Image processing method and image processing apparatus

Abstract

Provided are an image processing method and an image processing apparatus, which balance detection accuracy and processing speed in an image processing method applying machine learning. The image processing method according to the present disclosure includes the following steps: obtaining a plurality of images of different parts of a prescribed range in which an object is photographed (S2); applying a prescribed image processing method obtained by using machine learning to perform individual evaluation of whether each of the plurality of images includes a feature of a detection object (S3); and performing comprehensive evaluation. In the comprehensive evaluation, when the ratio X1 / Y1 of the number Y1 of the plurality of images in a partial range constituting the prescribed range or being a part of the prescribed range to the number X1 of the images evaluated as including the feature in the individual evaluation among the Y1 images is equal to or greater than a first prescribed value (S4), it is evaluated that the object includes a specific part of the detection object (S5).

Description

Technical Field

[0001] The present invention relates to an image processing method and an image processing apparatus. Background Art

[0002] There is known a defect inspection system, an abnormality inspection system, or a specific part detection system for inspecting a certain abnormal state or a part having characteristics different from other parts. In recent years, a method has been proposed for using machine learning to evaluate the presence of a part (hereinafter referred to as a "specific part") having characteristics different from most other regions, including defects.

[0003] Patent Document 1 describes the following specific part detection system: a specific part image including a specific part having arbitrary characteristics is extracted from a captured image of an object captured by a capturing unit, and the category of the specific part is identified by machine learning with the specific part image as input.

[0004] Prior Art Documents

[0005] Patent Documents

[0006] Patent Document 1: International Publication No. 2019 / 003813 Summary of the Invention

[0007] Problems to be Solved by the Invention

[0008] The specific part to be detected may have various shapes. When the specific part has a large shape or is formed in a long strip shape, the specific part image extracted from the photographed image also becomes large.

[0009] However, if such a specific part is to be detected by machine learning, the learning time required to generate a learned model becomes long. In addition, the calculation time required to detect the specific part using the learned model increases. Although it is expected that such problems will be solved by using an information processing device with high computing power, since it causes an increase in cost, it becomes a disadvantage when installed in a production line for applications such as online inspection.

[0010] In addition, when the size ratio of the specific part image is larger than the image size of the image to which machine learning is to be applied, the specific part image has to be extracted in such a way that a part of the specific part is not included, resulting in a decrease in the detection accuracy of the specific part.

[0011] Therefore, in order to solve the above problems, an object of the present invention is to provide an image processing method and an image processing apparatus that take into account both detection accuracy and processing speed in an image processing method using machine learning.

[0012] Means for Solving the Problems

[0013] The image processing method involved in this disclosure includes the following steps: obtaining a plurality of images, where the plurality of images are images of an object, and each of the plurality of images is an image of a different part of a specified range of the object; respectively applying a specified image processing method obtained using machine learning to the plurality of images to perform individual evaluations on whether each of the plurality of images contains the features of a detection object; and based on whether the features are included in two or more of the plurality of images, performing a comprehensive evaluation on whether the object contains a specific part of the detection object. Here, in the comprehensive evaluation, when the ratio X1 / Y1 of the number Y1 of the plurality of images constituting the specified range or a partial range as a part of the specified range and the number X1 of the images among the Y1 images that are evaluated as containing the features in the individual evaluation is equal to or greater than a first specified value, it is evaluated that the object contains a specific part of the detection object.

[0014] According to such an image processing method, a specified image processing method obtained using machine learning is applied to perform individual evaluations on whether each of the plurality of images contains the features of a detection object, and based on whether the features are included in two or more of the plurality of images, a comprehensive evaluation on whether the object contains a specific part of the detection object is performed. Specifically, based on the number Y1 of at least any one of the plurality of images corresponding to the specified range and the plurality of images constituting a partial range as a part of the specified range, and the number X1 of the images evaluated as containing the features, the presence or absence of a specific part is evaluated. That is, based on the number of images among the plurality of images constituting the specified range or a partial range of the object that have an estimated score indicating the meaning of containing the features in the individual evaluation equal to or greater than a specified value, it is evaluated whether the object contains a specific part of the detection object. Therefore, even when the size of the specific part image is larger than the size of the image for which machine learning is to be applied, it is possible to take into account the shortening of the learning time and the operation time and the detection of a specific part larger than the image size by performing individual evaluations using this image size and performing comprehensive evaluations based on the number of the plurality of images containing the features within the specified range or the partial range.

[0015] In addition, the "evaluation" in this disclosure includes the following: determining the presence or absence of a specific part or determining the presence or absence of a specific part together with probability information, and outputting probability information (estimated score) related to the presence or absence of a specific part. Further, the "evaluation" in this disclosure also includes the following: when there are multiple types of specific parts, outputting the above information related to the presence or absence of a specific part and / or information related to the category of the specific part.

[0016] In addition, different or identical evaluations may be performed before or after the above-mentioned individual evaluation and comprehensive evaluation. For example, after performing the comprehensive evaluation for a specified range, one or more comprehensive evaluations may be further performed for a partial range that is part of the specified range.

[0017] In addition, one aspect of the present disclosure further includes the following steps: segmenting a base image obtained by photographing a specified range of the object to generate the plurality of images having the same image size.

[0018] In addition, in one aspect of the present disclosure, in the comprehensive evaluation, when the ratio X1 / Y1 is less than the first specified value and is equal to or greater than a second specified value smaller than the first specified value, and the ratio X2 / Y2 of the number Y2 (<Y1) of the plurality of images constituting the second partial range to the number X2 of the images evaluated as including the feature in the individual evaluation among the Y2 images is equal to or greater than a third specified value, it is evaluated that the object includes a specific part to be detected, where the second partial range is the specified range constituted by the Y1 images or a part of the partial range.

[0019] In addition, in one aspect of the present disclosure, in the comprehensive evaluation, when there is an image A evaluated as including the feature in the individual evaluation, and in the individual evaluation, it is evaluated that the feature is included in two or more of the plurality of images located around the image A, it is evaluated that the object includes a specific part to be detected.

[0020] In addition, in one aspect of the present disclosure, in the comprehensive evaluation, when there is an image A evaluated as including the feature in the individual evaluation, there is an image B evaluated as including the feature in the individual evaluation and located around the image A, there is an image C evaluated as not including the feature in the individual evaluation and not located around the image A but around the image B, there is an image D evaluated as including the feature in the individual evaluation and not located around the image A, the image B, or the image C but around the image C, and there is an image E evaluated as including the feature in the individual evaluation and not located around the image A, the image B, the image C, or the image D but around the image D, it is evaluated that the object includes a specific part to be detected.

