
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The creation of training data for machine learning models in image segmentation is time-consuming, and existing methods require expert knowledge in image processing to achieve accurate preliminary segmentation, making it difficult for annotators without such knowledge to perform efficient annotation.
A computer system that generates data for image segmentation by performing segment division and integration processes multiple times, allowing users to select representative images without detailed parameter adjustments, thereby facilitating efficient annotation.
Enables efficient and accurate generation of training data for image segmentation, reducing the time and expertise required for creating high-quality training images.
Smart Images

Figure 2024158427000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a system and method for generating data for use in training models that perform image segmentation. [Background technology]
[0002] Patent Document 1 describes an image segmentation system for assisting in training a machine learning model, and also describes the use of preliminary segmentation (coloring results) generated using existing non-machine learning image processing methods during annotation. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2021-002338 A Summary of the Invention [Problem to be solved by the invention]
[0004] Training a machine learning model requires teacher data, and when training a machine learning model for image segmentation, teacher data (teacher images) is required in which objects in the image are colored according to their type. The task of creating teacher data, also known as annotation, is necessary for creating a machine learning model, and since it is a time-consuming task, it is desirable to make it more efficient.
[0005] Patent Document 1 describes the use of results obtained by using a partially trained segmentation and discrimination model for annotation. If a partially trained segmentation and discrimination model can be used, it is possible to obtain relatively good results, and it is expected that annotation work can be made more efficient. However, since training data is necessary even for partial training, this method can only be applied after a certain amount of training data has been created.
[0006] Patent Document 1 also describes performing preliminary segmentation using a non-machine learning image processing method, which can be applied even to the very first annotation (teacher data creation). On the other hand, as described in the document that in many cases preliminary segmentation is rough or coarse, it is generally difficult to perform good segmentation using a non-machine learning image processing method. This is partly because non-machine learning image processing methods require parameter settings, and good results cannot be obtained unless the parameters are set appropriately. When an image processing expert sets the parameters, even if the effect of changing the parameters can be grasped, the expert is not necessarily familiar with the objects in the image, and it is difficult to determine how the objects should be colored as a result of the preliminary segmentation. On the other hand, the annotator, who is in charge of annotation, can determine how the objects in the image should be colored, but is not necessarily familiar with image processing, and it is difficult to set the parameters appropriately.
[0007] Due to the above circumstances, there is a demand for a technology that can be applied even when creating initial training data and that allows even annotators with no knowledge of image processing to perform appropriate preliminary segmentation. [Means for solving the problem]
[0008] A representative example of the invention disclosed in the present application is as follows: That is, a computer system that generates data to be used in learning a model for performing image segmentation, which receives an input of an image, and generates a plurality of segment images by executing a segment division process for dividing the image into a plurality of segments a plurality of times while changing a first parameter used in the segment division process on the image, generates a plurality of segment images by executing a segment integration process on a segment image selected from the plurality of segment images a plurality of times while changing a second parameter used in the segment integration process for integrating at least one segment included in the segment images, and generates a plurality of segment integrated images, and outputs the segment integrated image selected from the plurality of segment integrated images as data to be used in learning the model. Effect of the Invention
[0009] According to the present invention, it is possible to support annotation work. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]
[0010] [Figure 1] FIG. 2 is a diagram showing a hardware configuration of the image coloring device according to the first embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a functional configuration of the image coloring device according to the first embodiment. [Diagram 3] 4 is a flowchart illustrating an example of a process executed by the image coloring device according to the first embodiment. [Figure 4] FIG. 2 is a diagram showing an example of a screen presented by the image coloring device of the first embodiment. [Diagram 5] 4A to 4C are diagrams illustrating an example of a segment integration process executed by the image coloring device according to the first embodiment. [Figure 6] FIG. 2 is a diagram showing an example of a screen presented by the image coloring device of the first embodiment. [Figure 7]FIG. 11 is a diagram showing a hardware configuration of an image coloring device according to a second embodiment. [Figure 8] FIG. 11 is a diagram illustrating an example of a functional configuration of an image coloring device according to a second embodiment. [Figure 9] 11 is a flowchart illustrating an example of a process executed by the image coloring device according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0011] Hereinafter, examples of the tissue image analysis device and tissue image analysis method according to the present invention will be described with reference to the attached drawings. Note that the present invention should not be interpreted as being limited to the description of the embodiments shown below. It will be easily understood by those skilled in the art that the specific configuration can be changed without departing from the concept or spirit of the present invention.
