JP2007183980A - Data setting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a data setting device for an image recognition device for setting optimal inspection data by dividing the surface color of a part to be inspected into several constituting colors, and calculating the statistic of each constituting color and the similarity of the constituting colors. <P>SOLUTION: The data setting device for setting the setting data of the image recognition device for recognizing the image for recognition of an object to be recognized based on setting data is provided with: a recognition image input part for designating a recognition area in the image for recognition where the object to be recognized is imaged; a constituting color processing part for extracting the position of the constituting colors of the object to be recognized based on the image in the recognition area, and for calculating statistic indicating the variation of the colors concerning the constituting colors; and an inspection data determining part for determining the extraction color range of the object to be recognized based on the statistic. The constituting color processing part comprises a constituting color statistic calculation part for calculating the statistic indicating the variation based on the value of each dimension of each pixel included in the constituting colors in a color space. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a data setting device and a data setting method for an image recognition device. In particular, the recognition of an image recognition device that captures an image for recognition of an object, specifies a color range (extracted color) of the object to be imaged at a specific position in the image, and extracts and recognizes the object The present invention relates to technology for creating business data.

  The appearance inspection machine is an apparatus that picks up an inspection image of an inspection object and performs pass / fail determination using inspection data (extracted color range, reference value). In general, an optical appearance inspection machine irradiates a test object with a white color and acquires an inspection image by capturing an image from a specific angle. Thereafter, an appearance inspection is performed on the inspection image by inspecting whether or not the inspection target portion is imaged with an appropriate color. In the appearance inspection of the printed circuit board, the board is imaged from above, and it is inspected whether the surface color of the component to be inspected is imaged at the component mounting position. As a result, it is possible to detect defects such as non-mounting or erroneous mounting of components.

  When using the appearance inspection machine, it is necessary to set “designation of inspection target location”, “extracted color range”, and “reference value (threshold)” in the inspection image as inspection data. The “appropriate color range” refers to the color range of the surface color of the inspection object. The color range is an upper limit value and a lower limit value for each dimension of the color space of the inspection image. Usually, the color space is represented by RGB of red, green, and blue. The reference value is for the area (extracted area) of the pixel corresponding to the color range, for example, the lower limit value of the ratio of the extracted area in the inspection target range specified by “designation of inspection target location”.

  In setting the above inspection data, a problem is a variation in the surface color of an inspection object caused by a variation in manufacturing. Even if the color variations are the same in appearance, there are variations in the same color. Moreover, even if it is the same test object, there is variation between samples. Therefore, it is necessary to set the extracted color range in accordance with this variation range.

  This is because if the range of the extracted color is narrow, it becomes “too much” to judge a non-defective product as a defective product, and if it is too wide, it causes “missing” to judge a defective product as a non-defective product. Similarly, in setting the reference value, if the threshold is too high, “too much” is caused, and if it is too low, “missing” is caused. “Overview” causes an increase in man-hours such as visual inspection in the subsequent process, and “missing” causes a decrease in inspection quality, and causes an increase in specific man-hours that are determined to be defective in an electric test or the like in the post-process. Therefore, setting inspection data is a general problem.

  As a known technique related to setting of a threshold value for distinguishing a recognition object from other objects, there is an invention described in Patent Document 1. In Patent Document 1, when a color of a recognition object of a color visual sensor is designated, the certainty factor of the color, a characteristic line of an area, and a characteristic line of another color are displayed, and the identification object is identified by the designated color. It is disclosed that it is possible to easily set a threshold value of a color designated as possible.

Further, as a known technique related to three-dimensional shape data processing, there is an invention described in Patent Document 2. In Patent Document 2, only a region having a color close to the skin color from the entire color image is determined based on the statistical processing result of the skin color of the color image as a model face region required when creating the face model. Discloses a three-dimensional shape data processing apparatus that can be automatically performed.
JP-A-7-44706 JP 2000-76454 A

  However, the data setting device of the conventional image recognition device has the following problems. The surface color of the inspection object is composed of several kinds of colors (component colors), and each component color has different variations. For this reason, the distribution of the entire surface color is a combination of variations in the constituent colors, resulting in a complex variation. This makes it difficult to know the variation tendency of the surface color necessary for optimizing the extraction color range. However, when the variation can be analyzed for each constituent color unit of the surface color, the variation tendency of the surface color can be easily known.

