JP2007333709A - Inspection standard determination method and device, and appearance inspecting apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce the number of trial times and a burden of an inspector for obtaining a psychological measurement curve by using a simple constitution, in order to able to determine an inspection standard with high effect, accuracy and simply at low cost, and in order to make a visual inspection automated by using the inspection standard, in an inspection standard determining method, an inspection standard determining apparatus and a visual inspecting apparatus. <P>SOLUTION: The inspection standard determining apparatus 1 determines the inspection standard which is used for determining as to whether a defect candidate area of a sample is a defect, by using the appearance feature quantity of the defect candidate area, based on the psychological measurement curve obtained, by digitizing the response of the inspector M to a stimulation consisting of the feature quantity. The psychological measurement curve is obtained by using a staircase method. The inspection standard determining apparatus 1 comprises an image forming means 2, an image presenting means 3; an input means 4, a presentation image determining means 5, a statistical processing means 6, a memory means 7, an external input/output means 8, and a central control means 10. By using the psychological measurement curve, human sensibilities, i.e., perceptive and sensory characteristics can be added to the inspection standards. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention is a reference for determining whether or not a defect candidate area of a sample has a defect based on the size of an appearance feature amount, and determines a reference that reflects human perception and sensory characteristics. The present invention relates to an inspection standard determination method, an inspection standard determination apparatus, and an appearance inspection apparatus that performs defect inspection of a sample using an inspection standard determined by the inspection standard determination apparatus.

  2. Description of the Related Art Conventionally, there is a defect inspection process for performing defect inspection for detecting a defective part of a product in quality control of an industrial product or the like, for example, inspection for determining that there is a defect if an attribute feature value of a defect is greater than a certain reference value. In order to guarantee the quality of industrial products, defect inspection of materials is performed on various materials and shapes, and both inspection efficiency and inspection accuracy are required. Most of these inspection processes require continuous tension that depends on human perception and sensation. Therefore, in order to prevent individual differences, physical condition, fatigue, fluctuations in inspection standards, oversights, etc. of inspectors, attempts to automate inspections are actively made from an early stage. In order to realize inspection automation, it is necessary to efficiently extract human perception and sensory characteristics in the extraction of parameters representing the characteristics of defect areas and the determination of inspection standards. I'm in a hurry.

  That is, a reference value for determining that there is a defect is necessary for defect inspection of industrial products. And if the reference value is based on human perception and sensory characteristics, as well as human sensitivity and preference, for example, depending on whether the surface is flawed, painted, like or dislike, In order for the product to be accepted by the user, it is necessary to determine the reference value in consideration of the user's judgment criteria.

  By the way, in the inspection process, the inspection of an inspection sample such as an industrial product, directly or via an inspection device, by an inspector to determine whether the inspection sample (hereinafter referred to as a sample) is defective is a sensory stimulus. It can be regarded as a human response. The feature quantity as a reaction target in the sample becomes a perceptual stimulus for humans. When the intensity of a sensory stimulus (hereinafter referred to as stimulus intensity) is represented by a one-dimensional variable, the response to the sensory stimulus generally shows a trend of a continuous monotonically increasing function with respect to the stimulus intensity. Moreover, this response generally occurs stochastically with respect to the stimulus intensity. The response to the stimulus intensity is a probability curve, which is called a psychometric curve.

  The determination of the inspection standard for determining that there is a defect based on human perception and sensory characteristics as described above is performed based on a psychological measurement curve. Therefore, when automating such defect inspection, it is necessary to obtain this psychological measurement curve for each feature amount of the sample. Such determination of inspection standards requires a different procedure from the determination of physical inspection standards that do not require human sensitivity, such as the mechanical strength and dimensional accuracy of materials.

The psychometric curve is measured based on a perceptual psychometric method, and various methods have been studied for a long time. For example, there are classically well-known measurement methods roughly divided into three methods, a constant stimulation method, a limit method, and an adjustment method devised by Fechner. A method using a probit method (probability unit method) in combination with one of these constant stimulus methods has been proposed as a method of measuring a psychometric curve for defect inspection by visual inspection (for example, non-patent literature). 1).
Masao Nakagawa, Hidetoshi Nakayasu, “Evaluation method of defect detection probability by multi-modality” material, Vol. 54, no. 1, pp2-7, 2005

  However, the method for measuring a psychological measurement curve using the constant stimulus method and the probit method as shown in Non-Patent Document 1 described above has the following problems. That is, in this method, when comparing a pair of stimulus intensities, a comparison is made by a number of trials that are repeated randomly, and the comparison of the pair is performed by a predetermined number of sets. Since the psychological measurement curve is obtained by obtaining the response, there is a problem that the number of trials is large and the efficiency is low. In addition, since the number of trials is large, there is a problem that the burden on the inspector who performs the comparison experiment in order to obtain the psychological measurement curve is large.

  The present invention solves the above-mentioned problems, and with a simple configuration, the number of trials for obtaining a psychological measurement curve and the burden on the inspector can be reduced, and it is highly efficient, accurate, and inexpensive and simple. It is an object of the present invention to provide an inspection standard determination method, an inspection standard determination apparatus, and an appearance inspection apparatus using the inspection standard that can determine an inspection standard.