[0021] The present disclosure discloses an image processing apparatus. The image processing apparatus includes: an image acquisition unit that acquires a plurality of images of different parts of a predetermined range of an object being photographed; an individual evaluation unit that individually evaluates whether each of the plurality of images includes a feature of a detection object by applying a predetermined image processing method obtained using machine learning; and a comprehensive evaluation unit that comprehensively evaluates whether the object includes a specific part of the detection object based on whether the feature is included in two or more of the plurality of images, wherein the comprehensive evaluation unit is configured to: when the ratio X1 / Y1 of the number Y1 of the plurality of images constituting the predetermined range or a partial range that is a part of the predetermined range to the number X1 of the images evaluated by the individual evaluation unit as including the feature among the Y1 images is equal to or greater than a first predetermined value, evaluate that the object includes a specific part of the detection object.

[0022] According to such an image processing apparatus, it is possible to individually evaluate whether each of a plurality of images includes a feature of a detection object by applying a predetermined image processing method obtained using machine learning, and to comprehensively evaluate whether an object includes a specific part of the detection object based on whether the feature is included in two or more of the plurality of images. Specifically, it is possible to evaluate the presence or absence of a specific part based on the number Y1 of the plurality of images constituting the predetermined range or a partial range that is a part of the predetermined range and the number X1 of the images evaluated as including the feature. Therefore, even when the size of the specific part image is larger than the size of the image to which machine learning is to be applied, it is possible to shorten the learning time and the operation time and to detect a specific part larger than the image size by performing individual evaluation using the image size and performing comprehensive evaluation based on the number of the plurality of images including the feature within the predetermined range or the partial range.

[0023] In addition, the image processing apparatus of the present disclosure may further include a unit that divides a base image obtained by photographing a predetermined range of the object to generate the plurality of images having the same image size.

[0024] In addition, in one aspect of the present disclosure, the comprehensive evaluation unit is configured to: when the ratio X1 / Y1 is less than the first predetermined value and is equal to or greater than a second predetermined value smaller than the first predetermined value, and the ratio X2 / Y2 of the number Y2 (<Y1) of the plurality of images constituting a second partial range to the number X2 of the images evaluated by the individual evaluation unit as including the feature among the Y2 images is equal to or greater than a third predetermined value, evaluate that the object includes a specific part of the detection object, wherein the second partial range is a part of the predetermined range or the partial range constituted by the Y1 images.

[0025] In addition, in one aspect of the present disclosure, the comprehensive evaluation unit may also be configured to: when there is an image A evaluated by the individual evaluation unit as including the feature, and there are two or more images among the plurality of images located around the image A that are evaluated by the individual evaluation unit as including the feature, evaluate that the object includes a specific part to be detected.

[0026] In addition, in one aspect of the present disclosure, the comprehensive evaluation unit may also be configured to: when there is an image A evaluated by the individual evaluation unit as including the feature, there is an image B located around the image A evaluated by the individual evaluation unit as including the feature, there is an image C not located around the image A but around the image B evaluated by the individual evaluation unit as not including the feature, there is an image D not located around the image A and the image B but around the image C evaluated by the individual evaluation unit as including the feature, and there is an image E not located around the image A, the image B, and the image C but around the image D evaluated by the individual evaluation unit as including the feature, evaluate that the object includes a specific part to be detected.

[0027] In addition, an image being "located" "around" another image means a positional relationship where there is no other image between one image and the other image. Typically, it means a positional relationship where the vertices or sides of one image face or share with those of the other image. An image not being "located" "around" another image means a positional relationship where there is still another image between one image and the other image.

[0028] In addition, in one aspect of the present disclosure, the specific part is a defect of the object.

[0029] In addition, the computer program of the present disclosure includes commands for causing a computer to perform the following operations: acquiring a plurality of images, where the plurality of images are images of an object, and each of the plurality of images is an image of a different part of a specified range of the object; applying a specified image processing method obtained by using machine learning to each of the plurality of images to perform an individual evaluation of whether each of the plurality of images includes a feature of the object to be detected; and performing a comprehensive evaluation of whether the object includes a specific part of the object to be detected based on whether the feature is included in two or more of the plurality of images. Here, in the comprehensive evaluation, when the ratio X1 / Y1 of the number Y1 of the plurality of images constituting the specified range or a partial range as a part of the specified range to the number X1 of the images evaluated as including the feature in the individual evaluation among the Y1 images is equal to or greater than a first specified value, it is evaluated that the object includes a specific part of the object to be detected.

[0030] A computer program can also be stored in a non - transitory storage medium. The non - transitory storage medium includes non - volatile semiconductor storage elements.

[0031] By causing a computer to execute such a computer program, the same effects as the above - described image processing method can be achieved.

[0032] In addition, the image processing system according to the present disclosure includes a server and the above - described image processing device. The server stores a first set of image data including features of various different types and a second set of image data not including features as teacher data for machine learning. The same label can be assigned to the first image data. A label different from the label assigned to the first image data is assigned to the second set of image data. Further, the image processing system according to the present disclosure includes the following mode: execution commands included in the computer program according to the present disclosure are executed collaboratively by a plurality of arithmetic devices each having a processor and a memory, and the plurality of arithmetic devices are connected to each other in a manner capable of communicating wirelessly or wiredly.

[0033] According to such an image processing system, the same effects as the above - described image processing device can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a functional block diagram of the image processing system according to the present embodiment.

[0035] Figure 2 is a block diagram showing the hardware configuration of the image processing system according to the present embodiment.

[0036] Figure 3 is a flowchart of the image processing method according to the present embodiment.

[0037] Figure 4 is a diagram schematically showing the results of individual evaluations performed on segmented images for a partial range.

[0038] Figure 5 shows an example of a pattern showing the positional relationship of a plurality of segmented images used in the comprehensive evaluation of the image processing method according to the modified example of the present embodiment.

[0039] Figure 6 shows an example of a pattern showing the positional relationship of a plurality of segmented images used in the comprehensive evaluation of the image processing method according to the modified example of the present embodiment.

[0040] Figure 7 is a diagram showing the relationship between the size of the specific part and the segmented image. DETAILED DESCRIPTION OF THE INVENTION

[0041] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The following embodiments are illustrative of the present invention and are not intended to limit the present invention solely to these embodiments.