[0012] In the configuration of the invention described below, the same parts or parts having similar functions are denoted by the same reference numerals in different drawings, and duplicated explanations may be omitted. When there are multiple elements having the same or similar functions, they may be described with different subscripts added to the same reference numerals. However, when it is not necessary to distinguish between multiple elements, the subscripts may be omitted.
[0013] The designations "first," "second," "third," and the like in this specification are given to identify components and do not necessarily limit the number, order, or content. Furthermore, numbers for identifying components are used in different contexts, and a number used in one context does not necessarily indicate the same configuration in another context. Furthermore, a component identified by a certain number is not prevented from also having the function of a component identified by another number.
[0014] In order to facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings, etc. may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings, etc.
[0015] One feature of the embodiment described in detail below is that the input image is subjected to rule-based segmentation and segment integration, and a number of representative output images are sequentially presented to the user, allowing the user to select one, resulting in a good color-coded image without the need for detailed parameter adjustment. EXAMPLES
[0016] FIG. 1 is a diagram showing a hardware configuration of an image coloring device 100 according to the first embodiment.
[0017] The image coloring device 100 comprises a processor 110, a memory 111, a storage device 112, and a network adapter 113. The hardware elements are connected to each other via a system bus 114 so as to be able to transmit and receive signals. In addition, a display device 115 and an input device 116 are connected to the image coloring device 100. The image coloring device 100 is connected to an image input device 101 and an image database 102 via a network 103 so as to be able to transmit and receive signals.
[0018] Here, "capable of transmitting and receiving signals" refers to a state in which signals can be transmitted and received between each other or from one side to the other, regardless of whether the transmission is electrically or optically wired or wireless.
[0019] The processor 110 controls the operation of each hardware element, and also executes a program 120 stored in the storage device 112 .
[0020] The memory 111 stores the program 120 executed by the processor 110, as well as intermediate results of the computational processing.
[0021] The storage device 112 is a hard disk drive (HDD) or a solid state drive (SSD) or the like, and stores a program 120 executed by the processor 110 and data required for executing the program. The storage device 112 stores the program 120 for realizing functional blocks described below.
[0022] The network adapter 113 is connected to a network 103 such as a LAN (Local Area Network), a telephone line, the Internet, etc. Various data handled by the processor 110 may be transmitted to and received from the outside via the network 103.
[0023] The display device 115 is, for example, a liquid crystal display or a touch panel, and displays the analysis results of the image coloring device 100. The input device 116 is, for example, a keyboard, a mouse, a touch panel, and the like, and allows the user to give operational instructions to the image coloring device 100. A pointing device such as a track pad or a track ball may be used instead of a mouse.
[0024] The image input device 101 is a device that captures and inputs an image, and is, for example, a digital camera, a scanning electron microscope, an optical microscope, etc. The image database 102 is a database that stores the images captured by the image input device 101, etc.
[0025] Next, a functional configuration of the image coloring device 100 will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of a functional configuration of the image coloring device 100 according to the first embodiment.
[0026] The image coloring device 100 has, as functional blocks, a segment division unit 201, a segment divided image output unit 202, a segment integration unit 203, and a segment integrated image output unit 204. A program 120 for realizing these functional blocks is stored in the storage device 112. Note that the storage device 112 stores images captured by the image input device 101.
[0027] The segmentation unit 201 executes image processing for dividing an image into regions with similar features based on the brightness, color, and contours of the image. The image processing is performed according to rules based on specified parameters. For example, the quick shift method and the Felzenszwalb method can be used as the rules.
[0028] Hereinafter, the image processing executed by the segment division unit 201 will be referred to as segment division processing, and an image generated by the segment division processing will be referred to as a segmented image.
