JP2023066489A - Evaluation system, evaluation method, and evaluation program - Google Patents

Abstract

To provide a novel evaluation system, evaluation method, and evaluation program for evaluating a surface of an object.SOLUTION: An evaluation system according to one example includes: an acquisition unit that acquires an input image showing a surface of an object; a first estimation unit that inputs the input image to a first learned model and calculates a first estimated value for the surface; and a second estimation unit that inputs the input image to a second learned model different from the first learned model and calculates a second estimated value for the surface.SELECTED DRAWING: Figure 1

Description

One aspect of the present disclosure relates to an evaluation system, an evaluation method, and an evaluation program.

Patent Literature 1 describes an inspection apparatus that uses a plurality of images obtained by imaging a single inspection target to perform a visual inspection of the inspection target. This inspection apparatus estimates inspection results for each of a plurality of images based on a learning model that predefines the relationship between predefined images and inspection results. Based on this, it has a comprehensive determination unit that calculates a comprehensive inspection result.

Japanese Patent No. 6708695

A new technique for evaluating the surface of an object is desired.

An evaluation system according to one aspect of the present disclosure includes an acquisition unit that acquires an input image showing the surface of an object, and a first trained model that inputs the input image into a first trained model to calculate a first estimated value for the surface. 1 estimating unit; and a second estimating unit that inputs the input image to a second trained model different from the first trained model and calculates a second estimated value for the surface.

An evaluation method according to one aspect of the present disclosure is performed by an evaluation system that includes at least one processor. This evaluation method includes the steps of acquiring an input image showing the surface of the object, inputting the input image into a first trained model to calculate a first estimated value of the surface, and converting the input image into a first and inputting a second trained model, different from the one trained model, to compute a second estimate for the surface.

An evaluation program according to one aspect of the present disclosure includes steps of acquiring an input image showing the surface of an object, and inputting the input image into a first trained model to calculate a first estimated value of the surface. and inputting the input image to a second trained model different from the first trained model to calculate a second estimate of the surface.

According to one aspect of the present disclosure, a new technique for evaluating the surface of an object can be provided.

It is a figure showing an example of functional composition of an evaluation system. It is a figure which shows an example of the hardware constitutions of the computer used with an evaluation system. It is a flow chart which shows an example of processing in an evaluation system. FIG. 4 is a diagram showing an example of processing for obtaining a plurality of input images from an original image; FIG. 4 is a diagram showing an example of a diffuse reflection image and a specular reflection image; FIG. 4 is a diagram showing an example of estimation and evaluation of the surface of an object;

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are denoted by the same reference numerals, and overlapping descriptions are omitted.

[System configuration]
The evaluation system 1 according to the embodiment is a computer system that analyzes the surface of an object based on an input image of the surface of the object. An object is any tangible object. In one example, the evaluation system 1 performs its analysis to evaluate the treatment applied to the surface of the object. Examples of treatments evaluated include, but are not limited to, painting and polishing.

For its analysis, the evaluation system 1 feeds the input image into a trained model to calculate an estimate for the surface of the object. A trained model is a computational model that processes input images and outputs estimates. A trained model is generated in advance by machine learning. Machine learning refers to a method of autonomously discovering laws or rules by learning iteratively based on given information. A trained model is built using algorithms and data structures. In one example, the trained model is built by a neural network such as a convolutional neural network (CNN). An example of a CNN includes, but is not limited to, VGG16, a 16-layer CNN model. It should be noted that the trained model used in the evaluation system 1 is a computational model that is estimated to be optimal, and is not necessarily the “actually optimal computational model”. Trained models are portable between computer systems. Therefore, the evaluation system 1 may use a trained model generated by another computer system. Alternatively, the evaluation system 1 may further have a function of generating trained models. Generating a trained model corresponds to the learning phase, and using the trained model by the evaluation system 1 corresponds to the operation phase.