[0042] Figure 1 It is a functional block diagram of the image processing system 10 according to this embodiment. This image processing system 10 is an image processing system for determining whether a foreign object has entered the inside of a polyester bottle (an example of an "object") containing a liquid. However, as will be described later, the present invention can be applied to various objects and various specific parts.

[0043] The image processing system 10 includes an image processing device 20, and the image processing device 20 has an imaging unit 12, an image segmentation unit 14, an individual evaluation unit 16, and a comprehensive evaluation unit 18. Further, the image processing system 10 also includes a server device connected to the image processing device 20 via a network N. The server device includes a database DB1 for storing images of normal samples and a database DB2 for storing images of abnormal samples. In this embodiment, a normal sample refers to a sample that does not include a specific part, that is, a sample in which the object is normal and belongs to a qualified product. In addition, an abnormal sample refers to a sample that includes a specific part, that is, a sample in which the object is abnormal and belongs to a non-conforming product.

[0044] The imaging unit 12 captures part or all of the object to obtain an image (an example of a "basic image"). The imaging unit 12 can be composed of, for example, an image sensor such as a linear sensor or an area sensor. For example, when the object has a shape that is rotationally symmetric about a central axis (such as a container having an internal space for containing a content), by rotating the object about the central axis and capturing the side surface using a linear sensor, an image corresponding to the entire side surface of the object (an example of a "specified range") can be obtained. In addition, the object is not limited to a cylindrical shape. In the case of this embodiment, the imaging unit 12 is composed of a linear sensor, and grayscale data having a luminance value of 8 bits (256 grayscales) is obtained for each pixel of 5000 pixels × 5000 pixels as an image of the object. In addition, the imaging unit 12 may also obtain color data such as RGB data as an image of the object. Further, instead of capturing an image of the object by the imaging unit 12, an image of the object may be obtained from a database or the like for storing images of the object.

[0045] The image segmentation unit 14 (an example of the "image acquisition unit") divides the image acquired by the imaging unit 12 into a plurality of images of the same size, thereby acquiring a plurality of images of different parts of a predetermined range in which an object is photographed (hereinafter, each of the divided images may be referred to as a segmented image). The segmented images are set to have a size suitable for an image processing method using machine learning. Here, the size of the segmented images does not need to be set to include the size that the specific part to be detected may have, and may also be set smaller than the size that the specific part may have. For example, when the size of the predetermined range is 5000 pixels × 5000 pixels and the size of the segmented images is 75 pixels × 75 pixels, the image segmentation unit 14 acquires approximately 4400 segmented images. By setting the size of the segmented images smaller in this way to be suitable for machine learning, the operation time of machine learning can be shortened compared to the case where the image size is set to include the specific part. Also, there is a great advantage as follows: since the number of image sheets required for learning can be reduced, the learning time can be shortened while making it easier to collect images of abnormal samples.

[0046] The individual evaluation unit 16 individually evaluates whether each of the plurality of segmented images contains a feature to be detected by applying a predetermined image processing method obtained using machine learning to each of the plurality of segmented images.

[0047] Here, the specific part refers to a part having a feature different from the surrounding area from at least one of the viewpoints of shape, brightness, and color, and the feature is the specific part itself or a part of the specific part. Therefore, the feature refers to a part having a feature different from the surrounding area from at least one of the viewpoints of shape, brightness, and color.

[0048] The specific part and the feature may be a defect, or may be a manually set mark, pattern, etc. The specific part and the feature related to the shape refer to a part having a characteristic different from the surrounding area in terms of the characteristics of the shape (for example, volume, area, height, particle diameter, aspect ratio, roundness, contour shape, etc.). Similarly, for brightness, it refers to a part having a characteristic different from the surrounding area in terms of characteristics related to brightness such as brightness and histogram, and for color, it refers to a part having a characteristic different from the surrounding area in terms of characteristics related to color such as chromatogram, center wavelength, and maximum wavelength.

[0049] The specific part and the feature in the present embodiment are dust, human hair, air bubbles, and dirt on the container that may be contained in the liquid inside the polyester bottle.

[0050] The individual evaluation unit 16 of this embodiment includes a learned model composed of a convolutional neural network. The convolutional neural network includes an input layer for inputting data of a segmented image, an output layer for outputting the presence or absence of features, and an intermediate layer for combining the two layers. The intermediate layer includes a plurality of combinations of convolutional layers and pooling layers, and a combination layer. The convolutional layer is used to slide a filter and perform a convolution operation of calculating the product sum of repeated parts as a feature amount to generate a graph with a feature amount. The pooling layer is used to extract the maximum value of the two-dimensional arranged data output from the convolutional layer. The combination layer is used to combine internal parameters as weight coefficients and provide an evaluation result to the output layer. In addition, the neural network can be composed of not only a convolutional neural network (Convolutional Neural Network), but also an RNN (Recurrent Neural Network), an Elman network, a Jordan network, an ESN (Echo State Network), an LSTM (Long Short Term Memory network), a BRNN (Bi-directional RNN), etc. In addition, the output layer outputs probability information indicating the possibility that the segmented image of the object has features (sometimes also referred to as "classification probability"). However, as the output by the output layer, a binary result such as the presence or absence of features can be output, or the output can be made according to the types of features.

[0051] The server device includes a database DB1 for storing sample images of normal objects and a database DB2 for storing sample images of abnormal objects. The database DB1 is used to store data of segmented images that do not include specific parts. In this embodiment, the database DB1 stores data of a 75-pixel × 75-pixel segmented image obtained by photographing the side of a PET bottle as an object, which does not include specific parts and features, as data of the segmented image.

[0052] The database DB2 is used to store a second image data group including specific parts or features. In this embodiment, the database DB2 stores data of a 75-pixel × 75-pixel segmented image obtained by photographing the side of a PET bottle as an object, which includes at least one of specific parts and features, as data of the segmented image. In this embodiment, the specific parts and features are, for example, human hair, dust, dirt on the container, or a part of them. Therefore, the database DB2 stores data of a plurality of segmented images including human hair, data of a plurality of segmented images including dust, data of a plurality of segmented images including dirt on the container, etc., respectively.