[0029] The segmentation unit 201 changes parameters and executes the segmentation process multiple times to generate multiple segmented images. The parameters may be changed randomly or based on a predetermined rule. The present invention is not limited to the method of changing the parameters.
[0030] The segmented image output unit 202 outputs one or more representative images from among the multiple segmented images. The segmented image output unit 202 accepts a user's selection for the output images, and outputs one or more segmented images similar to the selected segmented image based on the selection result.
[0031] The segment integration unit 203 executes image processing to evaluate boundaries between segments of a segmented image, remove unnecessary boundaries, and integrate the segments. The image processing includes a differentiation process for the input image, a superposition process for superimposing the differentiated image and the segment boundary line, an index calculation process for calculating the degree of unnecessaryness of the boundaries by comparing the edge components at the two boundaries with the edge components inside each of the two segments, and a boundary removal process for removing the boundaries based on the degree of unnecessaryness and integrating the segments.
[0032] Hereinafter, the image processing executed by the segment integration unit 203 will be referred to as a segment integration process, and an image generated by the segment integration process will be referred to as a segment-integrated image.
[0033] The segment integration unit 203 changes the threshold value for the degree of unnecessaryness of the boundary and executes the segment integration process multiple times to generate multiple segment integrated images. Note that the threshold value may be changed randomly or based on a predetermined rule. The present invention is not limited to the method of changing the threshold value.
[0034] The segment integrated image output unit 204 outputs one or more representative images from the segment integrated images. The segment integrated image output unit 204 accepts a user's selection for the output images, and outputs one or more segment integrated images similar to the selected segment integrated image based on the selection result.
[0035] Each functional block shown in Fig. 2 is realized by executing a program by the processor 110. Note that each functional block shown in Fig. 2 may be realized by using dedicated hardware such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field-Programmable Gate Array).
[0036] It should be noted that, regarding the functional blocks shown in FIG. 2, a plurality of functional blocks may be integrated into one functional block, or one functional block may be divided into two or more functional blocks.
[0037] Next, a description will be given of a process executed by the image coloring device 100. Fig. 3 is a flowchart illustrating an example of a process executed by the image coloring device 100 of the first embodiment.
[0038] The segmentation unit 201 acquires an image (step S301). The image may be acquired from the storage device 112, or may be acquired from the outside (the image input device 101 or the image database 102) via the network adapter 113. The image may be, for example, a grayscale image captured by an electron microscope, or a color image captured by an optical microscope or a camera.
[0039] The segment division unit 201 executes a segment division process (step S302).
[0040] Here, in the rule-based segmentation method, there are one or more setting parameters. The segmentation unit 201 changes the setting parameters and executes the segmentation process multiple times. As a result, multiple segmented images are generated.
[0041] The segmented image output unit 202 displays one or more representative images from among the segmented images generated by the segment division unit 201 (step S303).
[0042] The segmented image output unit 202 determines whether or not the selection of the segmented images is completed based on a user operation (step S304).
[0043] If the selection of the segmented image is not completed (step S304: NO), the segmented image output unit 202 returns to step S303 and executes the same process. If the selection of the segmented image is completed (step S304: YES), the image coloring device 100 proceeds to step S305.
[0044] Here, a screen presented by the image coloring device 100 when executing the segment division process will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of a screen presented by the image coloring device 100 of the first embodiment.
[0045] A screen 401 includes a field for setting image division (a setting field), a field for selecting an image (a selection field), and a field for displaying the final selection result (a confirmation field).
[0046] In the setting field, information on the input image, segmentation algorithm, parameters, etc. is set. In Fig. 4, the file name of an image containing a spherical object is specified. Furthermore, QuickShift is selected as the segmentation algorithm, and the kernel size is set as a setting parameter in the range of 3 to 20. Different setting parameters are displayed for each segmentation algorithm.
[0047] After the user sets the image and setting parameters in the setting field, the user presses the "Execute" button. It is assumed that "Manual" is selected in the image selection checkbox. The operation when "Automatic" is selected will be explained in Example 2. In response to this operation, the segment division unit 201 executes segment division processing. Furthermore, the segment division image output unit 202 selects a representative image from the multiple segment division images, and displays it in the first selection row of the selection field.