  Further, in the inspection data setting, a range including all the constituent colors cannot be set as the extraction color range. There are cases where the constituent color of a small area of a certain inspection object A and the main constituent color of another inspection object B are the same color. In this case, if the constituent color is included in the extraction range of a certain inspection object A, it is determined that the inspection object A is a non-defective product even when another inspection object B is mounted instead of the inspection object A. It is. That is, when setting inspection data, it is necessary to consider not only the variation of each component color but also the relationship (similarity) with the component colors of other inspection objects.

  Moreover, in patent document 1, since the component color of a recognition target object is analyzed and the variation of the component color is not calculated | required based on the variation between the samples of a recognition target image, the said subject cannot be solved. Patent Document 2 can extract a predetermined color of a face using a statistical processing result, but extracts constituent colors of parts other than the face and obtains variations between samples of the constituent colors. Therefore, the above problem cannot be solved.

  The present invention has been made in view of the above problems, and separates the surface color of the inspection target portion into several constituent colors, and calculates the statistic of each constituent color and the similarity between the constituent colors. An object of the present invention is to provide a data setting device for an image recognition device capable of setting optimum inspection data.

  According to the present invention, in the data setting device that sets the setting data of the image recognition device that recognizes the recognition image of the recognition target object based on the setting data, the recognition in the recognition image obtained by capturing the recognition target object A recognition image input unit for designating a region, and a component color processing unit for extracting a position of a component color of the recognition object based on an image in the recognition region and calculating a statistic indicating a color variation regarding the component color And an inspection data determination unit that determines an extraction color range of the recognition object based on the statistic, and the component color processing unit is configured for each dimension in the color space of each pixel included in the component color. There is provided a data setting device including a constituent color statistic calculation unit that calculates a statistic indicating variation based on a value.

  According to the present invention, by providing several inspection images that represent variations in the surface color of the inspection object, it is possible to analyze the tendency of the variation in the surface color and to set the extraction color range. Further, it is possible to select a component color to be extracted and set a reference value while confirming the extraction area of another inspection object. As a result, adjustment of inspection data has been conventionally performed over a long period of time, but inspection data can be adjusted in a short time. Conventionally, when determining inspection data, it has been necessary for a person to empirically determine the reference value setting and the selection of the extracted color for adjusting the balance of “missing” and “too much”. According to the invention, it is possible to adjust inspection data while checking the occurrence status of `` missing '' and `` overlooking '' when selecting a constituent color or setting a reference value, thereby realizing less skill in creating inspection data The

  FIG. 1 shows the principle of the present invention. As shown in FIG. 1, the data setting device includes a recognition image input unit 2, a constituent color processing unit 4, and an inspection data determination unit 6. The recognition image input unit 2 specifies the recognition area 14 of the recognition image 12 of the recognition object 10. The component color processing unit 4 extracts the component color of the recognition target object 10 based on the image in the recognition area 14, and calculates a statistic indicating the color variation related to the component color. The inspection data determination unit 6 determines the extraction color range of the recognition object based on the statistics. As described above, since the extracted color range of the recognition target is based on the statistic, it is possible to set optimal inspection data.