[0053] In addition, the specific part and the feature vary depending on the object to be detected. For example, when a bubble is the object to be detected, the data of the segmented image including the bubble is stored in the database DB2 as the second image data including the specific part and the feature. However, when a PET bottle containing carbonated beverage is the object to be detected, the bubble is not the object to be detected, so the data of the segmented image including the bubble is stored in the database DB1 as the first image data including the specific part or the feature.

[0054] The individual evaluation unit 16 obtains a learned model including internal parameters for evaluating the presence or absence of a feature by learning using the data of the segmented images stored in the database DB1 and the database DB2 as teacher data. The internal parameters are, for example, the bias value in the convolutional layer, the weight coefficient in the combination layer, and the like. The internal parameters change due to the addition of teacher data.

[0055] The learned model in the present embodiment is a learned model configured to output information related to whether a feature is included in an input image of a specified size. The learned model in the present embodiment is generated by machine learning using teacher data obtained by attaching labels indicating the presence or absence of a feature to the data of a plurality of segmented images. In addition, the learned model may output information related to whether a feature is included in the input image and / or information related to the category of the feature. Such a learned model is generated by machine learning using teacher data obtained by attaching labels indicating the presence or absence of a feature and / or the category of the feature to the data of a plurality of segmented images.

[0056] The individual evaluation unit 16 applies the learned model obtained by machine learning as described above to each segmented image, and performs individual evaluation on whether each of more than 1000 segmented images includes a feature, thereby obtaining probability information indicating the possibility of the presence or absence of a feature.

[0057] The comprehensive evaluation unit 18 evaluates the presence or absence of a specific part based on two or more segmented images evaluated by the individual evaluation unit 16 as possibly including a feature among a plurality of segmented images. More specifically, the comprehensive evaluation unit 18 obtains the number Y1 of segmented images within a specified range and the number X1 of segmented images evaluated by the individual evaluation unit 16 as possibly including a feature. When the ratio X1 / Y1 thereof is equal to or greater than a specified threshold, it is evaluated that the object includes a specific part.

[0058] Figure 2is a block diagram showing the hardware configuration for implementing the image processing apparatus 20. As described above, the imaging unit 12 captures an object to obtain an image of a specified range of the object, and is constituted by, for example, a line camera. The processor 22 executes a computer program (including a learned model for evaluating the presence or absence of a specific part) stored in the storage unit 24, thereby executing each arithmetic process shown in the present embodiment. Therefore, the processor 22 and the storage unit 24 cooperate to function as the image segmentation unit 14, the individual evaluation unit 16, and the comprehensive evaluation unit 18. The processor 22 is constituted by, for example, an ASIC (Application Specific Integrated Circuit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), or a quantum computer having a plurality of arithmetic cores. The storage unit 24 stores each command and other information including a computer program (including a learned model for evaluating the presence or absence of a feature) for executing each arithmetic process shown in the present embodiment. The storage unit 24 is constituted by a nonvolatile semiconductor storage element (non-temporary storage element) such as a NAND flash memory, a FeRAM, or an MRAM that can electrically record and read information, or a magnetic storage element such as an HDD (hard disc drive). The RAM 26 is constituted by volatile semiconductor storage elements such as an SRAM (Static Random Access Memory) and a DRAM (Dynamic Random Access Memory) for temporarily storing data and other information used for performing each arithmetic process shown in the present embodiment. The display unit 28 includes a display for displaying the arithmetic result of the processor 22. The arithmetic result includes the evaluation results of the individual evaluation unit 16 and the comprehensive evaluation unit 18. The communication I / F unit 30 is connected to the database DB1 and the database DB2 via the network N, and can receive information from the database DB1 and the database DB2. The input unit 32 includes an input device such as a keyboard for an inspector to input information to the image processing system 10. These configurations are connected to each other via a bus so as to be able to transmit and receive data. However, a part of the configurations may be remotely provided via the network N, or may be integrated with other configurations. In addition, a part of the functions executed by the processor 22 may be executed by other configurations such as the imaging unit 12. And the server device may also include Figure 2The hardware structure shown. Also, database DB1 and database DB2 can also be configured as information processing devices integrally formed with image processing device 20. Additionally, each information processing device that implements image segmentation unit 14, individual evaluation unit 16, and comprehensive evaluation unit 18 of image processing device 20 can also be provided separately from image processing device 20, and they are connected via network N in a manner capable of electrical communication to form image processing system 10.

[0059] An image processing method using the above-described image processing system 10 will be described. Figure 3 It is a flowchart of the image processing method shown in this embodiment.

[0060] First, imaging unit 12 captures an object to obtain a base image (step S1). Specifically, while rotating a PET bottle as the object around the central axis, the side surface of the PET bottle is imaged using the line camera of imaging unit 12, thereby capturing a base image corresponding to a specified range. In this embodiment, imaging unit 12 captures an image of 5000 pixels × 5000 pixels as the base image corresponding to the specified range. Additionally, in this embodiment, imaging unit 12 obtains grayscale data having a luminance value of 8 bits (256 gradations) for each pixel of 5000 pixels × 5000 pixels as the base image. Here, the container of the PET bottle has translucency. Therefore, imaging unit 12 obtains an image of the container of the PET bottle and the liquid contained therein.

[0061] Next, image segmentation unit 14 divides the base image corresponding to the side surface of the PET bottle obtained by imaging unit 12 into divided images of 75 pixels × 75 pixels, and obtains 1000 or more divided images (step S2).

[0062] Then, individual evaluation unit 16 individually evaluates each divided image based on a learned model generated using machine learning, and thus evaluates whether each divided image contains features (step S3).

[0063] Figure 4This is a diagram schematically showing the results of individual evaluations performed on divided images of a partial range that is part of a specified range. The partial range in this diagram contains 80 divided images that are 8 rows in the vertical direction of the paper surface and 10 columns in the horizontal direction of the paper surface. Additionally, the numerical values shown in each divided image represent probability information indicating the likelihood that each divided image has a feature, obtained by the individual evaluation unit 16. In the present embodiment, the closer the numerical value is to 0, the higher the likelihood of having a feature, and the closer the numerical value is to 1, the lower the likelihood of having a feature. For example, since the probability information of the divided image DI1 is 0.041, the probability that the divided image DI1 does not have a feature is 4.1% (the probability of having a feature is 95.9%). Since the probability information of the divided image DI2 is 0.444, the probability that the divided image DI2 does not have a feature is 44.4% (the probability of having a feature is 55.6%).