Users of rating system 1 may enjoy various advantages. For example, by introducing the evaluation system 1, it becomes possible to automate manual inspections and reduce the number of inspection man-hours. In addition, it is possible to eliminate variations in judgment based on human senses and stabilize the quality of examination. Furthermore, since the analysis is performed by image processing, the surface of the object can be inspected without physically touching the surface.

FIG. 1 is a diagram showing an example of the functional configuration of the evaluation system 1. As shown in FIG. In one example, the evaluation system 1 includes a selection unit 10, an acquisition unit 11, an estimation unit 12, and an evaluation unit 13 as functional modules. The selection unit 10 is a functional module that selects images to be processed by the evaluation system 1 . The acquisition unit 11 is a functional module that acquires an input image representing the surface of an object. The estimation unit 12 is a functional module that inputs an input image to a trained model and calculates an estimated value. The evaluation unit 13 is a functional module that evaluates the surface of the object based on the estimated values.

In one example, the estimator 12 comprises a first estimator 14 and a second estimator 15 . The first estimation unit 14 is a functional module that inputs the input image to the first trained model and calculates a first estimated value regarding the surface of the object. The second estimation unit 15 is a functional module that inputs an input image to a second trained model different from the first trained model and calculates a second estimated value for the surface. That is, the estimation unit 12 may process one input image with two types of trained models and calculate two types of estimated values.

Each of the first trained model and the second trained model performs one of classification and regression based on the input image. Classification may be object detection, estimating the location or region of a given object in the input image. Alternatively, the classification may be a presence determination that estimates whether a given object is present in the input image. Thus, a trained model that performs classification may perform object detection and may perform presence determination. For example, a first trained model may perform classification based on input images and a second trained model may perform regression based on input images. In this case, the first estimating unit 14 inputs the input image to the first trained model and calculates the class indicating the physical property of the surface as the first estimated value. That is, the first estimation unit 14 may solve a multi-class problem, for example, a two-class problem. The second estimation unit 15 inputs the input image to the second trained model, and calculates continuous values indicating the physical properties of the surface as second estimated values. Examples of physical properties include, but are not limited to, the state of painting or polishing applied to the surface. The coating state of the surface may be represented by an index related to the amount of coating. For example, the coating state may be expressed qualitatively, such as see-through or dripping. "Transparent" refers to a state in which a part of the substrate is exposed due to insufficient coating. Dripping refers to a state in which a portion of the coating rises like a drop due to excessive coating. The coating state may be expressed quantitatively such as film thickness.

Rating system 1 can be implemented by any kind of computer. The computer may be a general-purpose computer such as a personal computer or a server for business use, or may be incorporated in a dedicated device that executes specific processing. The evaluation system 1 may be implemented by one computer, or may be implemented by a distributed system having a plurality of computers.

FIG. 2 is a diagram showing an example of the hardware configuration of the computer 100 used in the evaluation system 1. As shown in FIG. In this example, computer 100 comprises main body 110 , monitor 120 and input device 130 .

The main body 110 is the device that performs the main functions of the computer. The main body 110 has a circuit 160 which has at least one processor 161 , a memory 162 , a storage 163 , an input/output port 164 and a communication port 165 . Storage 163 records programs for configuring each functional module of main body 110 . The storage 163 is a computer-readable recording medium such as a hard disk, nonvolatile semiconductor memory, magnetic disk, or optical disk. The memory 162 temporarily stores programs loaded from the storage 163, calculation results of the processor 161, and the like. The processor 161 configures each functional module by executing a program in cooperation with the memory 162 . The input/output port 164 inputs and outputs electrical signals to/from the monitor 120 or the input device 130 according to instructions from the processor 161 . The input/output port 164 may input/output electrical signals to/from other devices. Communication port 165 performs data communication with other devices via communication network N according to instructions from processor 161 .

Monitor 120 is a device for displaying information output from main body 110 . The monitor 120 may be of any type as long as it can display graphics, and a specific example thereof is a liquid crystal panel.