  FIG. 2 is a configuration diagram of a data setting device of the image processing device according to the embodiment of the present invention. The data setting device is a device for automatically setting the extraction color range of each recognition target object, for example, an electronic component (resistor, IC, capacitor, etc.) mounted on a substrate, such as a personal computer or a workstation. Each function of the data setting process is realized by a program executed on the hardware. As shown in FIG. 2, the data setting device includes an inspection image input unit 50, a constituent color processing unit 52, a constituent color storage unit 54, and an inspection data determination unit 56. The inspection image input unit 50 inputs a recognition target image captured by a camera (not shown) and a portion to be inspected which is a recognition area of the recognition target, and stores them in a storage device such as a disk (not shown). The recognition target image is an image in which the recognition target object is captured by a camera, and one or a plurality of recognition target objects are captured in one image. The recognition target image is represented by a value of each dimension in the color space for each pixel. For example, it is expressed by gradations (256, etc.) in each dimension of (R, G, B) in the RGB color space. The inspection target portion is an area including the recognition target object in the recognition target image, and is specified by a set of position coordinates of the image.

The constituent color processing unit 52 includes constituent color separating means 60, constituent color statistic calculating means 62, and inter-sample constituent color statistic calculating means 64. FIG. 3 is a flowchart of the constituent color separation means 60 in FIG. Hereinafter, the operation of the component color separation means 60 will be described with reference to FIG. In step S <b> 2, all pixels of the recognition target object at the inspection target location are extracted from the inspection image input unit 50. In step S4, the color μ (μ ₁ , μ ₂ ,..., Μ _N ) having a high appearance frequency in the inspection target location is searched from the pixels in all inspection target locations. The color μ having a high appearance frequency indicates a color having the highest appearance frequency in a color space represented by an N-dimensional color. Such a high appearance frequency color μ is estimated to be an average color of certain constituent colors. In step S6, the high appearance frequency color μ and its similar color are searched from all the pixels in the inspection target location, and the position of the corresponding pixel is stored. A similar color refers to a color that is close in distance to the high appearance frequency color μ in the color space. For example, when each dimension is expressed by 256 gradations, the position of the corresponding pixel is stored in the range of μ _i ± 20 (i = 1,..., N) in all dimensions. Thereby, it is possible to obtain an approximate position of the constituent color such that the high appearance frequency color is an average value. The reason for searching for similar colors is that, in general, the constituent colors are not completely single colors, but vary due to the imaging environment and the manufacturing of the recognition target.

In step S8, for each searched pixel, the standard deviation of the color of the pixel and neighboring pixels is calculated for each dimension, and the average value σ ( σ ₁ , σ ₂ ,..., σ _N ). The neighboring pixel refers to, for example, one peripheral pixel (all eight pixels) with the searched pixel as a central pixel. Since each color component color follows a normal distribution in each dimension, the range of variation of the component color can be known by using the standard deviation value σ (σ ₁ , σ ₂ ,..., Σ _N ). For example, in the range of μ ± 3σ, approximately 99% of the constituent colors are searched.

In step S10, a pixel corresponding to the variation range of the component colors calculated in step S8 (a range of μ _i ± 3σ _i in all dimensions) is searched. The reason for re-searching the position of the constituent color by calculating the standard deviation from the surrounding pixels of the constituent color is that the variation in color differs in each dimension of each constituent color.

  In step S12, noise removal is performed. This utilizes the property that the constituent colors of the recognition target object are fixed at the same position to some extent. First, regarding the searched pixels, adjacent pixels are regarded as connected components to form one block. For each chunk, pixel chunks whose area is less than or equal to the integrated value (for example, 5%) relative to the sum of the areas of all retrieved chunks are removed. Thereby, for example, when a color similar to the color of the substrate is included in the constituent colors, if a part of the substrate is included as the recognition target portion, the block of pixels on the substrate is removed. On the other hand, an unsearched pixel whose internal area with respect to the area of the cluster existing inside the cluster of searched pixels is equal to or less than a certain value is defined as a pixel of the constituent color.

  In step S14, the area ratio S of the area of the pixel selected in steps S10 and 12 to the search target location is calculated. Through the above analysis, the constituent colors for the frequently appearing colors calculated in step S4 can be acquired. The area ratio S of the constituent colors is stored in the constituent color information storage unit 54. In step S16, it is determined whether or not the sum of the constituent color area ratios is equal to or greater than a certain value (95%). If it is above a certain value, the analysis is terminated. If it is less than the certain value, the process proceeds to step S18. In step S18, the pixel to be inspected excluding the pixels of the acquired component color is analyzed. In steps S4 to S14, other main constituent colors and their positions can be acquired by performing the processing on the pixels that are not retrieved as the constituent colors.