[0064] The image processing apparatus 20 may also be configured to be able to display the probability information of each divided image on the display unit 28. For example, it may be configured to display each divided image in a distinguishable color by classifying the probability information using three thresholds, so as to easily show how the divided images evaluated as having features are distributed within the specified range or partial range. In addition, the threshold used as a reference for evaluating as having a feature can be appropriately set according to the detection object. For example, the individual evaluation unit 16 may be configured to evaluate as not having a feature when the probability of not having a feature is 90% or more.

[0065] The reasons for being evaluated as having a feature are diverse. For example, there are cases where it is evaluated as having a feature due to noise. Additionally, there are cases where it is evaluated as having a feature due to dust, human hair, etc. For example, when human hair H( Figure 4 ) is mixed in, it is sometimes evaluated as having a feature over multiple adjacent divided images such as the divided images DI3 to DI7.

[0066] After that, the comprehensive evaluation unit 18 obtains the number Y1 of divided images within the specified range and the number X1 of divided images evaluated by the individual evaluation unit 16 as being likely to contain a feature, and determines whether X1 / Y1, which is the ratio of the two, is equal to or greater than a specified threshold (an example of the "first specified value") (step S4).

[0067] For example, when Y1 is 5000 and X1 is 10, X1 / Y1 is 0.002. When this value is equal to or greater than the specified threshold ("yes"), it is evaluated that the detection object contains a specific part (step S5). When this value is less than the threshold ("no"), it is evaluated that the detection object does not contain a specific part (step S6).

[0068] When the object to be inspected is evaluated as including a specific part, the display unit 28 outputs information to that effect and prompts the inspector to perform a visual inspection (step S7). At this time, the display unit 28 is configured to display the position within a specified range of the divided image evaluated as including a specific part or feature. By being configured in this way, the inspector can easily identify the area evaluated as including a specific part or feature. Therefore, the visual inspection time can be reduced. In addition, when the object to be inspected is evaluated as including a specific part, the image processing apparatus 20 may also output information to that effect to a controller (such as a PLC: programmable logic controller, etc.) for controlling the controlled object on the production line. The controller that receives the information may also control the production line so that the object evaluated as including a specific part is discharged outside the production line system as a defective product or a candidate for a defective product.

[0069] In addition, even when the object to be inspected is evaluated as not including a specific part, a visual inspection may be further performed.

[0070] Alternatively, when the object to be inspected is large, the image processing method including steps S1 to S7 according to the present embodiment may be repeatedly executed for different ranges. On the other hand, when the specific part is a defect and the parts where such a defect may occur are concentrated in a part of the area, even if the object to be inspected is large, the image processing method including steps S1 to S7 according to the present embodiment may be implemented for that area, and the object to be inspected may be inspected by a different method (such as a visual inspection) for other areas.

[0071] According to the image processing system 10 and the image processing method as described above, the size of the divided image can be set to be small to be suitable for machine learning. Therefore, compared with the case where the image size is set to be large in a manner that includes a specific part, the learning time and the operation time of machine learning can be shortened. And the inventors of the present application focused on the fact that noise, etc. that should not be evaluated as a specific part or feature rarely straddle multiple divided images, while dust, human hair, etc. that should be evaluated as a specific part or feature straddle multiple divided images relatively more often. Thus, an image processing system 10 and an image processing method are designed that evaluate the presence of a specific part not based on a single divided image having a feature but based on multiple divided images having a feature. Therefore, the possibility of misidentifying the presence of a specific part due to noise, etc. and thus performing excessive detection can be reduced, and overlooking dust, human hair, etc. that should be evaluated as a specific part or feature can be suppressed, thereby improving the detection accuracy.

[0072] In other words, it can also be said that an image processing system 10 and an image processing method are provided, which set the size of the divided image in such a way that the specific part straddles a plurality of divided images, thereby taking into account both the detection accuracy and the processing speed.

[0073] In the case of an object for mass production, when even if only one of the objects has a specific part, a big problem may occur, by using the image processing system 10 and the image processing method according to the present embodiment to generate a divided image in a size smaller than the specific part, the overlooking rate of the specific part can be reduced.

[0074] In addition, in the image processing method of the present embodiment, when adopting a method using a model obtained by machine learning that can autonomously generate an estimation rule for input data based on teacher data (training data), by using a divided image with a small number of pixels as teacher data, the learning time can be shortened.

[0075] In particular, if it is assumed that the evaluation method of the present embodiment is installed on a production line and used for product inspection, in order to flexibly respond to batch changes or model changes of products to be manufactured, it is preferable that the time for generating a learned model is short. As described above, by suppressing the increase in learning time, an evaluation method suitable for installation on a production line can be realized.

[0076] Furthermore, from the viewpoint of further improving the above effects, as a first method, it is preferable to use an image processing method based on rules that can also be applied to image processing with few elements that should be changed corresponding to batch changes or model changes of products to be manufactured and a large amount of data.

[0077] In addition, it can also be set such that multiple objects can be simultaneously photographed by moving multiple objects while photographing the area photographed by the photographing unit 12. Thereby, multiple objects can be efficiently evaluated. At this time, the image processing system 10 can also be configured to be able to evaluate the entire side surface by rotating the object around the central axis while moving. In addition, while photographing the first object using the photographing unit 12, the presence or absence of features or specific parts of the second object can be evaluated using the individual evaluation unit 16 or the comprehensive evaluation unit 18 at the same time.

[0078] [First Variation Example]

[0079] Next, a first variation example of the first embodiment will be described. In addition, for those skilled in the art, it is reasonable to understand that the variation examples described below have the same or similar structures as the first embodiment or other variation examples, and the description will be omitted or simplified, and the description will be centered on the different parts.

[0080] The image processing apparatus and image processing method according to the first embodiment perform a comprehensive evaluation for a specified range based on the ratio of the number Y1 of a plurality of images constituting the specified range to the number X1 of images evaluated as including features in the individual evaluations among the Y1 images.

[0081] The image processing method according to the first modification example does not perform a comprehensive evaluation for a specified range but for a partial range that is a part of the specified range, based on the ratio of the number Y1 of a plurality of images constituting the partial range to the number X1 of images evaluated as including features in the individual evaluations among the Y1 images.

