JP2005309914A - Method for processing color image and image processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To stabilize an extraction result and a measurement result by eliminating the need of making a user specify a reference color and to perform easy and highly accurate processing even in the case of processing a part having a plurality of kinds of colors. <P>SOLUTION: Pieces of image data about each color component generated by a color camera 2 are stored in image memories 12, 13 and 14 of a main body part 1, respectively. A CPU 10 makes the surface of an object in a color image constituted of the three kinds of image data into a measurement object area and divides the area into a plurality of small areas. Then, color difference of average colors is calculated for every set of adjacent small areas after calculating the average colors for every small areas and frequencies in which the color difference exceeds a threshold are counted. In addition, when pieces of image data constituting a reference color image are registered in reference image memories 15, 16 and 17, a color abnormality part is extracted by calculating the color difference to corresponding pixels of the reference image. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention uses a color image composed of density data of red, green, and blue color components to extract an object having a predetermined color or pattern, or to measure the size, posture, etc. of the object. The present invention relates to a method and apparatus for performing processing such as detecting an abnormality occurring in a color as a defect.

  In typical color image processing performed conventionally, by allowing a user to specify a predetermined color, three-dimensional image data obtained by combining density data of each color component is converted into a one-dimensional color difference with respect to the specified color. And processing using the size of the color difference data is performed (see Patent Document 1).

  In addition, when extracting a portion where a color abnormality has occurred as a defect, it is also proposed to register a normal color as a reference color in advance and obtain a color difference from the reference color for each pixel constituting the measurement target image. (See Patent Document 2).

JP-A-7-203476 JP-A-11-108759

  FIG. 9 shows an outline of processing for replacing each pixel on a color image with one-dimensional color difference data in a relationship in a virtual space (hereinafter referred to as “color space”) with three axes corresponding to each color component. It is. The color designated by the user corresponds to a specific point Q in this color space. Here, when the color difference is expressed by the distance from the point Q in this color space, the colors corresponding to the points on the surface of the sphere centered on the point Q are all expressed by the same value. In FIG. 9, the value converted to each density value of 150, 100, and 50 with the value of the point Q being 255 is schematically shown in the outline of the cross section of the sphere.

  When the user designates a color, the color to be designated may vary depending on the user or the same user. When the designated color changes, the position of the point Q also changes, so that the color difference value in each pixel also differs. That is, as in Patent Documents 1 and 2, when the color at a predetermined position on the image is set as a reference color, the density distribution on the image converted as a color difference varies depending on the selection of the reference color. Furthermore, there is a possibility that even the measurement result of the object and the detection result of the defect may be changed due to such fluctuation of the density distribution.

  In addition, when different colors are converted to the same value, for example, when performing correlation calculation processing between the converted image and the model image (an image in which the color image is replaced with a color difference), the actual color differs. However, there is a problem that if the distribution state of the density indicating the color difference is similar, the degree of similarity increases.

  In addition, since the conventional method uses a specific color as a reference color, the color of the measurement target, such as a character string in which each constituent character is represented in a different color, or an object in which a pattern with multiple colors is represented. It becomes difficult to deal with when there are multiple types.

  The present invention has been made paying attention to each of the above problems, and when extracting or measuring a portion having a color different from the surroundings, the extraction result or measurement result can be obtained by eliminating the need for the user to specify a reference color. The purpose is to stabilize.

  A second object of the present invention is to facilitate processing when extracting or measuring a portion having a plurality of kinds of colors and to obtain a highly accurate processing result.

  The present invention is applied to a method of processing a color image composed of density data of red, green, and blue color components. In the first method according to the present invention, a process of setting two areas having the same size on a color image to be measured, a process of obtaining a color difference between areas based on an average color in each area, and the color difference A process for comparing with a predetermined threshold value is executed for a plurality of cycles while changing the setting position of at least one of the areas, thereby determining whether or not there is a portion in which the color has changed on the image. ing.

  In the above method, the color of each constituent pixel of the color image to be measured is represented by a combination of red, green, and blue density data. The average color can be obtained by averaging the density value of each pixel in the area for each color component. That is, it can be considered that the color indicated by the average density value of the red component, the average density value of the green component, and the average density value of the blue component is an average color.

  The above method can be executed when extracting a part having a color different from the base color part, such as extracting a defect having an abnormal color from an image of an object with a small color change. For example, if an area having a size corresponding to the size of the defect to be detected is set when detecting a defect, the average color when there is a defect in one area and no defect in the other area is Can be considered different between regions.