  In order to achieve the above object, the invention of claim 1 determines a reference for the feature amount for determining whether or not the region is a defect based on the size of the appearance feature amount of the defect candidate region of the sample. A method for determining an inspection standard, wherein a standard sample having a defect candidate area in which the feature amount is set to a predetermined standard value, and a value different from each other by gradually increasing or decreasing the feature amount by a predetermined amount from the standard value Preparing a plurality of target samples having defect candidate regions, and presenting the standard sample and one of the target samples to an inspector, and comparing the feature amounts of the defect candidate regions of the two samples with each other And when the answer is that the feature value of the target sample is larger than the feature value of the target sample presented, When the target sample having the feature value of the threshold value is presented to the inspector together with the standard sample, and the answer is that the feature value of the target sample is smaller, the feature value of the presented target sample A step of presenting a target sample having a feature value larger by one level to the inspector together with the standard sample, and comparing the feature amounts of the defect candidate areas of the two presented samples with each other and returning the magnitude And by repeating the process of presenting the target sample according to the answer of the inspector a predetermined number of times, the frequency distribution of the answer to the feature amount is obtained, and the statistical parameters including the average value and the standard deviation of the frequency distribution are calculated. And determining an inspection standard that quantifies the visual judgment of the inspector for the feature amount based on the statistical parameter. It is intended.

  According to a second aspect of the present invention, in the method for determining an inspection standard according to the first aspect, a plurality of frequency distributions obtained by responses of a plurality of inspectors are acquired, and a plurality of statistical parameters calculated from the frequency distributions are statistically processed. The method further includes a step of determining an inspection standard relating to the feature amount.

  The invention according to claim 3 is an inspection standard determination device for determining a criterion relating to the feature amount for determining whether or not the region is a defect based on the size of the appearance feature amount of the defect candidate region of the sample. An image of a standard sample having a defect candidate area in which the feature amount is set to a predetermined standard value, and a defect candidate area in which the feature amount is increased or decreased stepwise from the standard value by a predetermined amount to be different from each other. An image generation means for generating images of a plurality of target samples, an image of a standard sample generated by the image generation means and one of the images of the target sample are presented to an inspector, and the two sample images presented An image presentation means for prompting the inspector to input a comparison result of the magnitude of the feature amount of the defect candidate area, and for the inspector to input the comparison result answer A feature quantity having a value smaller by one step than the feature quantity of the target sample presented in the case where the reply of the comparison result input by the input means and the input means is a reply that the feature quantity of the target sample is larger And when the answer of the comparison result is a reply that the feature quantity of the target sample is smaller, the feature quantity having a value one step larger than the feature quantity of the target sample presented Presenting image determining means for selecting an image of a target sample having the image of the standard sample and an image of the target sample selected by the presenting image determining means to the inspector using the image presenting means Obtaining a frequency distribution of responses to the feature amount by repeating a predetermined number of times to obtain an answer of the comparison result input by the input means, and obtaining an average value of the frequency distribution Statistical processing means for calculating a statistical parameter including a quasi-deviation, and the statistical processing means determines an inspection standard that quantifies the visual inspection of the inspector with respect to the feature amount based on the statistical parameter To do.

  According to a fourth aspect of the present invention, in the inspection standard determining apparatus according to the third aspect, the statistical processing means holds a plurality of statistical parameters by a plurality of inspectors and performs statistical processing of the plurality of statistical parameters. Is used to determine the inspection standard in which the visual judgment of a plurality of inspectors is digitized.

  According to a fifth aspect of the present invention, a defect inspection of a sample is performed by determining whether or not a defect candidate region of the sample is a defect by using the inspection standard determined by the inspection standard determining device according to the third or fourth aspect. It is an appearance inspection apparatus that performs

  According to the first aspect of the present invention, a technique based on the so-called staircase method (step method, up-down method), which can obtain an estimated average with a relatively small number of trials and the number of samples, is applied to defect inspection. Since the psychological measurement curve can be obtained, the number of trials for obtaining the psychological measurement curve and the burden on the inspector can be reduced, and the inspection standard can be easily determined with high efficiency, high accuracy and low cost.

  According to the invention of claim 2, since the inspection standard is determined based on the perception and sensory characteristics of a plurality of inspectors, a highly accurate inspection standard without bias can be obtained.

  According to the invention of claim 3, the psychological measurement curve is obtained by applying the steer case method, which can obtain an estimated average with a relatively small number of trials and the number of samples, to defect inspection. It is possible to reduce the number of trials to obtain and the burden on the inspector, and to easily determine the inspection standard with high efficiency, high accuracy and low cost.

  According to the fourth aspect of the invention, since the inspection standard is determined based on the perception and sensory characteristics of a plurality of inspectors, a highly accurate inspection standard without bias can be obtained.

  According to the invention of claim 5, since defect inspection by appearance inspection can be automated using inspection standards obtained efficiently, with high accuracy and at low cost, individual differences, physical condition, fatigue, inspection standards of inspectors It is possible to eliminate problems such as false inspections due to fluctuations and oversights.

  Hereinafter, an inspection standard determination method, an inspection standard determination apparatus, and an appearance inspection apparatus using the inspection standard according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a block configuration of an inspection standard determination apparatus 1 according to the embodiment. The inspection standard determination device 1 determines whether or not the region is a defect based on the appearance feature amount of the defect candidate region of the sample, for example, the size of a scratch or crack, the difference in color, etc. The inspection standard regarding the feature amount is determined based on the psychological measurement curve. That is, by using a psychological measurement curve, human sensitivity, that is, perceptual characteristics and sensory characteristics can be added to the inspection standard.