[0082] The size of the partial range can be appropriately set. For example, the partial range is set in such a way as to include the size of a specific part assumed to be a detection object. The comprehensive evaluation unit 18 can also be configured, for example, as follows: when the size of a typical specific part is included in 16 divided images (a partial range composed of 16 divided images in 4 rows and 4 columns) in many cases, the partial range is set to be composed of the same 16 divided images in 4 rows and 4 columns, and when the number of images evaluated as including features in the individual evaluations is 4 or more, that is, when Y1 is 16, X1 is 4, and X1 / Y1 is 0.25 or more, it is evaluated as including a specific part.

[0083] [Second Modification Example]

[0084] Next, a second modification example of the first embodiment will be described.

[0085] The image processing method according to the second modification example performs an evaluation that is a combination of the comprehensive evaluation described in the first embodiment and the comprehensive evaluation described in the first modification example.

[0086] Specifically, the comprehensive evaluation unit 18 can also be configured such that in step S4, although X1 / Y1 is less than the threshold value ("no") but is equal to or greater than a second threshold value (an example of "second specified value") that is smaller than the threshold value, and the ratio X2 / Y2 of the number Y2 (<Y1) of a plurality of divided images constituting a second partial range that is a part of the specified range composed of Y1 divided images to the number X2 of images evaluated as including features in the individual evaluations among the Y2 images is equal to a third threshold value (an example of "third specified value"), it is evaluated as including a specific part.

[0087] By adopting such a configuration, even when X1 is small within the specified range, it is possible to capture the situation where the divided images evaluated as including features are concentrated in a part of the region, and thus it is possible to suppress the omission rate of specific parts.

[0088] [Third Modification Example]

[0089] Next, a third modification example of the first embodiment will be described.

[0090] The image processing method according to the third modification example performs comprehensive evaluation based on patterns of the positional relationships of a plurality of divided images evaluated as including features in individual evaluations.

[0091] In Figure 5 Examples of patterns of the positional relationships of a plurality of divided images are shown. In this figure, black cells indicate divided images evaluated as including features in the individual evaluation (step S3) of the divided images, and white cells indicate divided images evaluated as not including features.

[0092] As shown in this figure, Figure 5 (A) of Figure 5 and (B) of Figure 5 show the following pattern: Among the eight divided images surrounding the divided image (an example of "Image A") located at the center of the divided images evaluated as including features within a partial range composed of nine divided images in a 3-row and 3-column arrangement, there are two divided images evaluated as including features. Figure 5 (A) of Figure 5 and (B) of Figure 5 both show patterns in which there is one divided image evaluated as including features in each column (in the vertical direction) and including two adjacent divided images in the diagonal direction.

[0093] (A) of

[0094] (B) of

[0095] In addition, patterns obtained by rotating the illustrated pattern by 90 degrees, and patterns obtained by making the pattern symmetric with respect to the top and bottom or left and right of the paper surface may also be used.

[0096] Preferably, when the specific part is linear and has a width narrower than the size of the divided image, comprehensive evaluation is performed as in the former case. On the other hand, when the specific part is linear but has a width wider than the size of the divided image, comprehensive evaluation can also be performed as in the latter case. That is, it is preferable to set a pattern of the positional relationship of a plurality of divided images evaluated as including features in the individual evaluation according to the assumed shape of the specific part.

[0097] Alternatively, such comprehensive evaluation based on the pattern of the positional relationship of a plurality of divided images evaluated as including features in the individual evaluation is performed in step S4 of the first embodiment. When X1 / Y1 is above a specified threshold and the pattern shown in (A) of Figure 5 or Figure 5 (B) etc. of

[0098] is detected, it is evaluated that the object to be detected includes a specific part.

[0099] [Fourth Modification Example]

[0100] Next, a fourth modification example of the first embodiment will be described.

[0101] The similarity between the image processing method according to the fourth modification example and that of the third modification example is that comprehensive evaluation is performed based on the pattern of the positional relationship of a plurality of divided images evaluated as including features in the individual evaluation.

[0102] The inventor of the present application focused on the following aspect: Even in cases other than those described in the third modification example (a case where, among the eight divided images around the divided image (an example of "Image A") located at the center of the divided images evaluated as including features within a partial range composed of nine divided images in 3 rows and 3 columns, there are two divided images evaluated as including features), there are sometimes specific parts. Such a case is a pattern where two divided images (an example of "Image A" and "Image B") evaluated as including features in the individual evaluation are adjacent, two other divided images (an example of "Image D" and "Image E") evaluated as including features in the individual evaluation are adjacent, and Image A and Image B are not adjacent to Image C and Image D, but are separated by one divided image (an example of "Image C") evaluated as not including features in the individual evaluation.

[0103] In other words, the comprehensive evaluation unit 18 is configured such that: when there is an image A that is evaluated as including a feature in individual evaluation, and there is an image B that is located around the image A and is evaluated as including the feature in individual evaluation, and there is an image C that is not located around the image A but around the image B and is evaluated as not including the feature in individual evaluation, and there is an image D that is not located around the images A and B but around the image C and is evaluated as including the feature in individual evaluation, and there is an image E that is not located around the images A, B, and C but around the image D and is evaluated as including the feature in individual evaluation, it is evaluated that the object includes a specific part to be detected.

[0104] In Figure 6 such a pattern is illustrated. In addition, patterns obtained by rotating the illustrated pattern by 90 degrees and patterns obtained by making the pattern symmetric with respect to the top-bottom or left-right of the paper surface are omitted from description. In these drawings, black units show divided images that are evaluated as including a feature in individual evaluation (step S3) of the divided image, and white units show divided images that are evaluated as not including the feature.

[0105] In Figure 6 the (A) of

[0106] shows a pattern that locally includes a pattern in which images B, C, and D are arranged in a straight line in the row direction or column direction. Image A can be any one of images A1 to A3. Image E can be any one of images E1 to E3. Figure 6 the (B) of

[0107] shows a pattern that locally includes a pattern in which images B and C among images B, C, and D are arranged in a straight line in the row direction or column direction, and image D is arranged in an inclined direction with respect to image C. Image A can be any one of images A1 to A3. Image E can be any one of images E1 to E5. Figure 6 the (C) of

[0108] shows a pattern that locally includes a pattern in which images B and D among images B, C, and D are arranged at the same position in the row direction or column direction, and image C is arranged in an inclined direction with respect to images B and D. Image A can be any one of images A1 to A3. Image E can be any one of images E1 to E4. Figure 6 the (D) of

[0109] In addition, Figure 6 the pattern shown in (E) is equivalent to the pattern obtained by rotating the pattern shown in (B) of Figure 6 by 90 degrees and then symmetrically reversing it left and right. Therefore, it is substantially the same as the pattern of (B) of Figure 6 and thus the description thereof is omitted.