The color difference between the two regions can be obtained, for example, as the distance between the average colors in the color space described above. Alternatively, each average color may be converted into one point in the virtual space by brightness, saturation, and hue, and the distance between these points may be obtained. Alternatively, the color difference may be calculated by replacing each average color with chromaticity or luminance specified by the International Lighting Commission.
In any case, when the color difference between the two areas exceeds the threshold value, it can be determined that the colors in these areas are different.

  Furthermore, in the first method, while changing the setting position of at least one of the two areas, a process for obtaining a color difference between the areas and a process for comparing the obtained color difference with a threshold value are repeatedly executed. . Here, the setting of the area includes a process of dividing a measurement target area on the image into a plurality of areas in advance and selecting any two of the areas. In this case, the process of changing the combination of areas to be selected can be regarded as the process of changing the set position of the area.

  In the first method, the color difference can be obtained by fixing the setting position of one area and sequentially setting the other area so as to be adjacent to the fixed area in the upper, lower, left, and right directions. In this case, if any color difference exceeds the threshold value, it can be determined that a color different from the surroundings appears in the fixed region. When only one of the color differences exceeds the threshold value, it can be considered that different colors are included in the region where the setting position is changed. Therefore, by changing the setting position of the area to be fixed in various ways, for each setting position, a process for obtaining a color difference and a process for comparing the obtained color difference with a threshold value are executed for a plurality of cycles, so that different colors can be obtained. It is possible to clearly discriminate the area that has it.

  However, in the first method, it is not always necessary to specify a portion having a different color, and it is only necessary to determine the presence / absence of a portion having a different color based on the number of times that the color difference exceeds the threshold value.

  In the first method, the color difference can be calculated and compared while moving two adjacent regions along a predetermined direction. In this case, the boundary between the areas when the color difference exceeds the threshold value can be extracted as the position of the color boundary. According to this processing, it is possible to detect the position of the defect or to distinguish the boundary between the object and the background.

  In the second method according to the present invention, while scanning a color image to be processed along a predetermined direction, processing for obtaining an average color along a direction orthogonal to the scanning direction is repeatedly executed at predetermined intervals. Next, in the color space described above, a point corresponding to a color boundary is extracted from these points based on the positional relationship of the points corresponding to the plurality of average colors obtained when scanning the color image, and the points The scanning position when the average color corresponding to is obtained is determined as the position where the color has changed.

  This method can be applied to the use of extracting the boundary between two kinds of colors on an image, such as extracting the position of the contour of the object when the color change of both the background and the object is small. For example, a line having a predetermined length can be set along a direction orthogonal to the scanning direction, and an average color of each pixel on the line can be obtained. In this case, the length of the line can be arbitrarily set according to the size of the object. Further, instead of the line, a rectangular area having a predetermined width in the scanning direction may be set, and an average color in the area may be obtained. In either case, the average color can be calculated for each pixel. However, the present invention is not limited to this, and it may be calculated every several pixels. When the rectangular area is set, the average color may be obtained every time scanning is performed by the width of the area.

  In the above scanning, it can be considered that the average color extracted while the color does not change greatly is distributed in the same kind of color range in the color space. On the other hand, when the color changes greatly at a predetermined position, it can be considered that the average color at that position is located at a place away from the distribution of each average color obtained so far. Therefore, in the color space, the points corresponding to each average color are tracked in the order of scanning, and when a point located at a distance exceeding a predetermined threshold is found, that point corresponds to the boundary of the color Can be extracted as

Further, when scanning a rectangular area, it can be considered that the ratio of two colors included in the rectangular area gradually changes in accordance with the movement of the area near the boundary between the colors. Therefore, when two points (assumed to be A and B) that are located farthest from the points corresponding to each average color in the color space are extracted, one point A has a region in front of the boundary. It can be considered that it corresponds to the average color when only the color is included, and the other point B corresponds to the average color when the region includes only the color behind the boundary. The points corresponding to the other average colors are likely to be arranged in the order of extraction from point A to point B. Therefore, the points A and B are connected by a line segment, the position of the intersection point by the perpendicular line dropped from the other point to the line segment AB is sequentially tracked, and the intersection point of the perpendicular line and the line segment AB is determined to be a predetermined reference position (for example, a line segment). When the value exceeds the midpoint of AB, the point of interest at that time can be extracted as a point corresponding to the color boundary.
Even when a line is scanned, the same method as described above can be applied if the color on the image changes gradually.