  The inspection standard determination device 1 uses a so-called staircase method (also called a staircase method, an up-down method, or an up-down method) to obtain a psychological measurement curve. The psychological measurement curve is a result of quantifying the response of the inspector M to the stimulus composed of the external feature amount of the defect candidate area. The steer case method is characterized by a method for presenting an image of a defect candidate region of a sample to an inspector M, and a psychological measurement curve is obtained efficiently and accurately with a relatively small number of trials and the number of samples. be able to.

  In order to execute the processing as described above, the examination standard determination apparatus 1 includes an image generation unit 2, an image presentation unit 3, an input unit 4, a presentation image determination unit 5, a statistical processing unit 6, and a storage unit 7. And an external input / output means 8 and a central control means 10.

  The image generating means 2 includes an image of a standard sample having a defect candidate area with a feature value set to a predetermined standard value, and a defect candidate having a feature value increased or decreased step by step from the standard value to a different value from each other. Generating images of a plurality of target samples having regions.

  The image presenting means 3 displays the standard sample image and one of the target sample images generated by the image generating means 2 on a display 31 such as a CRT or LCD, and presents and presents them to the inspector M. The inspector M is urged to input an answer of the comparison result of the feature amount of the defect candidate areas of the two sample images.

  The input unit 4 is a unit for the inspector M to input an answer of the comparison result to the inspection standard determination device 1. As the input means 4, for example, an input user interface in a normal computer such as a mouse, joystick, light pen, and keyboard can be used, or a dedicated input switch such as a voice input or a manual switch can be provided and used. .

  The presented image determination means 5 has a value one step smaller than the feature value of the target sample presented last time when the answer of the comparison result input by the input means 4 is a reply that the feature value of the target sample is larger. Select an image of a target sample having a feature quantity, and if the answer of the comparison result is a reply that the feature quantity of the target sample is smaller, the feature quantity that is one step larger than the feature quantity of the target sample presented previously Select an image of the target sample with

  The statistical processing means 6 presents the image of the standard sample and the image of the target sample selected by the presentation image determination means 5 to the inspector M using the image presentation means 3 and answers the comparison results input by the input means 4. Is acquired a predetermined number of times. Thus, the statistical processing means 6 obtains the frequency distribution of responses to the feature quantity, calculates a statistical parameter including the average value and standard deviation of the frequency distribution, and checks the feature quantity based on the statistical parameter. The inspection standard in which the visual judgment of the worker M is digitized is determined. The psychological measurement curve described above is uniquely determined by these statistical parameters, as will be described later.

  The statistical processing means 6 holds a plurality of statistical parameters by a plurality of inspectors M by using the storage means 7 and the like, and performs a statistical processing of the plurality of statistical parameters to visually check the plurality of inspectors M. It is possible to determine the inspection standard in which the judgment is quantified.

  The storage means 7 stores data such as sample images to be presented, answers to comparison results that are responses of inspectors, software for operating the apparatus, and the like. The data stored in the storage unit 7 is read out as necessary.

  The external input / output means 8 inputs processing commands, data, software, and the like from an external device or a user of the apparatus to the inspection standard determination device 1 and determines the statistical parameters and inspection standards that are processing results as the inspection standard. It is means for outputting from the apparatus 1 to an external device. The external input / output means 8 can use a user interface for input in a normal computer such as a mouse or a keyboard as input means, and can use a wired or wireless network system, magnetic medium, optical medium, IC chip as input / output means. An input / output device using a medium such as a medium can be used.

  The central control means 10 is a means for controlling each of the means 2 to 8 described above, and includes a microprocessor (MPU), a peripheral processing chip, and the like. That is, the central control means 10 is configured as an electronic computer having a general configuration including a CPU, a memory, an external storage device, a display device, an input device, and the like, and a set of processes or functions on the electronic computer. Can do.

  Next, the outline of the processing will be described with reference to FIG. 2 in addition to FIG. FIG. 2 is a flowchart of the inspection standard determination method according to the embodiment. In the inspection standard determination method, first, the image generating means 2 and a standard sample having a defect candidate region having a feature amount set to a predetermined standard value and a feature amount are gradually increased or decreased from the standard value by a predetermined amount. A plurality of target samples having defect candidate areas having different values are prepared (# 1).

  Next, a single attempt is made to stimulate and seek a response. That is, the image presenting means 3 presents the standard sample and one of the target samples to the inspector M, and compares the feature amounts of the defect candidate areas of the two samples with each other to answer the magnitude (# 2). The answer by the inspector M is acquired by the input means 4.

  Next, the presented image determination means 5 has a feature value that is one step smaller than the feature value of the target sample presented last time when the answer is that the feature value of the target sample is larger. Select the target sample, and if the answer is that the feature quantity of the target sample is smaller, select the target sample that has a feature quantity that is one step larger than the feature quantity of the target sample presented last time. The target sample to be presented is changed (# 3).

  Next, the central control means 10 determines whether or not the number of trials for obtaining a response has reached a predetermined number (# 4). If the number of trials has not yet reached the predetermined number (No in # 4), The process returns to step # 2.