[0110] Such comprehensive evaluation based on the positional relationship of patterns of a plurality of divided images evaluated as including features in individual evaluations can be performed together with or independently of the evaluation that X1 / Y1 is above a specified threshold value performed in step S4 of the first embodiment.

[0111] According to the image processing method and the image processing apparatus as described above, even if a feature that should be included in the divided image corresponding to image C is missed in detection, it is possible to detect a specific part based on the surrounding divided images, and thus it is possible to suppress overlooking of the specific part and improve the detection accuracy.

[0112] [Method for generating teacher data]

[0113] Next, a method for generating teacher data stored in database DB1 and database DB2 will be described.

[0114] Conventionally, in the case of performing a sensory inspection on an object having a complex structure with noise generation factors, in order to suppress over-detection, the following operations have been performed: shielding parts with noise generation factors such as boundary parts of the parts and then performing imaging.

[0115] However, if a specific part is truncated due to shielding, the detection performance of the specific part is reduced. Therefore, in the present embodiment, when capturing a base image by photographing an object composed of a plurality of parts in order to generate teacher data, a specified range of the object is photographed in a manner that includes a plurality of parts and their boundary parts without using shielding. Here, the plurality of parts include a case where there is one part composed of a plane, one part composed of a curved surface connected thereto, one component composed of a certain material, and another component composed of another material and connected to the component.

[0116] Then, the base image is divided into a plurality of divided images to generate teacher data. Divided images that do not include specific parts are stored in database DB1. Here, when constituting a specified range that includes a plurality of parts and their boundary parts, these divided images are generated as teacher data divided images without performing noise removal processing such as filtering. Therefore, these divided images sometimes include noise.

[0117] After that, when capturing an object to obtain a base image in step S1, the imaging unit 12 captures a predetermined range of the object in a manner that includes a plurality of parts and their boundary parts, and a segmentation image is obtained based on this in step S2.

[0118] Then, in step S3, each segmentation image is individually evaluated based on a learned model generated using machine learning. When evaluating whether each segmentation image includes a feature, the learned model uses the segmentation image that constitutes a predetermined range including a plurality of parts and their boundary parts as teacher data, and thus can perform individual evaluation of whether each segmentation image that constitutes the predetermined range includes a feature with high accuracy.

[0119] Segmentation images including specific parts or features are stored in the database DB2. These segmentation images also constitute a predetermined range including a plurality of parts and their boundary parts. In Figure 7 An example of specific parts H2 to H4 and an example of a segmentation image generated based on this are shown. As shown in this figure, a specific part is sometimes larger than the size of the segmentation image (in other words, the size of the segmentation image is set to be smaller than a part of the specific part). In addition, even if a specific part is smaller than the size of the segmentation image, the specific part sometimes straddles multiple segmentation images. By performing comprehensive evaluation based on two or more segmentation images by the comprehensive evaluation unit 18 shown in this embodiment, the detection accuracy of such specific parts can be improved.

[0120] [Application Scope]

[0121] The image processing system and the image processing apparatus according to the present disclosure can be applied to evaluate various objects. For example, in the manufacturing process of a transparent film of polyvinylidene chloride (PVDC), it can be used to detect foreign matters, defects, wrinkles, etc. that may be contained in the transparent film as specific parts.

[0122] In addition, in the manufacturing process of a polymer film for a flat panel display or the like, it can be used to detect foreign matters, defects, etc. that may be contained in the polymer film as specific parts.

[0123] In addition, in the manufacturing process of glass products or the like, it can be used to detect cracks, defects, etc. that may be contained in the glass as specific parts. In this case, the glass product can be illuminated and imaged, and the specific part can be detected by using the fact that the brightness at the specific part is higher than the brightness of other parts.

[0124] In addition, in the manufacturing process of a separation membrane for chromatography, it can be used to detect deformations and bubbles on the membrane surface as specific parts. In this case, the separation membrane can be illuminated and imaged, and the specific part can be detected by using the fact that the brightness at the specific part is lower or higher than the brightness of other parts.

[0125] In addition, the present disclosure can be variously modified without departing from the gist of the present disclosure. For example, a part of the constituent elements in a certain embodiment can be added to other embodiments within the normal creative ability of those skilled in the art. In addition, a part of the constituent elements in a certain embodiment can be replaced with corresponding constituent elements in other embodiments.

[0126] For example, the image processing method according to the present disclosure may also include the following steps:

[0127] Obtaining a plurality of images, the plurality of images being images of an object, and each of the plurality of images being an image of a different part of a specified range of the object;

[0128] Individually evaluating, for each of the plurality of images, whether each of the plurality of images includes a feature of a detection object by applying a specified image processing method obtained using machine learning; and

[0129] Based on whether the feature is included in two or more of the plurality of images, comprehensively evaluating whether the object includes a specific part of the detection object,

[0130] wherein, in the comprehensive evaluation, based on the positional relationship of the plurality of images that constitute the specified range or a partial range that is a part of the specified range in the individual evaluation, evaluating whether the object includes a specific part of the detection object.

[0131] For example, the image processing apparatus according to the present disclosure may also include:

[0132] An acquisition unit that acquires a plurality of images, the plurality of images being images of an object, and each of the plurality of images being an image of a different part of a specified range of the object;

[0133] An individual evaluation unit that individually evaluates, for each of the plurality of images, whether each of the plurality of images includes a feature of a detection object by applying a specified image processing method obtained using machine learning; and

[0134] A comprehensive evaluation unit that comprehensively evaluates whether the object includes a specific part of the detection object based on whether the feature is included in two or more of the plurality of images,

[0135] wherein the comprehensive evaluation unit is configured to: based on the positional relationship of the plurality of images that constitute the specified range or a partial range that is a part of the specified range in the individual evaluation, evaluate whether the object includes a specific part of the detection object.

[0136] In such an image processing method and an image processing apparatus, it may also be that

[0137] each of the plurality of images is an image of a rectangular area having a long side and a short side,

[0138] the comprehensive evaluation is configured such that when at least two of the images evaluated as including the feature are adjacent in the inclination direction in the individual evaluation, it is evaluated that the object includes a specific part to be detected.