[0048] For example, the segmented image output unit 202 selects and displays a predetermined number of segmented images based on the number of segments. In Fig. 4, three segmented images, a segmented image with a large number of segments, a segmented image with a medium number of segments, and a segmented image with a small number of segments, are displayed as representative images.
[0049] The user looks at the representative images and selects the image that is closest to what the user expects from the options. In Figure 4, options A, C, and E, which correspond to the images themselves, as well as options B and D, which correspond to images between the two images, are displayed. As shown in Figure 4, option D is selected in the first selection.
[0050] When the user presses the "Continue" button, the segmented image output unit 202 selects new segmented images based on the selected segmented image and displays the new segmented images in the second selection row. For example, the segmented image output unit 202 selects a predetermined number of segmented images based on the similarity to the selected segmented image.
[0051] In FIG. 4, a segmented image selected based on a segmented image close to option D is displayed. An image with an intermediate number of segments between options C and E in the first selection is displayed as option C in the second selection. Option C is selected in the second selection. If the "Continue" button is pressed, segmented images are displayed using a similar procedure. By repeating this process multiple times, the image desired by the user can be narrowed down with high precision.
[0052] In FIG. 4, the third to fifth selections are omitted since they were made in the same manner, and the sixth selection shows the case where option C is selected and the "Done" button is pressed.
[0053] In the confirmation field, the input image and the segmented image of option C from the sixth selection are displayed side by side. The user checks the segmented image, and if there are no problems, presses the "OK" button to complete the selection of the segmented image.
[0054] This concludes the description of the screens displayed when the segment division process is executed. Returning to the description of FIG.
[0055] In step S305, the segment integration unit 203 executes a segment integration process on the selected segmented image (step S305). In the segmentation process, if an attempt is made to obtain a result that reflects the detailed shape of the object, the segments become fine and contain many unnecessary boundaries. Therefore, the segment integration process is executed to integrate the unnecessary boundaries.
[0056] FIG. 5 is a diagram for explaining an example of the segment integration process executed by the image coloring device 100 according to the first embodiment.
[0057] In segmented image 502 generated from enlarged image 501 of the input image, in addition to segment dividing lines corresponding to the contour of the spherical object in enlarged image 501 of the input image, segment dividing lines also exist inside the spherical object.
[0058] When making the segments correspond to each other so that each sphere can be identified, it is desirable to remove the segment dividing lines inside the sphere and integrate the segments. Therefore, the segment integration unit 203 calculates an index (degree of unnecessaryness) for identifying the unnecessary segment dividing lines. Specifically, the degree of unnecessaryness is calculated by the following procedure.
[0059] (Step 1) The segment integration unit 203 performs differentiation processing on the input image to generate a differentiated image 503 .
[0060] (Step 2) The segment integration unit 203 overlays the segment dividing lines of the segmented image 502 on the differential image 503 to generate an overlaid image 504 .
[0061] (Step 3) The segment integration unit 203 calculates the degree of unnecessaryness of the boundary between the segments contained in the two images.
[0062] As an example, in the case of two segments 505 to be compared, one segment hatched with a grid and the other segment hatched with diagonal lines, the segment integration unit 203 averages the differential image values of the parts corresponding to the segment boundary line between them (the dotted line part between the two segments) to calculate the boundary component value. The segment integration unit 203 also averages the differential image values of the parts corresponding to the area hatched with a grid or diagonal lines to calculate the segment interior component value. The segment integration unit 203 calculates the ratio of the boundary component value to the segment interior component value as the degree of unnecessaryness. For example, when the ratio obtained by dividing the segment interior component value by the boundary component value is used as the degree of unnecessaryness, it can be determined that the larger the value, the better it is to remove the boundary and integrate the segments.
[0063] Here, the degree of integration of segments can be changed by varying the threshold value for the degree of unnecessaryness. In step S305, the segment integration unit 203 generates a plurality of segment integrated images by varying the threshold value.
[0064] This concludes the description of the segment integration process. Returning to the description of FIG.
[0065] The segment integrated image output unit 204 displays one or more representative images from among the multiple segment integrated images generated by the segment integration unit 203 (step S306).