  In the second and subsequent trials, the image presenting means 3 presents the changed target sample to the inspector M together with the standard sample, and compares the feature amounts of the defect candidate areas of the two presented samples with each other. (# 2).

  The central control means 10 repeats steps # 2, # 3, and # 4 described above, and if the number of trials reaches a predetermined number (Yes in # 4), the process proceeds to step # 5.

  As described above, the statistical processing means 6 repeats the process of presenting the target sample according to the answer of the inspector M a predetermined number of times, thereby obtaining the frequency distribution of the answers to the feature amount, and the average value of the frequency distribution and Statistical parameters including standard deviation are calculated (# 5).

  Finally, the statistical processing means 6 determines an inspection standard that quantifies the visual inspection of the inspector M with respect to the feature amount based on the statistical parameter (# 6). The inspection standard that has been determined is a standard for determining whether or not a defect candidate area of a sample has a defect based on the size of the appearance feature quantity, and automates defect inspection by visual inspection. Can be used to

  In this inspection standard determination method, the above-described processing from step # 1 to step # 6 is repeated for a plurality of inspectors M under the control of the central control means 10, and a plurality of inspectors M obtained are obtained. And a plurality of statistical parameters calculated from the frequency distribution can be respectively held using the storage means 7 or the like. The statistical processing means 6 performs a process of statistically processing a plurality of statistical parameters calculated from the respective frequency distributions, and determining an inspection standard in which the visual judgments of the plurality of inspectors M regarding the feature amount are quantified. be able to.

  Hereinafter, the psychological measurement curve acquired by the above-described inspection standard determination apparatus 1 and the inspection standard determination method, the relationship between the psychological measurement curve and the inspection standard, and the staircase method used for obtaining the psychological measurement curve will be described.

(Psychological measurement curve)
When an image (inspection image) including a defect candidate area is presented to the inspector M, the image is used for the inspector M observing the inspection image to determine whether the object at the inspection site is defective. The requested sensory stimulus becomes a response, and the response of the examiner M becomes a response to the sensory stimulus. Such a response becomes a probability curve indicating a response to the stimulus intensity x, that is, a psychological measurement curve, as shown in FIG. 3, when the feature quantity representing the intensity of the presented perceptual stimulus is represented by a one-dimensional variable. .

  The psychological measurement curve shown in FIG. 3 is a characteristic quantity x (for example, the intensity of the intensity expressed by the inspector M, for example, when the perceptual stimulus has a single attribute (for example, the defect when the defect is a defect). This is regarded as the detection probability P that is recognized (or detected) that the size x) of the scratch is larger than a certain value x. The detection probability P is a function of the stimulus intensity x and is expressed as P = Φ (x).

  In perceptual psychology and sensitivity tests, normally, a cumulative distribution function according to a normal distribution is preferably used as the distribution representing the detection probability Φ (x). The detection probability Φ (x) is given by the following equation (1). The statistical parameters μ and σ represent the mean and standard deviation of x, respectively, Φ (*) is a cumulative distribution function, and φ (*) is a probability density function according to a normal distribution.

(Psychometric curve and inspection standard)
The psychological measurement curve shown by the above equation (1) and FIG. 3 is a characteristic curve that reflects human perception characteristics and sensory characteristics with respect to the feature amount x obtained by quantifying a single perceptual stimulus. In this psychometric measurement curve, P = 0.5 means that the probability of responding to the perceptual stimulus intensity x giving a value of P = 0.5 is 50%, and the probability of not responding is 50%.

The response probability of 50% means that the inspector M recognizes a defect area as a defect or recognizes that it is not a defect, or the probability is half. Therefore, the sensory stimulation intensity _{x 50} giving a value of P = 0.5, (also referred to as threshold) inspection standard _{x th,} that _is, to _x th ₌ x _50,. In addition, a certain margin value Δx may be subtracted from the perceptual stimulus intensity x ₅₀ to obtain a stricter reference value, x _th = x ₅₀ −Δx.

By giving such an inspection standard x _th to an appearance inspection apparatus capable of measuring the perceptual stimulus intensity x, it is possible to perform defect inspection based on a judgment standard that takes human sensitivity into consideration, and to automate the defect inspection. it can.

(Stair case method)
Inspection standard determination apparatus 1 and the inspection standard determination method according to the embodiment, as a method for obtaining the psychometric curve described above, gradually changing the a standard sample set the stimulation intensity in advance to a constant value x ⁰ sensory stimulus intensity x A so-called pair comparison method is used in which one of the plurality of target samples is paired and the inspector M compares the pair of features.

The paired comparison method, x is from stronger than x ^0, or, x is from weaker than x ^0, that, according to the inspector M, alternatively responses (i.e., responses by the so-called 2 method) based on the accumulated data of Thus, the parameters μ and σ of the above equation (1) are obtained. Since the probability distribution that the psychological measurement curve follows is assumed to be a normal distribution, the psychological measurement curve is uniquely determined by the parameters μ and σ.

  In the staircase method, in the paired comparison described above, until the response of the inspector M is reversed, the stimulus intensity of the target sample is changed to the stronger or weaker one after another and presented to the inspector M. At the point of time, the trial of presenting the direction of intensity change in reverse is repeated for a predetermined number of trials, and the response from the inspector M is represented as a time series chart (referred to as a staircase diagram) of the response to the change of the stimulus intensity x. get. Acquisition of such a staircase diagram is performed according to the procedure of steps # 1 to # 5 shown in FIG. 2 and the description thereof.