  According to a third method of the present invention, there is provided a process for obtaining a color difference for each set of corresponding pixels between a color image to be measured and a reference color image registered in advance, and calculating the color difference to a predetermined threshold value. By executing the process of comparing with the reference image, a portion where a color different from the reference color image appears in the measurement target image is extracted.

  This method can be used when measuring a plurality of objects having different colors or an object having a plurality of types of colors such as patterns. When this method is executed, it is desirable to correct in advance a positional deviation or a rotational deviation between the color image to be measured and the reference color image.

  In the above method, since a color difference is obtained for each corresponding pixel with respect to a reference color image having a good color distribution state, a color distribution different from that of the reference color image is generated in the color image to be measured. However, the difference can be extracted with high accuracy. In this case, since the user only performs the process of registering the reference color image, it is possible to obtain a stable processing result by registering an image having an appropriate color distribution for the region to be measured. it can.

  Next, in the fourth method according to the present invention, a measurement target area having the same size as the reference color image registered in advance is provided on the color image to be measured, and the image in the measurement target area and the reference color image are provided. For each of the above, a process for obtaining an average color and a process for obtaining a distance and a direction for the point corresponding to the average color are performed for each point corresponding to the color of each constituent pixel in the color space. Further, the similarity between the image in the measurement target area and the reference color image is obtained using the distance and direction obtained for each of the constituent pixels, and when the similarity exceeds a predetermined threshold, the measurement target area Is determined to correspond to the reference color image.

  Similarly to the third method, the fourth method can also be used when a plurality of objects having different colors or a single object expressing a plurality of types of colors are to be measured. . The reference color image in this method is the same size as the measurement target color image or smaller than the measurement target color image, and can be considered as an image including the target object to be extracted.

  In the fourth method, for each of the image in the measurement target region and the reference color image, points corresponding to the average color of the image and the color of each constituent pixel are set in the color space described above. Then, for each point corresponding to the constituent pixel, the distance and direction from the point corresponding to the average color are obtained. This distance and direction can be expressed as a vector from a point corresponding to the average color to a point corresponding to the color of the constituent pixel. In addition, it is desirable that the distance and direction of the reference color image are obtained before the measurement process and registered in the memory.

  The similarity based on the above distance and direction is set to increase as the relationship between the distance and direction between each pixel in the measurement target region becomes closer to the relationship between the distance and direction between each pixel of the reference color image. Is desirable. For example, the similarity is obtained by a correlation calculation using the distance DPi obtained for each pixel of the measurement target region, the distance DMi obtained for each pixel of the reference image, and the direction shift amount θi between corresponding pixels of each image. Can be calculated. Further, for each corresponding set of pixels, by calculating the sum of the difference between the distances DPi and DMi and the direction shift amount θi, and subtracting the sum of the calculation results for each pixel from a predetermined reference value (for example, 100), Similarity can be obtained. In any case, the similarity value can be increased as the color distribution of the image in the measurement target region approaches the color distribution of the reference color image.

  According to the above method, when extracting a region where a plurality of types of colors are distributed from a color image to be measured, a color image in which a correct distribution state appears is registered as a reference color image, and By executing the above method while sequentially changing the setting position, it is possible to extract the measurement target region when the highest similarity (greater than the predetermined threshold value) is obtained as the corresponding region.

  Also, if the position of the object is extracted in advance by a method such as edge extraction, the measurement target region is set based on the extraction result, and the above method is executed, the suitability of the color distribution of the object is accurately determined. be able to.

  An image processing apparatus for carrying out the first method includes an image input unit for inputting a color image composed of density data of red, green, and blue color components, and a predetermined image on the color image input by the image input unit. After setting the measurement target area of the size, the area setting means for dividing the measurement target area and setting a plurality of small areas, and after obtaining the average color for each small area, for each set of adjacent small areas And calculating means for obtaining a color difference based on the average color, and comparing the color difference obtained by the computing means with a predetermined threshold value, and when the color difference exceeds the threshold value, Discriminating means for discriminating that a color difference occurs between the small areas.

  An image processing apparatus for performing the second method includes an image input unit that inputs a color image composed of density data of each color component of red, green, and blue, and a color image that is input from the image input unit. Average color calculation means for obtaining an average color along a direction orthogonal to the scanning direction while scanning along a predetermined direction, and points corresponding to a plurality of average colors obtained by the average color calculation means in the color space Based on the positional relationship, the color changed the extraction position for extracting the point corresponding to the color boundary from these points, and the scanning position when the average color corresponding to the point extracted by the extraction means was obtained Discriminating means for discriminating the position.