Input device 130 is a device for inputting information into main body 110 . The input device 130 may be of any type as long as desired information can be input, and specific examples thereof include operation interfaces such as a keypad, mouse, and operation controller.

The monitor 120 and the input device 130 may be integrated as a touch panel. For example, the main body 110, the monitor 120, and the input device 130 may be integrated like a tablet computer.

Each functional module of the evaluation system 1 is realized by loading an evaluation program into the processor 161 or memory 162 and causing the processor 161 to execute the program. The evaluation program includes codes for realizing each functional module of the evaluation system 1. FIG. The processor 161 operates the input/output port 164 or communication port 165 according to the evaluation program to read and write data in the memory 162 or storage 163 .

The evaluation program may be provided after being permanently recorded in a non-temporary recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory. Alternatively, the evaluation program may be provided via a communication network as a data signal superimposed on a carrier wave.

[Evaluation method]
As an example of the evaluation method according to the present disclosure, an example of a processing procedure executed by the evaluation system 1 will be described with reference to FIG. FIG. 3 is a flow chart showing an example of processing in the evaluation system 1 as a processing flow S1. That is, the evaluation system 1 executes the processing flow S1.

In step S11, the acquisition unit 11 acquires an input image showing the surface of the object. For example, the acquisition unit 11 acquires an input image showing a painted or polished surface. The input image may be a still image, or may be one frame input image forming a video. Acquisition unit 11 may receive an input image sent from a camera or another computer. Alternatively, the acquisition unit 11 may accept an input image input by a user, or read an input image from a given storage device based on user input. In one example, the acquisition unit 11 performs object recognition on the input original image, and cuts out a region in which the target object is shown from the image. The obtaining unit 11 obtains the clipped area as an input image.

In one example, the acquisition unit 11 acquires an original image of the surface of the object, divides the original image, and acquires a plurality of input images. FIG. 4 is a diagram showing an example of the processing. In this example, the acquisition unit 11 acquires an original image 200 representing the surface of the object 210 . The acquisition unit 11 cuts out a region in which the object 210 is shown from the original image 200, divides the cut out region, and acquires six input images 221-226.

The dimensions of the input image may be set based on various factors such as the type of object, the type of surface, the method of treatment applied to the surface, and the like. For example, the acquisition unit 11 may acquire an input image of 448 (pixels)×448 (pixels).

In one example, the acquisition unit 11 acquires either one of a diffuse reflection image and a specular reflection image as an input image. A specular reflection image is an image from which specular reflection components of light are extracted. A diffuse reflection image is an image from which diffuse reflection components of light are extracted.

FIG. 5 is a diagram showing a diffuse reflection image and a specular reflection image obtained from a common original image of the surface of an object. As shown in this figure, the appearance of the surface state differs between the diffuse reflection image and the specular reflection image. As an example, when the glossiness of the surface is metallic, that is, when the glossiness of the surface is high, the specular reflection image clearly shows the condition of the surface. Typically, when a paint mixed with metal powder such as metal flakes is used, the gloss of the surface becomes metallic. On the other hand, when the glossiness of the surface is non-metallic, that is, when the glossiness of the surface is low, the surface condition appears more clearly in the diffuse reflection image. As another example, specular images are suitable for surface flaw inspection. Diffuse reflection images, on the other hand, are suitable for processes such as inspecting for foreign matter or dirt on the surface and evaluating film thickness.

Considering such properties of each image, the acquisition unit 11 may acquire either a diffuse reflection image or a specular reflection image. In one example, the acquisition unit 11 acquires the image selected by the selection unit 10 . For example, the selection unit 10 selects one of the regular reflection image and the diffuse reflection image based on the type of coating of the object, and the acquisition unit 11 acquires the selected image as the input image. The selection unit 10 may perform the selection by instructing the camera to capture one of the specular reflection image and the diffuse reflection image. Alternatively, the selection unit 10 may select one of the specular reflection image and the diffuse reflection image received by the evaluation system 1 . Alternatively, the selection unit 10 may perform the selection by converting the received original image into one of the specular reflection image and the diffuse reflection image. In any case, the acquisition unit 11 acquires the image selected by the selection unit 10 as the input image. In one example, the selection unit 10 may select a specular reflection image when the glossiness of the surface is metallic, and the obtaining unit 11 may obtain the specular reflection image.