The component color statistic calculation means 62 calculates the average value μ (μ ₁ , μ ₂ ,..., Μ _N ) of each component color from the pixels of each component color obtained by the component color separation means 60 and the pixels of each component color. Statistics of standard deviation σ (σ ₁ , σ ₂ ,..., Σ _N ) are calculated and stored in the component color information storage unit 54. The constituent color processing unit 60 and the constituent color statistic calculating unit 62 perform the above processing on the inspection target portions of different recognition target images (samples) for the same recognition target object input from the inspection image input unit 50. Thereby, the average value and standard deviation of each constituent color are obtained for a plurality of samples with respect to the same recognition object, and these are stored in the constituent color information storage unit 54.

The inter-sample constituent color statistic calculation means 64 calculates the constituent color variation between samples for several prepared samples for the same inspection object. For each sample, the average value, the standard deviation value, and the area ratio of each component color are calculated by the component color separation unit 60 and the component color statistic calculation unit 62 as described above, and stored in the component color information storage unit 54. Has been. For the same constituent color of each sample, the average value of the constituent color average value of the sample in each dimension (average value between samples μ _S ), the standard deviation of the sample constituent color average value (standard deviation between samples σ _S ), and the area ratio An average value (constituent color average area ratio S _M ) and an average value of standard deviations of constituent colors of the samples (constituent color standard deviation σ _M ) are calculated, and an inter-sample average value μ _S , an inter-sample standard deviation σ _S , and a constituent The color average area ratio S _M and the constituent color standard deviation σ _M are stored in the constituent color information storage unit 54.

FIG. 4 is a diagram showing each statistic. The horizontal axis represents color intensity (gradation), and the vertical axis represents frequency (number of pixels). Here, one dimension with constituent colors is shown. A bold curve indicates a normal distribution in which the average value is the average value μ _S between samples _{and the} standard deviation is the standard deviation σ _S between samples. For each sample, each sample color varies within a certain range around μ, for example, μ ± 3σ _M , and the average value of the entire sample is within a certain range around the average value μ _S between samples, for example, μ _S It shows a variation in the range of ± 3σ _S. Here, since the constituent colors of each sample are assumed to be a normal distribution, in the range of each μ ± 3σ _M , it means that about 99% of the constituent colors of the sample are within this range. To do. Further, in the range of μ _S ± 3σ _S , it means that approximately 99% of the average value of the sample is within this range.

The component color information storage unit 54 stores data (average value μ, standard deviation σ, area ratio S) of each sample unit obtained by the component color separation unit 60 and the component color statistic calculation unit 62 for each inspection object. ), And the inter-sample constituent color statistic calculation unit 64 stores the inter-sample average value μ _S , the inter-sample standard deviation σ _S , the constituent color average area ratio S _M , and the constituent color standard deviation σ _M.

  The inspection data determination unit 56 includes an extraction color range calculation unit 70, an extraction area calculation unit 72, and an inspection target color selection unit 74. The extracted color range calculation means 70 uses the statistics obtained by the inter-sample constituent color statistics calculation means 64 to specify the extraction color range.

FIG. 5 is a diagram showing the extracted color range. The horizontal axis indicates color intensity, and the vertical axis indicates frequency. Although the extracted color range is specified for each dimension of the constituent colors, FIG. 5 shows one dimension. As described above, each sample of each constituent color varies within a range of μ ± 3σ _M , for example. In the entire sample, the average value of the sample varies in the range of μ _S ± 3σ _S. Therefore, the overall variation is in the range of μ _S ± (3σ _S ± 3σ _M ). Accordingly, the extracted color range has a lower limit value of μ _S − (3σ _S + 3σ _M ) and an upper limit value of μ _S + (3σ _S + 3σ _M ). If the extraction range is smaller than this range, the extraction area may decrease and “overlook” may occur. If the extraction range is larger than this range, the extraction area will increase and other inspection objects may be extracted. It will cause “missing”. The extraction area calculation unit 72 calculates an extraction area ratio of each constituent color of another inspection object with respect to an extraction color range of each constituent color set for an arbitrary inspection object.