  An image processing apparatus for carrying out the third method includes an image input means for inputting a color image composed of density data of red, green, and blue color components, a memory for registering a reference color image, A calculation means for obtaining a color difference for each set of corresponding pixels between a color image input from the image input means and a reference color image in the memory, and a color difference obtained by the calculation means are predetermined. And a discriminating means for discriminating that a color different from the reference color image appears in the pixel from which the color difference is extracted when the color difference exceeds a threshold value.

  An image processing apparatus for carrying out the fourth method includes an image input means for inputting a color image composed of density data of red, green, and blue color components, a memory for registering a reference color image, A region setting unit that sets a measurement target region having the same size as the reference color image on the color image input from the image input unit, and each of the image in the measurement target region and the reference color image After the processing for calculating the average color in the image and the processing for obtaining the distance and direction for the point corresponding to the average color for each point corresponding to the color of each constituent pixel of the color space, the obtained configuration The calculation means for obtaining the similarity between the image in the measurement target area and the reference color image using the distance and direction for each pixel, and the similarity is compared with a predetermined threshold value to determine the similarity. When it exceeded had value, the measurement target region; and a discrimination means for discriminating that corresponds to the reference color image.

In the apparatus having each configuration described above, the image input means can be an interface circuit for the imaging means for generating a color image. When the imaging means is an analog camera, the image input means can include an A / D conversion circuit that digitally converts density data of each color component.
The image input means can also be configured as a communication interface circuit for inputting a color image generated in advance from an external device or a drive device for reading out a color image stored in a predetermined storage medium.

  Each means other than the image input means and the memory is preferably constituted by a computer in which a program is set. However, it is not always necessary to configure all the means by a computer, and some of the means can be configured by an ASIC (application specific IC).

  The image processing apparatus described above can be used not only for the purpose of extracting the position of an object, but also for inspections such as extracting defects that appear as color abnormalities and determining the suitability of pattern patterns. Any device can be provided with a means for displaying an extraction processing result, a good / bad determination result, and the like, and outputting the result to the outside.

  The third and fourth image processing apparatuses are preferably provided with means for receiving input or designation of a reference color image and registering the image in a memory. Further, it is desirable that the fourth image processing apparatus is provided with means for registering the distance and direction obtained by the calculation means for the reference color image in the memory.

  According to the present invention, when extracting or measuring a portion having a color different from the surroundings, it is not necessary to specify a reference color, so that a stable extraction result or measurement result can be obtained. Moreover, even if the base color of the object changes, it can be easily handled.

  Furthermore, according to the present invention, when extracting or measuring a portion including a plurality of kinds of colors, it is possible to easily cope with the problem by registering a reference color image having a good color distribution. In addition, it is possible to execute a process of extracting an abnormal portion of a color and a process of extracting an object with high accuracy without being affected by the number and types of colors included in the image.

  FIG. 1 shows a basic configuration of an image processing apparatus to which the present invention is applied. This image processing apparatus includes a camera 2 for generating a color image and a main body 1. The color image generated by the camera 2 is taken into the main body 1 to determine the presence / absence of a color abnormality part, Processes such as extracting the object contained in it are executed. The main body unit 1 includes a CPU 10, a memory 11 in which programs are stored, an output unit 18 for outputting processing results, and the like, and can be configured by a personal computer, for example.

  Furthermore, the main body 1 of this embodiment is provided with image memories 12, 13, and 14 for each color component of red (R), green (G), and blue (B). The image data of each color component output from the camera 2 is stored in the corresponding image memories 12, 13, and 14, respectively. Although not shown in this figure, an interface circuit, an A / D conversion circuit, and the like are provided in the input path to each of the image memories 12, 13, and 14. The output unit 18 includes a communication interface circuit for transmitting the processing result to an external device, a display interface circuit for displaying the processing result on a monitor (not shown), and the like.

In addition, in the main body 1, in order to register a reference image extracted from a color image of a model of an object, image memories 15, 16, 17 (hereinafter referred to as “reference image memories 15, 16, 17 ”).
The memory 11 includes an auxiliary storage device such as a hard disk in addition to the ROM and RAM. The reference image memories 15, 16, and 17 and the image memories 12, 13, and 14 can be physically set in the same storage device (the hard disk or the like).