Returning to FIG. 3, in step S12, the estimation unit 12 selects a trained model corresponding to the input image. In one example, the estimation unit 12 includes a trained model for calculating an estimated value based on a specular reflection image (a trained model for a specular reflection image) and a trained model for calculating an estimated value based on a diffuse reflection image (a diffuse A trained model corresponding to the input image is selected from among the trained models for the reflection image). It can be said that the trained model for the specular reflection image and the trained model for the diffuse reflection image are candidates for the trained model for processing the input image. In one example, the first estimating unit 14 selects the first trained model candidate for the specular reflection image and the first trained model candidate for the diffuse reflection image from the first trained model candidate corresponding to the input image. Choose a model. Further, the second estimating unit 15 generates a second trained model corresponding to the input image from the second trained model candidate for the specular reflection image and the second trained model candidate for the diffuse reflection image. to select. Thus, the estimator 12 (the first estimator 14 and the second estimator 15) can also function as a selector.

In step S13, the first estimation unit 14 inputs the input image to the first trained model to calculate the first estimated value. In one example, the first trained model is a trained model that performs classification based on input images. In this case, the first estimator 14 inputs the input image to the first trained model and calculates a class representing the physical property of the surface as the first estimated value. When multiple input images are acquired, the first estimation unit 14 calculates a first estimated value for each of the multiple input images.

In step S14, the estimation unit 12 determines whether the first estimated value satisfies a given condition. If the first estimated value satisfies the condition, the process proceeds to step S15, and if the first estimated value does not satisfy the condition, step S15 is not executed. In one example, the estimator 12 determines whether the first estimated value belongs to a specific class. If the first estimate is of that class, step S15 is performed, otherwise step S15 is omitted. When a plurality of input images are acquired, the estimation unit 12 executes the determination for each of the plurality of input images. In one example, step S15 is performed for at least one input image if the first estimate satisfies a given condition for the at least one input image. Alternatively, step S15 may be skipped for all input images if the first estimate does not satisfy the given condition for at least one input image.

In step S15, the second estimator 15 inputs the input image to the second trained model to calculate the second estimated value in response to the fact that the first estimated value satisfies the given condition. In one example, the second trained model is a trained model that performs regression based on the input image. In this case, the second estimator 15 inputs the input image to the second trained model, and calculates continuous values representing the physical properties of the surface as second estimated values. When multiple input images are acquired, the second estimation unit 15 can calculate a second estimated value for each of the multiple input images.

In step S16, the evaluation unit 13 evaluates the surface of the object based on at least one of the first estimated value and the second estimated value. When the first estimated value is calculated and the second estimated value is not calculated, the evaluation unit 13 evaluates the surface of the object based on the first estimated value. When the first estimated value and the second estimated value are calculated, the evaluation unit 13 may evaluate the surface of the object based on both estimated values, or the second estimated value may be evaluated without using the first estimated value. The evaluation may be performed based on estimates. That is, the evaluation unit 13 evaluates the surface based on the first estimated value in response to the first estimated value not satisfying the condition, and at least the first estimated value in response to the first estimated value satisfying the condition. Surfaces may be evaluated based on two estimates. The evaluation unit 13 may directly use at least one of the first estimated value and the second estimated value as the evaluation value. Alternatively, the evaluation unit 13 may determine a given index value or physical property value as an evaluation value based on at least one of the first estimated value and the second estimated value.