FIG. 6 is a flowchart of the extraction area calculation means 72. In step S100, the center of the extracted color range of one constituent color α arbitrarily selected from the inspection object is calculated. The extracted color range calculates the position. The inter-sample average value μ _Sα , the inter-sample standard deviation σ _Sα , and the constituent color standard deviation σ _Mα of the constituent color α stored in the constituent color storage unit 54, and the extracted color range is [μ _Sα − ( 3σ _Sα + 3σ _Mα ), μ _Sα + (3σ _Sα + 3σ _Mα )], the center value of the extracted color range is the average value μ _Sα between samples. The extracted color range can be set to an arbitrary range. In step S102, the inter-sample average value μ _Sβ , the sample stored in the constituent color storage unit 54 of one constituent color β of another inspection target arbitrarily selected different from the inspection target selected in step S100 during standard deviation sigma _{S [beta,} reads the configuration color standard deviation sigma _Emubeta, for each dimension of the color space, the center of the range of values of the extracted color range of composition colors α is the corresponding dimension of the sample between the variation mu _{S [beta} ± 3 [sigma] _{S [beta} configuration color β It is judged whether it is in. If it is within the range of variation between samples, the process proceeds to step S104. If it is outside the range of the variation between samples, the process proceeds to step S106.

  FIG. 7 is a diagram illustrating a method of calculating the upper limit value and the lower limit value of the extraction area when the extracted color center value is within the variation between samples. In step S104, as shown in FIG. 7, the sample normal distribution and the extracted color range obtained from the average value and standard deviation of the sample such that the average value A of the constituent colors β coincides with the center of the extracted color range are enclosed. Therefore, the extraction area ratio (for example, 80%) of this area to the entire area of the normal distribution of this sample is calculated, and this is set as the maximum (upper limit).

  FIG. 8 is a diagram illustrating a method of calculating the upper limit value and the lower limit value of the extraction area when the extracted color center value is outside the variation between samples. In step S106, a sample whose average value of the constituent color β is closest to the center of the extracted color range, for example, the average value and standard of the sample whose average value for the corresponding dimension is A as shown in FIG. Since the area of the hatched portion surrounded by the normal distribution of the sample and the extracted color range obtained from the deviation becomes the maximum, the ratio of the extracted area to the area of the entire normal distribution of this sample (for example, 45%) Is calculated and set as the maximum (upper limit).

  In step S108, as shown in FIGS. 7 and 8, the normal distribution for the corresponding dimension of the sample farthest from the center value of the extracted color range regardless of whether the extracted color center value is within or outside the inter-sample variation. Since the area of the hatched portion surrounded by B and the extracted color range is minimized, the ratio of the extracted area to the area of the entire normal distribution of this sample (for example, 5% as shown in FIG. (0% as shown in FIG. 8) is calculated and set as the minimum (lower limit).

  As described above, the upper limit and lower limit values of the extraction area ratio are calculated for each dimension of the constituent colors of any other inspection object with respect to each dimension of the extraction color range set for the arbitrary inspection object. Is done.

  The inspection target color selection unit 74 is a unit that selects an inspection target color from among the constituent colors of the inspection target. Basically, a constituent color having a large area ratio among the constituent colors of the inspection object or a constituent color unique to the inspection object that does not exist in other inspection objects may be selected as the inspection object color. However, if there are inspection objects with similar constituent colors, or if the inspection objects have constituent colors with various similar area ratios, it is necessary to select a constituent color according to the situation. is there. The optimum inspection target color requires that only the inspection object is extracted and no other inspection object is extracted.