  The response of the inspector M described above is inverted and raised (up and down) many times during a predetermined number of trials. In other words, when the number of inversions is small, it is inappropriate to set the standard sample or the stepwise change of the target sample.

  In the steer case method, a paired comparison with a target sample far away from an appropriately set standard sample does not occur. Further, the measurement is completed within a predetermined number of times. Therefore, a psychological measurement curve can be efficiently obtained with a relatively small number of trials and the number of samples.

  The psychological measurement curve obtained by using the paired comparison method and the staircase method (this is also referred to as a sensory test experiment) is performed, for example, as shown in FIG. And fixed. The inspector M, for example, looks at the pair of sample images 32 presented on the presentation display 31 placed in front of 60 cm in front of the eye, compares the perceptual stimulus intensity x, and inputs the comparison result to the input means such as the mouse 33. Use 4 to answer. The psychological measurement curve is obtained by accumulating the answers.

(Sample image and display)
For example, as shown in FIG. 5, the paired comparison sample images are presented side by side on the display 31. The left image is a standard sample a0 including a defect candidate area b having a standard stimulus intensity, and the right image is an image of a target sample a including a defect candidate area b having a predetermined stimulus intensity.

  The display 31 is provided with a sensory test start / end button 34 that can be operated by the mouse 33 and a button 35 for inputting a comparison result of large or small response to the test standard determination device 1. . The display 31 displays the scheduled total number of trials and the current number of trials, for example, 1/90.

(Method of sensory test)
The sensory test experiment by the staircase method is performed according to the following (i) to (v). The result of the sensory test experiment is obtained as a time series chart of responses, that is, a staircase diagram, as shown in FIG. Based on this chart, a frequency distribution with respect to the response is obtained, and a statistical distribution μ and σ for determining a psychometric measurement curve and a cumulative distribution function are obtained assuming a normal distribution for the frequency distribution. Note that the steps in FIG. 2 are associated with each other.

  (I) The initial stimulus intensity x and the change width d are set for the target sample a of the first trial, and the target sample a is displayed on the display 31 for display (# 1, # 2).

  (Ii) When the examiner M responds to the presented stimulus that the stimulus intensity x of the target sample a is “weaker” than the standard sample a0 (“◯” in FIG. 6), the target sample to be used in the next trial When the stimulation intensity x of a is increased (up) by d and conversely “strong” is responded (“×” in FIG. 6), it is decreased (down) by d (# 3).

  (Iii) Repeat in the same manner up to a preset repetition number N (# 4). As described above, the response step diagram shown in FIG. 6 is obtained. In FIG. 6, the horizontal axis represents the number of trials, and the vertical axis represents the stimulation intensity x. Responses for each number of trials are plotted by assigning ◯ when the examiner M determines that the stimulus of the target sample a is weak, and assigning × when the examiner M determines that the stimulus is strong.

  (Iv) For the result obtained as shown in FIG. 6, as shown in Table 1, a frequency distribution table relating to the number of ◯ and the number of x is created (# 5).

(V) For the above results, the smaller response frequency distribution (the frequency distribution for the response in the case of Table 1) out of the total number of ○ and the total number of × is adopted, and for the frequency distribution, the number of ○ or × for stimulation intensity x (the frequency) and f _i, using the following equation (2) to (4), calculates the average value S _N and the standard deviation σ of the stimulation intensity x (# 5).

However, the right side of the above equation (3), S ₀ uses the value of the stimulation intensity x for i = 0 in Table 1, decoding, if the number of ○ was f _i +, the number of × f _{When i} is used,-is used.

  In the above, the frequency distribution with a small number of responses is adopted for the following reason. For example, when the stimulus intensity x of the standard sample a0 is too large, the response is repeated up and down after a response in the down direction, and conversely, the stimulus intensity x of the standard sample a0 is small. If it is too long, the response repeats up and down at a certain value after a response in the up direction continues. Therefore, in any case, the smaller the number of responses, the better the symmetry of the frequency distribution, that is, the data with no bias.

  A psychological measurement curve obtained using the test standard determination method according to the embodiment described above, that is, a psychological measurement curve obtained using the staircase method, and a psychological measurement curve obtained using the probit method which is a conventional method, Compare The comparison target is the three attributes of defect candidates (described later), and the numerical comparison is performed by using the statistical parameters μ and σ of the psychometric measurement curve, and the average value and coefficient of variation of subjective equivalent value PSE (described later) by a plurality of inspectors M CV (coefficient of variation) is performed.

(Stimulation intensity digitization)
As three defect candidates, the attributes of surface defects in a fiber reinforced mat (mat FRP) product are taken up. Three attributes of the surface defect, that is, the size A, the flatness R, and the color density V are digitized so that they can be manipulated step by step. Surface defect area size (defect area area) A, flatness (defect area major axis to minor axis ratio, aspect ratio) R, color density (gray scale of defect area) obtained from the image of the mat FRP product The (concentration) V has an average value and a standard deviation as shown in Table 2.

Therefore, the defect candidate area is modeled by an ellipse using the average value and the standard deviation in Table 1. That is, the values of the stimulus intensity x _i ⁰ of the standard sample and the stimulus intensity x _{ij of the} target sample (where i = A, R, V, j = 1, 2,...) Are as follows: And expressed by mathematical formulas.