[0066] The segment integrated image output unit 204 determines whether or not the selection of the segment integrated image is completed based on a user operation (step S307).
[0067] If the selection of the segment integrated image is not completed (step S307: NO), the segment integrated image output unit 204 returns to step S306 and executes the same process. If the selection of the segment integrated image is completed (step S307: YES), the image coloring device 100 outputs the processing result (step S308) and ends the process.
[0068] The image coloring device 100 may output only the segment integrated image as the processing result, or may output the input image and the segment integrated image as the processing result. Also, the segment divided image may be included in the processing result.
[0069] Here, a screen presented by the image coloring device 100 when executing the segment integration process will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of a screen presented by the image coloring device 100 of the first embodiment.
[0070] A screen 601 includes a field for setting image integration (a setting field), a field for selecting images (a selection field), and a field for displaying the final selection result (a confirmation field).
[0071] The setting field displays the segmented images selected by the user. In addition, the number of segmented integrated images to be generated is set. This setting value can be reflected in the fineness of the threshold value for the degree of unnecessaryness of segment boundaries.
[0072] When the user operates the "Execute" button, the segment integration unit 203 executes the segment integration process. It is assumed that the image selection checkbox is selected as "Manual". The operation when "Automatic" is selected will be explained in the second embodiment. Furthermore, the segment integrated image output unit 204 selects a representative image from the multiple generated segment integrated images, and displays it in the first selection row of the selection column.
[0073] For example, the segment-integrated image output unit 204 selects and displays a predetermined number of segment-integrated images based on the number of remaining segments. In Fig. 6, a segment-integrated image having a large number of remaining segments, a medium number of remaining segments, and a small number of remaining segments are displayed as representative images.
[0074] In FIG. 6, the representation of the segments has been changed from dividing lines to differences in color tone, making it a format that corresponds to the color-coded images for annotation images (teacher images).
[0075] The user looks at the representative images and selects from the options the image that is closest to what the user expects. In Fig. 6, as in Fig. 4, options A, C, and E corresponding to the images themselves are displayed, as well as options B and D corresponding to images intermediate between the two images. As shown in Fig. 6, option E is selected in the first selection.
[0076] When the user presses the "Continue" button, the segment integrated image output unit 204 selects a new segment integrated image based on the selected segment integrated image, and displays the new segment integrated image in the second selection row. For example, the segment integrated image output unit 204 selects a predetermined number of segment integrated images based on the similarity to the selected segment integrated image.
[0077] In FIG. 6, a segment integrated image selected based on a segment integrated image that is close to option E is displayed. Option E from the first selection is displayed as option C in the second selection row. Option A displays an image with a larger number of segments than the segment integrated image, and option E displays an image with a smaller number of segments than the segment integrated image. Option C is selected in the second selection. If the "Continue" button is pressed, segment integrated images are displayed using a similar procedure. By repeating this process multiple times, the images requested by the user can be narrowed down with high precision.
[0078] In FIG. 6, the third to fifth selections are omitted since they are made in the same manner. In the sixth selection, option C is selected and the "Done" button is pressed.
[0079] In the confirmation field, the input image and the segment integrated image of option C from the sixth selection are displayed side by side. The user sets a save location and presses the "OK" button to complete the selection of the segment integrated image.
[0080] According to the first embodiment, the finally obtained segment integrated image can be used for annotation work (creating a teacher image for a machine learning model). Since most of the segment integrated image is colored as expected by the user, it is possible to generate the teacher image more efficiently and quickly than when generating the teacher image from the original image.
[0081] The image color-coding device according to the embodiment of the present invention has been described above. The present invention is not limited to the above embodiment, and the components can be modified without departing from the gist of the invention. EXAMPLES
[0082] In the first embodiment, the user manually selects images (segmented images and segmented integrated images). In the second embodiment, the image selection is performed automatically by utilizing the usage history of the image coloring device 100. Hereinafter, the second embodiment will be described focusing on the difference from the first embodiment.