FIG. 9 is a flowchart of the inspection object color selection unit 74. In step S150, an inspection object is selected. In step S152, an extraction color range is input. If no extracted color range is input, by default, μ _S ± (3σ _S + 3σ _M ) calculated by the extracted color range calculating means 70 is extracted color range for each constituent color of the selected inspection object. And In step S154, for each dimension of the extracted color range determined in step S152, the upper limit and lower limit values of the extraction area ratio by the above-described extraction area calculation means 72 for the corresponding dimension of each constituent color of the other inspection object. Is calculated. Note that the upper and lower limits of the extraction area ratio of each constituent color of the other inspection object in the extraction color range of the inspection object are products of the upper and lower limits of each dimension.

  FIG. 10 is a diagram illustrating an example of the display of the extracted color range of the constituent colors of the inspection object and the extraction area ratio of each constituent color of the other inspection objects. Here, the inspection object is the part 5, and the extraction area ratio is shown for the parts 1 to 4 and the part 5 of other inspection objects. In step S156, for each constituent color of the inspection object, the extracted color area of each constituent color of the other inspection object and the upper and lower limits of the constituent color are displayed on the display. For example, as shown in FIG. 10, when the component color of the component 5 is the component colors 1 to 3 and the extraction color range is the extraction color range 1 to 3, the area for each component color of the components 1 to 5 The upper and lower limit values in the extracted color ranges 1 to 3 of the component colors 1 to 3 of the component 5 to be inspected are displayed.

  In step S158, the setting data of the extracted color range of the constituent color of the inspection object of the inspection object is input. In the setting data, the person who performs the inspection looks at the contents shown in FIG. 10 displayed on the display and selects the optimum one so that there is no “missing” or “too much” of the inspection object. In step S160, the setting data of the threshold value of the constituent color to be inspected of the inspection object is input. It should be noted that the threshold of the extraction area of the extracted color that determines the pass / fail judgment may be set to be equal to or greater than the maximum extraction area ratio of other inspection objects in the extraction range of the extracted color. However, in some cases, there may be an inspection object having the same component color. In this case, for a constituent color having a small maximum extraction area ratio of another inspection object having the same constituent color as the inspection object, a value equal to or greater than the maximum extraction area ratio is set as a threshold.

Next, a data setting method will be described. Here, a case where a component mounted on a substrate is inspected as an inspection object will be described as an example. A substrate on which the inspection object is mounted is produced. A place including each component mounted on the board is imaged by a camera to generate an inspection target image. The inspection target part of the inspection target in the inspection target image is designated. The component color separation unit 60 separates the test object into the component colors, and the component color statistic calculation unit 62 calculates an average value and a standard deviation for each dimension unit in the color space, and stores them in the component color storage unit 54. . When the average value and the standard deviation of a plurality of samples are calculated for the same inspection object, the inter-sample average value and the standard deviation are calculated by the inter-sample constituent color statistic calculation means 64 from the average value and the standard deviation of the plurality of samples of the inspection object. μ _S , intersample standard deviation σ _S , constituent color average area ratio S _M, and constituent color standard deviation σ _M are stored in the constituent color storage unit 54.

The extracted color range calculation means 70 calculates the extracted color range μ _S ± (3σ _S + 3σ _M ) of the constituent colors of the inspection object. The extraction area calculation unit 72 calculates the upper limit and lower limit values of the extraction area ratio of each constituent color of the other inspection object in the extracted color range of the constituent color of the inspection object. The inspection target color selection means 74 displays the upper and lower limit values of the extraction area ratio of each component color of the other inspection object in the extraction color range of each component color of the inspection object on the display. The inspector sets the extraction color range and threshold value of the inspection object. By performing the above processing every time a substrate on which an inspection object is mounted is manufactured, the number of samples is increased, and statistical accuracy is improved. In addition, the above processing can be performed in parallel with the inspection of the substrate. That is, it is possible to perform inspection and data setting in parallel by using the inspection target image of the substrate in combination for data setting and inspection.

  The present invention includes the following supplementary notes.