When a plurality of input images are acquired, the evaluation unit 13 evaluates each of the plurality of partial regions of the surface corresponding to the plurality of input images based on at least one of the first estimated value and the second estimated value. do. When the input images 221-226 shown in FIG. 4 are acquired, the surface of the object 210 is divided into six subregions. In this case, the evaluation unit 13 evaluates each of these partial areas.

FIG. 6 is a diagram showing an example of estimation and evaluation of the surface of an object. In this example, the evaluation system 1 estimates and evaluates the state of paint applied to the surface 300 of the object. Acquisition unit 11 acquires an input image showing painted surface 300 . The first estimator 14 inputs the input image to the first learned model, and calculates a class indicating the painting state of the surface as a first estimated value. In this example, the first trained model outputs either class ID "0" indicating dripping of paint or class ID "1" indicating a state other than dripping for one input image. In response to the fact that the first estimated value is "1", the second estimator 15 inputs the input image to the second trained model, and uses continuous values indicating the state of painting on the surface as the second estimated value. calculate. In this example, the second trained model estimates the “non-sagging” paint condition for one input image by one numerical value within the range of −10 to +10. The lower limit value of -10 means that the paint is most transparent, and the upper limit value of +10 means that the amount of paint is so large that the paint is likely to drip. The first estimated value and the second estimated value are set in consideration of the fact that the state of the coating is approximately proportional to the amount of paint until dripping occurs, and once dripping occurs, the linear change ceases to hold. It is.

FIG. 6 shows five input images 311-315. In the input image 311, the degree of transparency of the paint on the surface 300 is relatively large. In the input image 312, the degree of transparency is relatively small, but it cannot be said that the paint is good. The input image 313 has just enough paint on the surface 300, so it can be said that this is a good paint condition. In the input image 314, the paint is partially raised because the amount of paint is slightly excessive. In the input image 315, the excessive amount of paint causes sagging on the surface 300. FIG.

In the example of FIG. 6, the first estimation unit 14 calculates class ID "1" for each of the input images 311 to 314 and class ID "0" for the input image 315 as the first estimated values. This corresponds to step S13 in the processing flow S1. In this example, it is assumed that the given condition for the first estimated value is "the first estimated value is 1". For each of the input images 311 to 314, in response to the first estimated value being 1 (YES in step S14), the second estimation unit 15 inputs the input image to the second trained model, One value between -10 and +10 is calculated as the second estimated value (step S15). For example, the second estimation unit 15 calculates a value relatively close to −10 as the second estimated value for the input image 311, and calculates a value relatively close to +10 as the second estimated value for the input image 314. do.

In the example of FIG. 6, the evaluation unit 13 evaluates the paint condition of the surface 300 based on at least one of a binary first estimated value and a second estimated value between -10 and +10. The evaluation unit 13 may directly determine at least one of the first estimated value and the second estimated value as the evaluation value. Alternatively, the evaluation unit 13 may determine index values such as “see-through”, “OK”, and “drip” as evaluation values based on at least one of the first estimated value and the second estimated value. Alternatively, the evaluation unit 13 may convert the second estimated value into a film thickness and determine the film thickness as the evaluation value. When the second trained model outputs the film thickness based on the input image, the evaluation unit 13 may directly determine the output value as the evaluation value.

In the example of FIG. 6, the first estimation unit 14 may estimate the painting state of the surface using three or more class IDs. For example, the first estimation unit 14 may estimate in detail at least one of paint transparency and paint dripping using two or more class IDs. The second estimation unit 15 may calculate the second estimated value for any one or more class IDs among the three or more class IDs. For example, the first estimating unit 14 may calculate the second estimated value in response to at least one class ID related to paint transparency, or the second estimated value in response to at least one class ID related to paint dripping. may be calculated.