[0083] 7 is a diagram showing a hardware configuration of the image coloring device 100 of the embodiment 2. The hardware configuration of the image coloring device 100 of the embodiment 2 is the same as that of the embodiment 1. However, in the embodiment 2, a usage history 701 consisting of an input image, a final segmented image, and a final segment integrated image is stored in the storage device 112.
[0084] The functional configuration of the image coloring device 100 of the second embodiment is partially different from that of the first embodiment. Fig. 8 is a diagram showing an example of the functional configuration of the image coloring device 100 of the second embodiment. The image coloring device 100 of the second embodiment newly includes an image selection inference unit 801.
[0085] The image selection inference unit 801 infers the selection of a segmented image and a segment integrated image. The image selection inference unit 801 automatically selects an image based on a model generated by machine learning using the usage history 701.
[0086] FIG. 9 is a flowchart illustrating an example of a process executed by the image coloring device 100 according to the second embodiment.
[0087] The processes from step S901 to step S903 and from step S905 to step S906 are the same as the processes from step S301 to step S303 and from step S305 to step S306.
[0088] In the second embodiment, in step S904, the image selection and inference unit 801 selects one segment image from among the multiple segment images based on the input image. In addition, in step S907, the image selection and inference unit 801 selects one segment integrated image from among the multiple segment integrated images based on the input image.
[0089] The user selects the "Auto" checkbox in the setting column of screen 401 and screen 601, and presses the "Execute button" to automatically select each image. In this case, the "Selection column" is not displayed, and the selection result is displayed in the "Confirmation column."
[0090] In step S908, the image coloring device 100 outputs the processing result and stores the usage history 701 in the storage device 112.
[0091] As described above, according to the second embodiment, image selection can be automated by learning the results of image selection performed based on the know-how of a user with specialized knowledge. This makes it easier to obtain a segment-integrated image, and further reduces the amount of work required for annotation work. In addition, when high accuracy is not required, image processing such as particle analysis can be applied to the obtained color-coded image, and it can be used to extract feature values such as particle diameter.
[0092] The present invention is not limited to the above-mentioned embodiment, but includes various modified examples. For example, the above-mentioned embodiment describes the configuration in detail to easily explain the present invention, and the present invention is not necessarily limited to the configuration including all the described configurations. Also, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.
[0093] In addition, each of the above configurations, functions, processing units, processing means, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. The present invention can also be realized by software program code that realizes the functions of the embodiments. In this case, a storage medium on which the program code is recorded is provided to a computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-mentioned embodiments, and the program code itself and the storage medium storing it constitute the present invention. Examples of storage media for supplying such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, SSDs (Solid State Drives), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, non-volatile memory cards, ROMs, etc.
[0094] Furthermore, the program code for realizing the functions described in this embodiment can be implemented in a wide range of program or script languages, such as assembler, C / C++, perl, Shell, PHP, Python, Java (registered trademark), etc.
[0095] Furthermore, the program code of the software that realizes the functions of the embodiments may be distributed over a network and stored in a storage means such as a computer's hard disk or memory, or in a storage medium such as a CD-RW or CD-R, and the processor of the computer may read out and execute the program code stored in the storage means or storage medium.
[0096] In the above-mentioned embodiment, the control lines and information lines are shown as those considered necessary for the explanation, and not all the control lines and information lines are shown in the product. All the components may be connected to each other. [Explanation of symbols]
[0097] 100 Image painting device 101 Image input device 102 Image Database 103 Network 110 Processor 111 Memory 112 Storage device 113 Network Adapter 114 System Bus 115 Display device 116 Input Devices 120 Programs 201 Segment division section 202 Segmented image output unit 203 Segment Integration Department 204 Segment Integrated Image Output Unit 401, 601 screen 701 Usage History 801 Image Selection and Inference Unit
Claims
1. 1. A computer system for generating data for use in training a model for performing image segmentation, comprising: Accepts image input, generating a plurality of segmented images by performing a segmentation process on the image a plurality of times while varying a first parameter used in the segmentation process to divide the image into a plurality of segments; generating a plurality of segment integrated images by executing the segment integration process a plurality of times on the segment division images selected from the plurality of segment division images while changing a second parameter used in the segment integration process for integrating at least one segment included in the segment division images; A computer system comprising: a computer that outputs the segment-integrated image selected from a plurality of the segment-integrated images as data to be used for training the model.