(Additional remark 1) In the data setting apparatus which sets the said setting data of the image recognition apparatus which recognizes the image for recognition of the recognition target object based on setting data,
A recognition image input unit for designating a recognition area in a recognition image in which a recognition object is captured;
A constituent color processing unit that extracts a position of a constituent color of the recognition object based on an image in the recognition area and calculates a statistic indicating a color variation related to the constituent color;
An inspection data determination unit that determines an extraction color range of the recognition object based on the statistics;
A data setting device comprising:

  (Additional remark 2) The said structural color process part has the said appearance frequency high from the average value of the pixel of the color with a high appearance frequency of the said image for recognition objects in the said recognition object, and the pixel of the similar color of the said high appearance frequency color. A component color analyzing unit for extracting a component color as a pixel having a certain variation range based on a standard deviation of neighboring pixels of each pixel of the color and the similar color pixel is included in the component color of the recognition object; The data setting device according to supplementary note 1, characterized by:

  (Supplementary Note 3) The component color processing unit calculates a statistic of a first average value and a first standard deviation value of each dimension value in a color space of each pixel included in the component color. The data setting device according to appendix 2, characterized by comprising:

  (Additional remark 4) The said structural color process part is the average of the said 1st average value about the some sample recognition object image in which the same recognition target object was imaged, and the average value between the said sample average values. An inter-sample constituent color statistic calculation unit that analyzes a variation range between samples based on a standard deviation value between samples that is a standard deviation and a constituent color average standard deviation value that is an average value of the first standard deviation values is provided. The data setting device according to supplementary note 3, which is characterized.

  (Additional remark 5) The said test | inspection data determination part is the extraction color of the said component color based on the said average value between samples, the said sample standard deviation value, and the said component color average standard deviation value about each component color of the said recognition target object. The data setting device according to appendix 4, further comprising an extraction color range calculation unit for calculating the range.

  (Additional remark 6) The said test | inspection data determination part sets the upper limit and lower limit of the probability by which the constituent color of the other recognition target object different from the said recognition target object is extracted within the extraction color range of the constituent color of each said recognition target object. The data setting device according to appendix 5, further comprising an extraction area calculation unit that calculates an average value between samples of the constituent colors of the other recognition target object, a standard deviation value between samples, and an average standard deviation value of the constituent colors .

  (Additional remark 7) The said extraction area calculation part assumes that distribution of the pixel value contained in the component color of each image for recognition object of the said other recognition target object is normal distribution, and is in the said extraction color range in the said distribution 7. The data setting device according to appendix 6, wherein the upper limit and the lower limit are calculated by calculating an area.

  (Additional remark 8) The said test | inspection data determination part displays the said upper limit and lower limit of each structural color of another recognition target object on a screen, when the said extraction color range of the said structural color of the said recognition target object is given. The data setting device according to appendix 7, further comprising an inspection object color selection unit.

  (Additional remark 9) The said structural color analysis part calculates the area ratio with respect to the area of the said target area | region of the area of the area | region which consists of a pixel contained in each said constituent color, The said test target color selection part is the said recognition target object. 9. The data setting device according to appendix 8, wherein an area ratio of the constituent colors and an area ratio of another recognition object different from the recognition object are displayed on the screen.

(Additional remark 10) In the data setting method which sets the said setting data of the image recognition apparatus which recognizes the image for recognition of the recognition target object based on setting data,
Designating a recognition area in a recognition image in which a recognition object is captured;
Extracting a constituent color of the recognition object based on an image in the recognition area, and calculating a statistic indicating a color variation related to the constituent color;
Determining an extraction color range of the recognition object based on the statistics;
A data setting method characterized by comprising:

It is a principle diagram of the present invention. It is a block diagram of the data setting apparatus by embodiment of this invention. 3 is a processing flowchart of a component color separation unit in FIG. 2. It is a figure which shows each statistic. It is a figure which shows the extraction color range. It is a flowchart of an extraction area calculation means. It is a figure which shows the case where an extraction color range exists in the dispersion | variation between samples. It is a figure which shows the case where an extraction color range exists outside the dispersion | variation between samples. It is a flowchart of a test object selection means. It is a figure which shows an example of a display.