Returning to FIG. 3, the evaluation unit 13 outputs the evaluation result in step S17. The evaluation unit 13 may display the evaluation results on the monitor 120 in the form of text, computer graphics (CG) moving images or still images. Alternatively, the evaluation unit 13 may store evaluation results in a recording medium such as the storage 163 . Alternatively, the evaluation unit 13 may transmit evaluation results to other computers. When a plurality of input images are acquired, the evaluation unit 13 may output an evaluation result indicating the correspondence between the position of the partial area within the surface of the object and the evaluation value for each input image. Based on the evaluation results, the surface state can be presented in detail for each partial area.

The evaluation system 1 can repeatedly execute the process flow S1. In each processing flow S1, the acquiring unit 11 may acquire a specular reflection image or may acquire a diffuse reflection image. That is, the acquisition unit 11 may acquire at least one of the specular reflection image and the diffuse reflection image as the input image.

The evaluation system 1 acquires both a specular reflection image and a diffuse reflection image as input images in step S11 for a certain surface, and executes the processing of steps S12 to S17 for each of the two types of input images. good. In this case, the evaluation system 1 may acquire a plurality of input images by image division for at least one of specular reflection images and diffuse reflection images, and perform estimation and evaluation for each input image. As described above, the specular reflection image and the diffuse reflection image show different surface states. Therefore, by estimating the surface of the object based on both the specular reflection image and the diffuse reflection image, it becomes possible to evaluate the surface from a plurality of viewpoints.

[effect]
As described above, the evaluation system according to one aspect of the present disclosure includes an acquisition unit that acquires an input image representing the surface of an object, and inputs the input image to the first trained model to obtain the first A first estimating unit that calculates an estimated value, and a second estimating unit that inputs the input image to a second trained model different from the first trained model and calculates a second estimated value for the surface.

An evaluation method according to one aspect of the present disclosure is performed by an evaluation system that includes at least one processor. This evaluation method includes the steps of acquiring an input image showing the surface of the object, inputting the input image into a first trained model to calculate a first estimated value of the surface, and converting the input image into a first and inputting a second trained model, different from the one trained model, to compute a second estimate for the surface.

An evaluation program according to one aspect of the present disclosure includes steps of acquiring an input image showing the surface of an object, and inputting the input image into a first trained model to calculate a first estimated value of the surface. and inputting the input image to a second trained model different from the first trained model to calculate a second estimate of the surface.

In this aspect, two estimated values are calculated using two types of trained models, respectively, for one input image. Employing two types of estimation allows the surface of the object to be properly evaluated.

In the evaluation system according to another aspect, the first trained model is a trained model that performs classification based on the input image, and the first estimation unit inputs the input image to the first trained model, As one estimate, a class may be calculated that indicates the physical properties of the surface. By solving a classification problem, the physical properties of an object's surface can be accurately estimated using multiple classes (categories).

In the evaluation system according to another aspect, the second trained model is a trained model that performs regression based on the input image, and the second estimation unit inputs the input image to the second trained model, 2 As an estimate, a continuous value indicative of the physical properties of the surface may be calculated. By solving a regression problem, detailed physical properties of the surface of the object can be estimated.

In the evaluation system according to another aspect, the second estimator may calculate the second estimated value in response to the first estimated value satisfying a given condition. Since the second estimated value is calculated only in specific cases, the surface of the object can be estimated in detail while suppressing the calculation load on the evaluation system.

The evaluation system according to another aspect further comprises an evaluation unit that evaluates the surface based on at least one of the first estimate and the second estimate, the evaluation unit responding that the first estimate does not satisfy the condition. Then, the surface may be evaluated based on the first estimate, and in response to the first estimate satisfying the condition, the surface may be evaluated based on at least the second estimate. A second estimated value is calculated according to the result of the first estimated value, and the estimated value used for the evaluation is selectively used according to such estimation. Therefore, it is possible to accurately evaluate the surface of the object while suppressing the calculation load on the evaluation system.

In the evaluation system according to another aspect, the acquisition unit may acquire at least one of a specular reflection image and a diffuse reflection image as the input image of the surface of the object. The specular reflection image and the diffuse reflection image represent the surface condition of the object more clearly than the normal image. By using such an image, the surface of the object can be estimated with higher accuracy.