  Hereinafter, four examples of processing that can be performed by the above-described image processing apparatus will be described, and detailed processing contents of each example will be described. The “image” referred to below is a combination of R, G, and B image data.

(1) Defect inspection 1
This inspection is to inspect whether or not a color different from the original color has occurred on the surface of a plain object due to defects such as dirt and spots. FIG. 2 shows the procedure of this inspection.

  This procedure starts when the image data from the camera 2 is stored in each of the image memories 12, 13, and 14. First, in the first step 1 (hereinafter, step is abbreviated as “ST”), a measurement target region is set for a target on an image. Although not described in detail in FIG. 2, when setting the measurement region, for example, a monochrome image is created from the input image data of each color component, and an edge of the monochrome image is extracted. Then, an image of the object is extracted. Then, based on the extraction result, a measurement target region is set in the original color image so as to include the inspection target portion of the target object and not to include a portion having another color such as a background portion. However, when the object does not greatly deviate, the setting position of the measurement target area may be registered in advance.

  In the next ST2, the measurement target area is divided into a plurality of small areas. The size of the small area is set in advance according to the size of the defect to be extracted. FIG. 3 shows a specific example, and a plurality of (12 in the illustrated example) small regions are set by equally dividing the measurement target region 20 for each of the x and y axes.

  In the next ST3, for each small area, the density data in each pixel in the small area is averaged for each color component of R, G, and B. The color represented by the average value of R, the average value of G, and the average value of B obtained by this averaging is the average color.

  Next, in ST4, the counter n is set to 1 and the maximum color difference value Dmax is set to 0, respectively. Then, paying attention to the n-th small area, the color difference Dn is calculated for the small area adjacent to the small area in any of the vertical and horizontal directions (ST5, ST6). This color difference Dn can be obtained as the Euclidean distance in the color space. That is, when the average color in the small area under consideration is (R0, G0, B0) and the average color in the adjacent small areas is (R1, G1, B1), the color difference Dn is expressed by the following equation (1). Can be calculated.

  In ST7, the color difference Dn obtained by the above equation (1) is compared with the maximum value Dmax. Since the initial value of Dmax is set to 0 in ST4, the first determination in ST7 is always “YES”. In this case, the process proceeds to ST8, and the maximum value Dmax is rewritten with the value of Dn.

  Thereafter, after performing ST6 to 8 on all the small regions adjacent to the small region of interest, the value of n is further updated to change the target of interest, and the above processing is performed (ST5 to 11). ). As a result, the color difference Dn is obtained for each set of adjacent small regions, and the maximum value among them is extracted as Dmax. In ST12, the maximum value Dmax is compared with a predetermined threshold value Dth. Note that the threshold value Dth can be set based on a color difference (desirably the maximum color difference) obtained from an image of an object having no defect prior to inspection.

  If it is determined in ST12 that Dmax is larger than the threshold value Dth, the process proceeds to ST13 and a defect determination (meaning that the object is defective) is performed. On the other hand, if Dmax is equal to or less than the threshold value Dth, the process proceeds to ST14 and a good determination (meaning that the object is free of defects) is performed. In ST15, the determination result of ST13 or ST14 is output from the output unit 18, and the process ends.

  According to the above processing, a defect inspection can be performed on a plain object without having to register the color of the object. Moreover, even when the color of the object changes during the inspection, the inspection can be promptly advanced by the same algorithm.

  In the procedure of FIG. 2, the maximum value Dmax of the color difference in the measurement target region is obtained and the presence or absence of a defect is determined based on the magnitude of the value of Dmax. Instead, each time the color difference Di is obtained, The value may be compared with the threshold value Dth, and the defect determination may be performed when the color difference Di exceeds the threshold value Dth.

  In addition, when a color difference exceeding the threshold value Dth is extracted, the position of the defect can be specified by checking the color difference with the other small areas for the set of small areas when the color difference is obtained. . For example, in the procedure of FIG. 2, when the maximum value Dmax finally obtained exceeds the threshold value Dth, the color difference with respect to the other small areas is checked for two small areas when Dmax is obtained, For all of the small areas, the small area from which the color difference exceeding the threshold value Dth or the color difference close to Dth is extracted is specified as the defective portion.

(2) Defect inspection 2
This inspection is to check whether or not a color that should not be originally included is included in an object in which characters, designs, patterns, and the like are represented by a plurality of types of colors, not plain. When performing this inspection, first, a model of an object having a good color distribution is imaged, an image in a predetermined region including the object is extracted from the obtained color image, and this is extracted from the reference image memory 15. , 16 and 17. Then, a color difference is obtained for each corresponding pixel between the image to be inspected and the reference image, and a pixel whose color difference exceeds a predetermined threshold value dn1 is extracted as a defective pixel.