_A value corresponding to the length of the ellipse major axis is used as the feature quantity xA representing the size A as the first attribute. The stimulation intensity x _A ⁰ of the standard sample a0 is expressed by the following expression (5), and the stimulation intensity x _Aj of the target sample is changed by the following expression (6).

However, μ _A and σ _A are the average and standard deviation of the areas shown in Table 2. h _A is a coefficient, and h _A = 0.019 was determined by preliminary experiments. The stimulus intensity _xAj of the target sample changes with a change width d determined by a change of j = 1, 2 _,.

As the characteristic amount x _R representing the flatness R is the second attribute, and minus the aspect ratio of the minor axis with respect to the long axis of the ellipse (flatness) from 1, expressed by the following equation. Here, μ _R and σ _R are the average and standard deviation of flatness shown in Table 2. h _R is a coefficient, and h _R = 0.055 was determined by preliminary experiments. The reason why the flatness is subtracted from 1 is to increase the stimulation intensity as the feature amount increases.
^{_{x R 0 = (1-μ}} R),
_{x Rj = (1-μ R} ) + σ R h R (j-3.5).

As the feature quantity x _V representing the color density (gray scale density) V is a third attribute, the white 0, as one black, expressed by the following equation. Here, mu _V, sigma _V is the average and the standard deviation of color density as shown in Table 2. h _V is a coefficient, and h _V = 0.028 was determined by preliminary experiments.
x _V ⁰ = μ _V ,
_{x Vj = μ V + σ V} h V (j-3.5).

The stimulation intensity x _i ⁰ of the standard sample and the stimulation intensity x _{ij of the} target sample set as described above (where i = A, R, V and 7 levels, j = 1, 2,..., 7) Table 3 shows the values. 7, the stimulation intensity with respect to the area A _x ^{A 0} and _{x Aj (j = 1,2, ...} , 7) indicates one standard sample a0 and seven target sample a containing defect candidate area b for presenting .

Further, in the examples, seven adult men and women aged 20 or over who have naked eyes or corrected visual acuity from 1.0 to 1.5 were designated as inspectors M (subjects, panels). Attempts _{n s} in staircase method was _n s = 90, respectively. In the probit method of the comparative example, for the stimulation intensity of the level number n _d = 7 shown in Table 3, a random trial with the number of repetitions n _r = 30 is performed, and the total number of trials n _p = n _d × n _r = 7 × 30 = 210 times.

(Staircase diagram)
FIG. 8, FIG. 9, and FIG. 10 show step diagrams of response results of the inspector M1 for the first attribute, the second attribute, and the third attribute, respectively. The vertical axis represents the stimulus intensity of each attribute, and the horizontal axis represents the number of trials. Horizontal thin line in the figure is a stimulation intensity x values of the standard samples a0, thick line is the average value of the stimulation intensity x which is a response to the n s ₌ 90 trials. 8 and 9, the number of ○ and × is the same number 45, and in FIG. 10, the number of ○ is 46 and the number of × is 44. The number of levels of stimulus intensity appearing on the staircase chart is 6 levels in FIG. 8, 7 levels in FIG. 9, and 5 levels in FIG. In addition, every attribute fluctuates up and down across the standard sample, and the number of responses to the stimulus intensity far from the average value is small. When the stimulation intensity is close to the average value, the number of up and down fluctuations is large, and the number of ○ and X is large. The results of the other six examiners M were similar to those in FIG.

  Further, in each of the above-mentioned staircase diagrams, the fact that the number of ○ and X is almost the same means that there is no bias in the number of responses of “strong” and “weak” stimulation intensity. Furthermore, a large number of up and down fluctuations observed around a level close to the average value indicates that the frequency of responses is distributed around the average value. These represent the characteristics of the steer case method when an appropriate sample is set.

(Psychological measurement curve)
A frequency distribution is obtained from the staircase diagram obtained as described above, a statistical parameter is derived, and a psychological measurement curve is determined. The psychological measurement curves for the inspector M2 are shown in FIG. 11, FIG. 12, and FIG. 13, and the psychological measurement curves for the other inspectors M3 are shown in FIGS. The statistical parameters were calculated using the above-described equations (3), (4), and (5). The solid line in FIGS. 11 to 16 shows the psychological measurement curve st by the steer case method in the test standard determination method according to the embodiment, and the broken line shows the psychological measurement curve pb by the probit method performed for comparison.

  The psychological measurement curve st by the staircase method shows a tendency to draw a gentler curve than the psychological measurement curve pb by the probit method. This tendency is the same for the other inspectors M.

  Table 4 shows average values and coefficient of variation CV by both the steer case method and the probit method corresponding to FIGS. 11, 12, and 13, and Table 5 corresponds to FIGS. 14, 15, and 16. The average value and the coefficient of variation CV by both the staircase method and the probit method are shown.

  The results of the average values obtained by the steer case method and the probit method show values that are substantially coincident with each other. As for the value of the variation coefficient CV, for the three attributes of size A, flatness R, and color density V, the result of the steer case method is larger than the result of the probit method. Further, in the comparison for each inspector M, the personality of the inspector is reflected, and the psychological measurement curve st by the inspector M2 has a steep slope with three attributes compared to the psychological measurement curve st by the inspector M3. It is a simple curve.