The evaluation system according to another aspect further includes a selection unit that selects one of the specular reflection image and the diffuse reflection image based on the type of coating on the surface, and the acquisition unit selects the one image from the input image. can be obtained as This configuration makes it possible to estimate the state of the surface according to the type of coating on the surface.

In the evaluation system according to another aspect, the selection unit may select the regular reflection image when the gloss state of the surface is metallic. The specular reflection image clearly shows the state of the metallic surface. Such a selection therefore allows a better evaluation of the surface.

In the evaluation system according to another aspect, the first estimation unit corresponds to the input image from the first trained model candidate for the specular reflection image and the first trained model candidate for the diffuse reflection image. and the second estimating unit selects the first trained model for the specular reflection image and the second trained model candidate for the diffuse reflection image from the second trained model candidates for the input image. A corresponding second trained model may be selected. Since a trained model is selected according to the type of input image, the surface of the object can be estimated with high accuracy while dealing with a plurality of types of images.

The evaluation system according to another aspect further includes an evaluation unit that evaluates the surface based on at least one of the first estimated value and the second estimated value, and the acquisition unit divides the original image showing the surface of the object. A first estimation unit calculates a first estimated value for each of the plurality of input images, and a second estimation unit calculates a second estimated value for each of the plurality of input images. , the evaluating unit may evaluate each of the plurality of partial regions of the surface corresponding to the plurality of input images based on at least one of the first estimate and the second estimate. Since the surface of the object is divided into subregions and each subregion is evaluated, the surface can be evaluated in more detail.

In the evaluation system according to another aspect, the acquisition unit acquires an input image showing the painted surface, the first estimation unit calculates a first estimated value representing the state of coating of the surface, and the second estimation unit may calculate a second estimate representing the surface coating condition. In this case, it becomes possible to appropriately evaluate the coating state of the surface of the object.

An evaluation system according to one aspect of the present disclosure includes an acquisition unit that acquires one of a specular reflection image and a diffuse reflection image as an input image of the surface of an object, and an estimated value for the surface based on the specular reflection image. Select the trained model corresponding to the input image from a trained model for specular reflection images that calculates , and a trained model for diffuse reflection images that calculates surface estimates based on diffuse reflection images A selection unit and an estimation unit that inputs an input image to the selected trained model and calculates an estimated value for the surface.

A specular image or a diffuse image shows the surface condition of an object more clearly than a normal image. In this aspect, since a trained model is selected according to the type of input image, it is possible to appropriately evaluate the surface of the object while dealing with these two types of images.

[Modification]
The above has been described in detail based on the embodiments of the present disclosure. However, the disclosure is not limited to the above examples. Various modifications can be made to the present disclosure without departing from the gist thereof.

The functional configuration of the evaluation system is not limited to the above example. For example, the estimator may estimate a single estimate rather than the first estimate and the second estimate. Alternatively, the estimator may also calculate a second estimate regardless of what the first estimate is. The estimation unit may calculate the estimated value using a predetermined trained model regardless of the input image. The evaluation part may be omitted.

The hardware configuration of the evaluation system is not limited to the implementation of each functional module by executing a program. For example, at least part of the functional module group described above may be configured by a logic circuit specialized for that function, or may be configured by an ASIC (Application Specific Integrated Circuit) integrated with the logic circuit.

The procedure of the method executed by at least one processor is not limited to the above examples. For example, some of the steps or processes described above may be omitted, or the steps may be performed in a different order. Also, two or more of the steps described above may be combined, and some of the steps may be modified or deleted. Alternatively, other steps may be performed in addition to the above steps.

When comparing two numerical values within a computer system or within a computer, either of the two criteria "greater than" and "greater than" may be used, and the two criteria "less than" and "less than" may be used. Either of the criteria may be used.