  FIG. 4 shows the detailed procedure of the inspection. In ST101, which is the first step in this procedure, the edge of the object is extracted by converting the color image to be processed into a monochrome image, and the reference image and the color image to be processed are added to the color image to be processed based on the extraction result. Set the measurement target area of the same size. In the following description, it is assumed that the model and the target object are accurately aligned when the reference image is superimposed on the set measurement target region. In consideration of a subtle difference in size, it is desirable to include a slight error in the threshold value dn1.

When the measurement target region is set, in ST102, the counter i is initialized to 1, and the number of defective pixels dn is initialized to 0. In the next ST103, the color difference Di with respect to the corresponding pixel on the reference image side is calculated for the i-th pixel. Here, when the color difference Di exceeds the predetermined threshold value D1, ST104 becomes “YES”, the process proceeds to ST105, and the number of defective pixels dn is updated to one larger value.
Note that the color difference Di obtained in ST103 is expressed as the Euclidean distance of the color in the color space for one pixel in the measurement target region and the corresponding pixel on the reference image, and is the same method as the equation (1). Can be sought.

  Hereinafter, similarly, by updating the value of i, while paying attention to each pixel in the measurement target region in order, the color difference Di with the corresponding pixel on the reference image is calculated for the target pixel, and the color difference Di is calculated. The number of pixels exceeding the threshold value D1 is counted as the defective pixel number dn (ST103 to 107). When the processing for all the pixels is completed, the process proceeds to ST108, and the number of defective pixels dn is compared with a predetermined threshold value dn1. Here, when the number of defective pixels dn exceeds the threshold value dn1, the process proceeds to ST109 to perform defect determination. On the other hand, if the number of defective pixels dn is equal to or less than the threshold value dn1, the process proceeds to ST110 to perform a good determination. Finally, in ST111, the determination result is output and the process is terminated.

According to the defect inspection described above, a defect can be easily extracted without performing individual inspection for each color of an object in which a plurality of types of colors are mixed. In addition, since the extraction is performed in units of pixels, even minute defects can be extracted.
However, only an aggregate in which a predetermined number or more of defective pixels (pixels having the color difference dn exceeding the threshold value dn1) may be extracted as defects. Further, the position and size of the defect may be extracted and output based on the coordinates of the defective pixel and the size of the aggregate.

(3) Position Measurement This measurement process is performed in order to detect the boundary between the background and the object (that is, the position of the contour line of the object) when both the background and the object are plain. This measurement process can also be used when extracting an object for setting a measurement target area in the defect inspection 1 described above.

This measurement process is based on the premise that the position of the contour line of the object is known to some extent, and the measurement target region is set in a range that may correspond to the contour line. FIG. 5 shows an example of setting the measurement target region. In the measurement target image 30, a rectangular measurement target region 32 that is long in the direction crossing the outline of the target 31 (in this example, the x-axis direction) is set. doing. In this embodiment, a window 33 having a predetermined width w (the length of the window 33 in the y-axis direction is the same as that of the measurement target area 32) is set at one end (the left end in the illustrated example) of the measurement target area 32. To do. Then, while moving the window 33 in the right direction, the average color in the window 33 is obtained each time the window 33 is moved by a distance corresponding to the width w. Further, by analyzing the average color obtained at each position, the x coordinate of the contour line of the object 31 is obtained.
The example in FIG. 5 is for extracting a contour line along the y-axis direction, and when extracting a contour line along the x-axis direction, a range that may correspond to the contour line. Then, a measurement target region that is long in the y-axis direction is set, and the y-coordinate of the contour line is obtained.

FIG. 6 shows the detailed procedure of this measurement process. In this example, it is assumed that the position of the contour line along the y-axis direction of the object is extracted based on the example of FIG.
In ST201, which is the first step, the measurement target area 32 is set in the manner shown in FIG. In the next ST202, the window 33 is set at the left end of the set measurement target region 32. In ST203, an average color in the window 33 is obtained. Further, the average color in the window 33 is obtained each time the window 33 is moved to the right by the number of pixels corresponding to the width w (ST 203 to 205). The average color may be calculated every time the window 33 moves by one pixel.