Explanation of symbols

2 Inspection Image Input Unit 4 Component Color Processing Unit 6 Inspection Data Determination Unit 10 Recognition Object 12 Recognition Object Image 14 Recognition Area

Claims (8)

In the data setting device for setting the setting data of the image recognition device that recognizes an image for recognition of a recognition object based on setting data,
A recognition image input unit for designating a recognition area in a recognition image in which a recognition object is captured;
A constituent color processing unit that extracts a position of a constituent color of the recognition object based on an image in the recognition area and calculates a statistic indicating a color variation related to the constituent color;
An inspection data determination unit that determines an extraction color range of the recognition object based on the statistics,
The component color processing unit includes a component color statistic calculating unit that calculates a statistic indicating a variation in the color based on a value of each dimension in a color space of each pixel included in the component color. Data setting device. In the data setting device for setting the setting data of the image recognition device that recognizes an image for recognition of a recognition object based on setting data,
A recognition image input unit for designating a recognition area in a recognition image in which a recognition object is captured;
A constituent color processing unit that extracts a position of a constituent color of the recognition object based on an image in the recognition area and calculates a statistic indicating a color variation related to the constituent color;
An inspection data determination unit that determines an extraction color range of the recognition object based on the statistics,
The component color processing unit calculates a statistic indicating the color variation with respect to individual representative values of a plurality of sample recognition target images in which the same recognition target object is imaged. Data setting device. In the data setting device for setting the setting data of the image recognition device that recognizes an image for recognition of a recognition object based on setting data,
A recognition image input unit for designating a recognition area in a recognition image in which a recognition object is captured;
A constituent color processing unit that extracts a position of a constituent color of the recognition object based on an image in the recognition area and calculates a statistic indicating a color variation related to the constituent color;
An inspection data determination unit that determines an extraction color range of the recognition object based on the statistics,
The constituent color processing unit includes a constituent color statistic calculating unit that calculates a statistic of a first average value and a first standard deviation value of each dimension value in a color space of each pixel included in the constituent color,
The component color processing unit is an average value between samples that is an average of the first average values and a standard deviation of the average value between samples for a plurality of sample recognition target images in which the same recognition target object is imaged. Data comprising an inter-sample constituent color statistic calculation unit for analyzing a variation range between samples based on an inter-sample standard deviation value and a constituent color average standard deviation value that is an average value of the first standard deviation values Setting device.   The inspection data determination unit calculates an extraction color range of the constituent color for each constituent color of the recognition target object based on the average value between samples, the standard deviation value between samples, and the average standard deviation value between constituent colors. 4. The data setting device according to claim 3, further comprising an extracted color range calculation unit.   The inspection data determination unit determines the upper and lower limits of the probability that the constituent colors of other recognition objects different from the recognition object are extracted within the extracted color range of the constituent colors of the recognition objects. 5. The data setting device according to claim 4, further comprising an extraction area calculation unit for calculating based on an inter-sample average value, an inter-sample standard deviation value, and a constituent color average standard deviation value of the constituent colors of the object.   The extraction area calculation unit calculates an area within the extraction color range in the distribution assuming that the distribution of pixel values included in the constituent colors of each recognition target image of the other recognition target object is a normal distribution. The data setting device according to claim 5, wherein the upper limit and the lower limit are calculated.   The inspection data determination unit displays the upper and lower limit values of each constituent color of another recognition target object on the screen when the extracted color range of the constituent color of the recognition target object is given The data setting device according to claim 6, further comprising a selection unit.   The constituent color analyzing unit calculates an area ratio of an area of a pixel composed of pixels included in each constituent color to an area of the target region, and the inspection target color selecting unit calculates the constituent color of the recognition target object. The data setting device according to claim 7, wherein an area ratio and an area ratio of another recognition object different from the recognition object are displayed on the screen.

Images