  11 to 16 described above, the psychological measurement curve based on the steer case method has a gentler slope than the psychological measurement curve based on the probit method. Therefore, the variation coefficient CV of the steer case method is larger than the variation coefficient CV of the probit method. Is also big. Further, in the result of the steer case method, the coefficient of variation CV (Table 4) by the inspector M2 is smaller than the coefficient of variation CV (Table 5) by the inspector M3. Due to variation.

  In summary, according to the test standard determination method according to the embodiment using the staircase method, a psychological measurement curve having an average value equivalent to the psychological measurement curve by the probit method is obtained.

(Evaluation by subjective equivalence)
The stimulus intensity x ₅₀ corresponding to P = 0.5 in the psychological measurement curve is also called a subjective equivalent value PSE (point of subjective equality), and this value is also a statistic. Therefore, for the attribute, the inspector M, and the staircase method and the probit method, the average value of the subjective equivalent value PSE and the coefficient of variation CV are obtained, respectively, and the staircase method and the probit method in the inspection standard determination method according to the embodiment are obtained. Compare results with.

  Table 6 shows the average value and the coefficient of variation CV of the subjective equivalent value PSE of all the inspectors M according to the inspection standard determination method (stair case method) and the probit method according to the embodiment. The average values obtained by both methods are substantially the same, and the coefficient of variation CV indicates that the result of the staircase method is larger than the result of the probit method. For reference, an image of a target sample having the average subjective equivalent value PSE shown in Table 6 is shown in FIG. The average value of the subjective equivalent value PSE obtained by both methods for the three attributes is as shown in Table 6 in numerical terms, but looks almost the same from FIG.

  As shown in Table 6, the results of the average values of the subjective equivalent value PSE are substantially the same for both methods, and the results obtained by the steer case method are substantially the same as the results obtained by the probit method. Regarding the variation coefficient CV, it can be said that the probit method has higher estimation accuracy than the steer case method. The result regarding the coefficient of variation CV is a result according to conventional knowledge.

(About efficiency)
The staircase method can efficiently obtain a psychometric measurement curve with a smaller number of trials than the probit method based on the constant stimulus method. Number of repetitions n _r in probit methods are usually required more than 30 times. In the above-described embodiment, the number of trials n _s in the staircase method is n _s = 90, and in the probit method of the comparative example, the number of trials n _p = n _d × n _r = 7 × 30 = 210. The ratio between the number of trials n _s by the steer case method and the number of trials by the probit method is n _p / n _s = 90/210 = 0.43.

  That is, in the test standard determination method according to the embodiment, the psychological measurement curve can be efficiently obtained with the number of trials of 43% or less of the conventional method. The test standard determination method according to the embodiment, and the test standard determination apparatus using the method can reduce the number of trials for obtaining the psychometric measurement curve, reduce the burden on the inspector, and with high efficiency and accuracy, and Inspection standards can be easily determined at low cost.

  Next, with reference to FIG. 18, an appearance inspection apparatus 11 according to the embodiment that performs an appearance inspection using the inspection standard determined by the inspection standard determination apparatus using the inspection standard determination method according to the above-described embodiment will be described. To do. The appearance inspection apparatus 11 includes an inspection reference input means 12 that receives the inspection reference determined by the inspection reference determination apparatus 1 described above, an image input means 13 that inputs a surface image of a sample 20 to be subjected to an appearance inspection, and an image input. A defect determination unit 14 that extracts a defect candidate region from the image input by the unit 13 and determines whether the defect candidate region is a defect based on the inspection standard input by the inspection standard input unit 12. I have.

  A surface image of the sample 20, for example, the above-described mat FRP product is captured by the camera 21. The surface image of the sample 20 is input to the appearance inspection apparatus by scanning the entire surface of the inspection region in a batch or by relatively moving the sample 20 and the camera 21 together. .

  The inspection standard input means 12 is capable of receiving data transmitted via a wired or wireless network connection with the inspection standard determination device 1 and receiving data using a portable medium such as a magnetic medium or a medium using an IC chip. Configured. In the inspection standard determination apparatus 1 described above, the external input / output means 8 is used to output inspection standards. The appearance inspection apparatus 11 performs defect determination inspection on these images by the defect determination means 14.

  Further, the appearance inspection apparatus 11 may transmit the image of the defect candidate area of the sample 20 to the inspection standard determination apparatus 1 after accumulating or in real time. For this transmission, the inspection standard input means 12 can be extended for output. The inspection standard determination device 1 can perform predetermined processing on the image data transmitted from the appearance inspection device 11 and present it to the inspector M, and can determine the inspection standard based on the above-described inspection standard determination method. .

  The present invention is not limited to the above-described configuration, and various modifications can be made. For example, the inspection standard determination apparatus 1 and the inspection standard determination method can be applied to inspections and evaluations depending on human perception and sensitivity, and appearance inspection of the above-described mat FRP products, various exterior materials, surface processed products, and the like. In addition, the inspection standard determination device 1 and the inspection standard determination method that are preferably used for the above are applicable not only to visual inspection and evaluation, but also to inspection and evaluation depending on human perception and sensitivity. For example, the present invention can be extended and applied as an apparatus and method for setting a medical effect degree and setting inspection standards for hearing, touch, touch, taste, and the like.