  The average color acquired by scanning the window 33 is stored in the memory 11 in association with labels (for example, P1, P2, P3...) Including numbers indicating the acquired order. Further, the position of the window 33 when each average color is obtained is expressed as an x coordinate of a representative point (such as the left end point) of the window 33. This x coordinate is also a label (including a number similar to the average color). For example, it is stored in the memory 11 in association with x1, x2, x3.

  When the window 33 moves to the right end of the measurement target region 32, in ST206, two average colors obtained at each position are combined, and the distance in the color space is calculated for each combination.

  In the next ST207, based on the distance calculation result, two colors located at a maximum distance in the color space are extracted, and a line segment L having these colors as endpoints is set.

  In the next ST208, the counter j is initialized to 1. This j corresponds to the average color label obtained by scanning the window. J = 1 at the time of initial setting corresponds to the average color acquired at the initial setting position of the window 33.

  In ST209, a perpendicular line from the point corresponding to the jth average color obtained to the line segment L is set in the color space. In ST210, the intersection of the perpendicular and the line segment L is extracted. In ST211, it is checked whether or not the intersection point exceeds the midpoint of the straight line L. Here, when it is determined that the middle point is not exceeded, j is incremented to focus on the average color obtained next, and ST209 to 211 are similarly executed.

  In the above processing, for any one of the average colors, when the intersection of the perpendicular line from the point corresponding to the average color and the line segment L exceeds the midpoint of the line segment L, the loop of ST209 to 212 is ended. Then, the process proceeds to ST213, and the position where the average color is obtained (the one corresponding to the label corresponding to the value of j among the set positions of the window) is determined as the position of the contour line. In ST214, the coordinates indicating the position of the boundary line are output as the determination result, and the process ends.

  FIG. 7 shows an example of the distribution of each average color in the color space. In this example, assuming that the window 33 is set at five locations, the points corresponding to the respective average colors are indicated as P1, P2, P3, P4, and P5 in the order of extraction. In the figure, O is the midpoint of the line segment L.

  When the measurement target region 32 is set so as to cross the outline of the target, one of the two colors having the largest distance is an average color when the window 33 is set as the background portion, and the other Can be considered to be the average color when the window 33 is set on the object. Further, while the window 33 passes the boundary between the background and the object, the ratio of both included in the window 32 is considered to change gradually. For this reason, it can be considered that the average color P1 acquired at the first set position and the average color P5 acquired at the last set position are the end points of the line segment L as shown in FIG.

  In the example of FIG. 7, when performing the processing of ST207 to 212, pay attention to the points P1, P2, P3, P4, and P5 corresponding to each average color in order, and the intersection of the perpendicular line from the point of interest to the line segment L Will be tracking. Here, the intersection point of the first point P1 is P1 itself, and the following intersection point moves toward the other end point P5 of the line segment L. In the example of FIG. 7, since the point P4 first exceeds the middle point O when viewed from the point P1, the position where the average color corresponding to this point P4 is obtained is determined as the position of the contour line. Note that the point for determining that the color has changed is not limited to the middle point O, and may be a point separated from the point P1 by an arbitrary distance (however, a point separated by a sufficient distance to be regarded as a change in color) There is a need to.).

  The above description is based on the premise that the measurement target region 32 is set so as to cross the outline of the target object. However, a misjudgment when the measurement target region 32 does not include an image of the target object. In order to prevent discrimination, when the line segment L is smaller than a predetermined threshold value, the processing may be stopped.

(4) Number inspection In this inspection, whether a predetermined number of objects are extracted by processing an image of an object in which a plurality of objects (for example, marks) including a plurality of kinds of colors are arranged. It is to determine whether or not. For this discrimination process, in this embodiment, an object having a model of an object having a good color distribution is imaged in advance, and an image including only one of the models is cut out from the obtained color image. Register as an image.

  Further, in this embodiment, after obtaining the average color for the reference image, the color of each pixel on the reference image is set to one point Mi in the color space, and the distance from the point Mμ indicating the average color to the point Mi And a direction from the point Mμ toward the point Mi. This distance and direction correspond to the length and direction of a vector starting from the point Mμ and ending at the point Mi. The direction can be expressed as an angle of the vector with respect to a reference plane (for example, a plane formed by the R axis and the G axis).

  The distance and direction obtained for each pixel are registered in the memory 11. At the time of inspection, a measurement target region having the same size as the reference image is scanned on the image to be inspected, and the similarity to the reference image is obtained using registered data in the memory 11 every time one pixel moves. Then, by counting the number of times that the degree of similarity exceeds a predetermined threshold C1, the number of objects is measured, and the suitability of the number is determined.