[0139] Description of Reference Numerals

[0140] 10: Image processing system; 12: Imaging unit; 14: Image segmentation unit; 16: Individual evaluation unit; 18: Comprehensive evaluation unit; 20: Image processing apparatus; 22: Processor; 24: Storage unit; 28: Display unit; 32: Input unit; DB1: Database; DB2: Database.

Claims

1. An image processing method, characterized in that, Comprising the following steps: Obtain a plurality of images, the plurality of images being images of an object, each of the plurality of images including different parts of a specified range of the object; Apply a specified image processing method obtained using machine learning to each of the plurality of images to perform an individual evaluation of whether each of the plurality of images includes a feature of a detection object; And Based on whether the feature is included in two or more of the plurality of images, perform a comprehensive evaluation of whether the object includes a specific part of the detection object, wherein, in the comprehensive evaluation, when the ratio X1 / Y1 of the quantity X1 to the quantity Y1 is equal to or greater than a first specified value, it is evaluated that the object includes a specific part of the detection object, the quantity Y1 being the number of the plurality of images constituting the specified range or a partial range as a part of the specified range, and the quantity X1 being the number of the images among the Y1 images that are evaluated as including the feature in the individual evaluation; wherein, in the comprehensive evaluation, when the ratio X1 / Y1 is less than the first specified value and is equal to or greater than a second specified value smaller than the first specified value, and the ratio X2 / Y2 of the quantity X2 to the quantity Y2 is equal to or greater than a third specified value, it is evaluated that the object includes a specific part of the detection object, where the quantity Y2 is the number of a plurality of images constituting a second partial range, the quantity X2 being the number of the images among the Y2 images that are evaluated as including the feature in the individual evaluation, the second partial range being a part of the specified range or the partial range constituted by the Y1 images, and Y2 < Y1.

2. The image processing method according to claim 1, wherein Further comprising the following steps: Segment a base image obtained by photographing the specified range of the object to generate the plurality of images having the same image size.

3. The image processing method according to claim 1 or 2, wherein in the comprehensive evaluation, when there is an image A that is evaluated as including the feature in the individual evaluation, and the feature is included in two or more of the plurality of images located around the image A in the individual evaluation, it is evaluated that the object includes a specific part of the detection object.

4. The image processing method according to claim 1 or 2, wherein the specific part is a defect of the object.

5. The image processing method according to claim 2, wherein the plurality of images generated by segmenting the base image have a size smaller than that of the specific part.

6. An image processing method, characterized in that, Comprising the following steps: Obtain a plurality of images, the plurality of images being images of an object, each of the plurality of images including different parts of a specified range of the object; Apply a specified image processing method obtained using machine learning to each of the plurality of images to perform an individual evaluation of whether each of the plurality of images includes a feature of a detection object; And Based on whether the feature is included in two or more of the multiple images, a comprehensive evaluation is performed on whether the object includes a specific part to be detected. In the comprehensive evaluation, if there is an image A that is evaluated as including the feature in the individual evaluation, and there is an image B that is evaluated as including the feature in the individual evaluation and is located around the image A, and there is an image C that is evaluated as not including the feature in the individual evaluation and is not located around the image A but around the image B, and there is an image D that is evaluated as including the feature in the individual evaluation and is not located around the image A and the image B but around the image C, and there is an image E that is evaluated as including the feature in the individual evaluation and is not located around the image A, the image B, and the image C but around the image D, then it is evaluated that the object includes a specific part to be detected.

7. An image processing apparatus, characterized in that, Comprising: An image acquisition unit that acquires multiple images of different parts of a specified range of an object; An individual evaluation unit that applies a specified image processing method obtained by using machine learning to each of the multiple images to perform an individual evaluation on whether each of the multiple images includes a feature to be detected; And A comprehensive evaluation unit that performs a comprehensive evaluation on whether the object includes a specific part to be detected based on whether the feature is included in two or more of the multiple images, wherein, when the ratio X1 / Y1 of the quantity X1 to the quantity Y1 is equal to or greater than a first specified value, the comprehensive evaluation unit evaluates that the object includes a specific part to be detected. The quantity Y1 is the number of the multiple images that constitute the specified range or a partial range that is part of the specified range, and the quantity X1 is the number of the images among the Y1 images that are evaluated as including the feature in the individual evaluation. wherein the comprehensive evaluation unit is configured to: When the ratio X1 / Y1 is less than the first specified value and is equal to or greater than a second specified value that is smaller than the first specified value, and the ratio X2 / Y2 of the quantity X2 to the quantity Y2 is equal to or greater than a third specified value, it is evaluated that the object includes a specific part to be detected. Here, the quantity Y2 is the number of the multiple images that constitute a second partial range, the quantity X2 is the number of the images among the Y2 images that are evaluated as including the feature in the individual evaluation, the second partial range is a part of the specified range or the partial range constituted by the Y1 images, and Y2 < Y1.

8. The image processing apparatus according to claim 7, wherein it further includes a unit that segments a base image obtained by photographing the specified range of the object to generate the multiple images having the same image size.

9. The image processing apparatus according to claim 7 or 8, wherein the comprehensive evaluation unit is configured to: In the case where there is an image A evaluated by the individual evaluation unit as including the feature, and the feature is included in two or more of the plurality of images located around the image A as evaluated by the individual evaluation unit, it is evaluated that the object includes a specific part to be detected.

10. The image processing apparatus according to claim 7 or 8, wherein: The specific part is a defect of the object.

11. An image processing apparatus, characterized in that, comprising: an image acquisition unit that acquires a plurality of images of different parts of a predetermined range of an object; an individual evaluation unit that individually evaluates whether each of the plurality of images includes a feature to be detected by applying a predetermined image processing method obtained by machine learning to the plurality of images; and a comprehensive evaluation unit that comprehensively evaluates whether the object includes a specific part to be detected based on whether the feature is included in two or more of the plurality of images, wherein the comprehensive evaluation unit is configured to: In the case where there is an image A evaluated by the individual evaluation unit as including the feature, there is an image B evaluated by the individual evaluation unit as including the feature and located around the image A, there is an image C evaluated by the individual evaluation unit as not including the feature and not located around the image A but around the image B, there is an image D evaluated by the individual evaluation unit as including the feature and not located around the image A, the image B, or the image C but around the image C, and there is an image E evaluated by the individual evaluation unit as including the feature and not located around the image A, the image B, the image C, or the image D but around the image D, it is evaluated that the object includes a specific part to be detected.

Images