The block lineblock diagram of the inspection standard deciding device concerning one embodiment of the present invention. The flowchart of the inspection standard determination method which concerns on one Embodiment of this invention. A diagram of a general psychometric curve. The figure of the inspector who is responding using the method and apparatus same as above. The figure of the sample presentation screen used using the method and apparatus same as the above. The staircase figure which shows the response of the inspector acquired using the method and apparatus same as the above. The figure which shows the example of the target sample and standard sample which are used in the inspection standard determination method and determination apparatus same as the above. The staircase figure which shows the other example of the response | inspection of the inspector obtained using the method and apparatus same as the above. The staircase figure which shows the further another example of the inspector's response obtained using the method and apparatus same as the above. The staircase figure which shows the further another example of the inspector's response obtained using the method and apparatus same as the above. The figure which shows the example of the psychological measurement curve obtained using the method and apparatus same as the above. The figure of the other example of the psychological measurement curve obtained using the method and apparatus same as the above. The figure of the further another example of the psychological measurement curve obtained using the method and apparatus same as the above. The figure of the further another example of the psychological measurement curve obtained using the method and apparatus same as the above. The figure of the further another example of the psychological measurement curve obtained using the method and apparatus same as the above. The figure of the further another example of the psychological measurement curve obtained using the method and apparatus same as the above. Psychometric curve diagram showing determined using the method and apparatus same as above and determined by the probit method. The block block diagram of the external appearance inspection apparatus which concerns on one Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Inspection standard determination apparatus 2 Image generation means 3 Image presentation means 4 Input means 5 Presented image determination means 6 Statistical processing means 11 Appearance inspection apparatus 20 Sample a Target sample a0 Standard sample b Defect candidate area M, M1-M3 Inspector x _th Inspection criteria σ, μ Statistical parameters

Claims (5)

An inspection criterion determination method for determining a criterion relating to the feature amount for determining whether or not the region is a defect according to the size of an appearance feature amount of a defect candidate region of a sample,
A plurality of standard samples having defect candidate areas in which the feature amount is set to a predetermined standard value; and a plurality of defect candidate areas in which the feature amount is gradually increased or decreased from the standard value by a predetermined amount to have different values. Preparing a target sample;
Presenting the standard sample and one of the target samples to an inspector, causing the feature quantities of the defect candidate areas of the two samples to be compared with each other, and answering the magnitude thereof;
When the answer is that the feature quantity of the target sample is larger, a target sample having a feature quantity that is one step smaller than the feature quantity of the presented target sample is added together with the standard sample to the inspector. And when the answer is that the feature quantity of the target sample is smaller, a target sample having a feature quantity that is one step larger than the feature quantity of the presented target sample is selected as the standard sample. And the step of presenting to the inspector, comparing the feature quantities of the defect candidate areas of the two presented samples with each other and answering the magnitude,
A step of obtaining a frequency distribution of responses to the feature amount by repeating a step of presenting a target sample according to the response of the inspector, and calculating a statistical parameter including an average value and a standard deviation of the frequency distribution And comprising
An inspection standard determination method, comprising: determining an inspection standard obtained by quantifying the visual inspection of the inspector with respect to the feature amount based on the statistical parameter.   The method further comprises a step of obtaining a plurality of frequency distributions obtained by answers from a plurality of inspectors, statistically processing a plurality of statistical parameters calculated from the frequency distributions, and determining an inspection standard related to the feature amount. The inspection standard determination method according to claim 1. An inspection standard determination device for determining a criterion for the feature amount for determining whether or not the region is a defect according to the size of an appearance feature amount of a defect candidate region of a sample,
An image of a standard sample having a defect candidate area in which the feature quantity is set to a predetermined standard value, and a defect candidate area in which the feature quantity is increased or decreased stepwise from the standard value by a predetermined amount to have different values. Image generating means for generating images of a plurality of target samples;
The standard sample image generated by the image generation means and one of the target sample images are presented to the inspector, and the result of comparing the comparison result of the feature amount of the defect candidate area between the two sample images presented Image presentation means to prompt the inspector to input,
An input means for the inspector to input the answer of the comparison result;
When the answer of the comparison result inputted by the input means is a reply that the feature quantity of the target sample is larger, the target sample has a feature quantity that is one step smaller than the feature quantity of the presented target sample. And when the answer of the comparison result is that the feature quantity of the target sample is smaller, the target sample has a feature quantity that is one step larger than the feature quantity of the presented target sample. Presenting image determining means for selecting an image of;
Presenting the image of the standard sample and the image of the target sample selected by the presented image determining means to the inspector using the image presenting means, and obtaining an answer of the comparison result input by the input means. Statistical processing means for obtaining a frequency distribution of answers to the feature amount by repeating a predetermined number of times, and calculating statistical parameters including an average value and a standard deviation of the frequency distribution,
The statistical processing means determines an inspection standard that quantifies the visual judgment of the inspector for the feature amount based on the statistical parameter.   The statistical processing means retains a plurality of statistical parameters by a plurality of inspectors, and determines an inspection standard that quantifies the visual judgment of the plurality of inspectors by performing statistical processing of the plurality of statistical parameters. The inspection standard determining apparatus according to claim 3.   A defect inspection of the sample is performed by determining whether or not the defect candidate area of the sample is a defect using the inspection standard determined by the inspection standard determination device according to claim 3 or 4. Appearance inspection device.