2. 2. The computer system of claim 1, providing a first interface for selecting the segmented image; selecting a plurality of the segment division images to be presented to a user from among the plurality of the segment division images based on the number of segments, and presenting the plurality of segment division images via the first interface; A computer system characterized in that, when a request for re-presentation of the segment division image, including specification of the segment division image, is received via the first interface, a plurality of the segment division images are selected based on the specified segment division image and presented via the first interface.
3. 2. The computer system of claim 1, providing a second interface for selecting the segment integrated image; selecting a plurality of the segment integrated images to be presented to a user from among the plurality of the segment integrated images based on the number of segments, and presenting the selected plurality of segment integrated images via the second interface; A computer system characterized in that, when a request for re-presentation of the segment integrated image, including specification of the segment integrated image, is received via the second interface, a plurality of the segment integrated images are selected based on the specified segment integrated image and presented via the second interface.
4. 2. The computer system of claim 1, maintaining a model for accepting an image as an input and selecting the segmented image and the segmented integrated image, the model being trained using a history including the image, the segmented images selected by a user, and the segmented integrated image selected by a user; selecting, using the model, one of the segmented images; A computer system comprising: a computer that uses the model to select one of the plurality of segment-integrated images.
5. 2. The computer system of claim 1, generating a differential image of said image; calculating an index for determining whether or not to integrate segments included in the selected segment division image using the differential image and the selected segment division image; A computer system comprising: a computer that integrates segments included in the selected segmented image based on the index.
6. A method for generating data used to train a model for performing image segmentation executed by a computer system, comprising the steps of: A first step in which the computer system accepts an input of an image; a second step of the computer system generating a plurality of segmented images by performing a segmentation process on the image a plurality of times while varying a first parameter used in the segmentation process to divide the image into a plurality of segments; a third step in which the computer system generates a plurality of segment integrated images by executing the segment integration process a plurality of times on the segment division images selected from the plurality of segment division images while changing a second parameter used in the segment integration process for integrating at least one segment included in the segment division images; A data generating method characterized by including a fourth step in which the computer system outputs the segment integrated image selected from the plurality of segment integrated images as data to be used for training the model.
7. 7. A method for generating data according to claim 6, comprising: The second step includes: the computer system providing a first interface for selecting the segmented image; the computer system selecting a plurality of the segment division images to be presented to a user from among the plurality of the segment division images based on the number of segments, and presenting the plurality of the segment division images via the first interface; A data generation method characterized by including a step in which, when the computer system receives a request for re-presentation of the segment division image, including specification of the segment division image, via the first interface, selecting a plurality of the segment division images based on the specified segment division image and presenting them via the first interface.
8. 7. A method for generating data according to claim 6, comprising: The third step includes: the computer system providing a second interface for selecting the segmented integrated image; the computer system selecting a plurality of the segment integrated images to be presented to a user from among the plurality of the segment integrated images based on the number of segments, and presenting the plurality of segment integrated images via the second interface; A data generation method characterized by including a step of: when the computer system receives a request for re-presentation of the segment integrated image, including specification of the segment integrated image, via the second interface, selecting a plurality of the segment integrated images based on the specified segment integrated image and presenting them via the second interface.
9. 7. A method for generating data according to claim 6, comprising: the computer system maintains a model for accepting an image as an input and selecting the segmented image and the segmented integrated image, the model being trained using a history including the image, the segmented images selected by a user, and the segmented integrated image selected by a user; the second step includes the computer system using the model to select one of the segmented images from the plurality of segmented images; The data generating method, wherein the third step includes a step of the computer system using the model to select one of the plurality of segment integrated images.
10. 7. A method for generating data according to claim 6, comprising: The third step includes: generating a differential image of the image by the computer system; a step of the computer system calculating an index for determining whether or not to integrate segments included in the selected segment division image, using the differential image and the selected segment division image; and a step of integrating segments included in the selected segmented image based on the index by the computer system.
Citation Information
Patent Citations
Method and system for image segmentation and identification
JP2021002338A