Reference Signs List 1 evaluation system 10 selection unit 11 acquisition unit 12 estimation unit 13 evaluation unit 14 first estimation unit 15 second estimation unit 200 original image 210 target object 221 226: input image, 300: surface of object, 311 to 315: input image.

Claims (14)

an acquisition unit that acquires an input image showing the surface of an object;
a first estimation unit that inputs the input image to a first trained model and calculates a first estimated value for the surface;
a second estimation unit that inputs the input image to a second trained model different from the first trained model and calculates a second estimated value for the surface;
A rating system with the first trained model is a trained model that performs classification based on the input image;
The first estimation unit inputs the input image to the first trained model, and calculates a class indicating physical properties of the surface as the first estimated value.
The evaluation system according to claim 1. the second trained model is a trained model that performs regression based on the input image;
The second estimation unit inputs the input image to the second trained model, and calculates a continuous value indicating physical properties of the surface as the second estimated value.
The evaluation system according to claim 1 or 2. The second estimation unit calculates the second estimated value in response to the first estimated value satisfying a given condition.
The evaluation system according to any one of claims 1-3. further comprising an evaluation unit that evaluates the surface based on at least one of the first estimated value and the second estimated value;
The evaluation unit
evaluating the surface based on the first estimate in response to the first estimate failing the condition;
evaluating the surface based at least on the second estimate in response to the first estimate satisfying the condition;
The evaluation system according to claim 4. The acquisition unit acquires at least one of a specular reflection image and a diffuse reflection image as an input image of the surface of the object.
The evaluation system according to any one of claims 1-5. a selection unit that selects one of the specular reflection image and the diffuse reflection image based on the type of coating on the surface;
the acquisition unit acquires the one image as the input image;
The evaluation system according to claim 6. wherein the selection unit selects the specular reflection image when the gloss state of the surface is metallic;
The evaluation system according to claim 7. The first estimating unit calculates the first trained model corresponding to the input image from the first trained model candidate for the specular reflection image and the first trained model candidate for the diffuse reflection image. Select a ready-made model,
The second estimating unit calculates the second trained model corresponding to the input image from the second trained model candidate for the specular reflection image and the second trained model candidate for the diffuse reflection image. select a ready-made model,
The evaluation system according to any one of claims 6-8. further comprising an evaluation unit that evaluates the surface based on at least one of the first estimated value and the second estimated value;
The acquisition unit acquires a plurality of input images by dividing an original image showing the surface of the object,
The first estimation unit calculates the first estimated value for each of the plurality of input images,
The second estimation unit calculates the second estimated value for each of the plurality of input images,
The evaluation unit evaluates each of the plurality of partial regions of the surface corresponding to the plurality of input images based on at least one of the first estimated value and the second estimated value.
Evaluation system according to any one of claims 1 to 9. The acquisition unit acquires the input image showing the painted surface,
The first estimation unit calculates the first estimated value representing the coating state of the surface,
The second estimation unit calculates the second estimated value representing the painting state of the surface.
The evaluation system according to any one of claims 1-10. an acquisition unit that acquires one of a specular reflection image and a diffuse reflection image as an input image of a surface of an object;
A trained model for specular reflection images that calculates an estimated value for the surface based on the specular reflection image, and a trained model for diffuse reflection images that calculates the estimated value for the surface based on the diffuse reflection image. a selection unit that selects a trained model corresponding to the input image from among
an estimating unit that inputs the input image to the selected trained model and calculates the estimated value for the surface;
A rating system with An evaluation method performed by an evaluation system comprising at least one processor, comprising:
obtaining an input image of the surface of the object;
inputting the input image into a first trained model to compute a first estimate for the surface;
inputting the input image into a second trained model different from the first trained model to calculate a second estimate for the surface;
evaluation methods, including obtaining an input image of the surface of the object;
inputting the input image into a first trained model to compute a first estimate for the surface;
inputting the input image into a second trained model different from the first trained model to calculate a second estimate for the surface;
An evaluation program that causes a computer to run

Images