  FIG. 8 shows the detailed procedure of the inspection. In this procedure, (x, y) is the coordinates of each pixel of the image to be processed, ax indicates the number of pixels in the x-axis direction, and ay indicates the number of pixels in the y-axis direction. En is the number of extractions of the object.

  In ST301, which is the first step, the extraction numbers En and y are set to initial values 0, and in the next ST302, x is set to 0. In ST303, the measurement target region is set starting from the position (x, y). That is, the measurement target region is set so that the position of the coordinates (0, 0) is the upper left vertex.

  In the next ST304, an average color in the measurement target area is calculated. This average color is also represented by an average value for each of the R, G, and B color components, and can be represented as one point Pμ in the color space.

  The next ST305 is executed for each pixel in the measurement target region. Here, the color of the pixel to be processed is considered as one point Pi in the color space, and the distance from the point Pμ indicating the average color and the direction from the point Pμ to the point Pi are obtained. The distance and the direction correspond to the length and direction of the vector having the point Pμ as the start point and the point Pi as the end point, as in the case of the reference image.

  In ST306, using the distance and direction obtained in ST305 and the distance and direction of the reference image registered in the memory 11, the similarity C (x, y) is obtained based on the following equation (2). . In Equation (2), n is the total number of pixels in the measurement target area. D (Pi, Pμ) is the distance between the point Pi and the point Pμ, and d (Mi, Mμ) is the distance between the point Mi and Mμ on the reference image side. Further, θi is an angle formed by the vector formed by the points Pμ and Pi and the vector formed by the points Mμ and Mi, and is a numerical value in the range of 0 to 180 degrees.

  When the similarity C (x, y) is obtained in this way, in ST307, the similarity C (x, y) is compared with a predetermined threshold C1. If it is determined that the similarity C (x, y) exceeds the threshold value C1, the process proceeds to ST308, and the extraction number En is updated to one larger value.

  Thereafter, by moving the values of x and y in the ranges of 1 to ax and 1 to ay, respectively, the measurement target region is scanned pixel by pixel, and the correlation value C (x, y is obtained for each scan by the same processing as described above. ) And the number of times that the correlation value C (x, y) exceeds the threshold value C1 is counted as the extraction number En (ST302 to 312). When the scanning is completed, the process proceeds to ST313, and it is checked whether or not the extraction number En matches the reference number E0 registered in advance. If En = E0, the process proceeds to ST314, and it is determined that the object to be inspected is a non-defective product. On the other hand, if En ≠ E0, the process proceeds to ST315, and it is determined that the object to be inspected is a defective product. Thereafter, the process proceeds to ST316, and the determination result is output and the process is terminated.

  According to the above equation (2), the difference C (x, y) increases as the difference between d (Pi, Pμ) and d (Mi, Mμ) decreases and θi approaches 0. . Therefore, by setting the threshold value C1 to a sufficiently high value, an image region close to the color distribution state of the reference image can be extracted as an object.

  However, in equation (2), if the relative color change with respect to the average color is the same, the similarity C (x, y) also increases, so that a region having a different hue from the reference image is also extracted as an object. It will be. When it is necessary to extract only a region having the same hue as that of the reference image, the similarity C (x, y) may be corrected by the following equation (3).

In equation (3), dmax is the maximum distance that can be obtained in the color space. For example, when the density value of each color component is represented by 8 bits, dmax = (255 ² × 3). d (Pμ, Mμ) is a distance between the average color Pμ obtained from the measurement target region and the average color Mμ obtained from the reference image. According to the corrected similarity C ′ (x, y) according to the equation (3), even if the relationship of the colors Pi in the measurement target region to the average color Pμ is similar to the relationship in the reference image, the average color The similarity value decreases as Pμ moves away from the reference average color Mμ.

  In the procedure of FIG. 8, the number of objects is obtained from the number of times the degree of similarity C (x, y) exceeds the threshold value C1, but instead, the degree of similarity C (x, y) is calculated. The value of (x, y) when the threshold value C1 is exceeded may be output as the position of the object. Also, the suitability of the color distribution of the object can be determined based on the value of the similarity C (x, y).

It is a block diagram which shows the structural example of the image processing apparatus concerning this invention. It is a flowchart which shows the procedure of the defect inspection 1. It is a figure which shows the example of a division | segmentation of a measurement object area | region. It is a flowchart which shows the procedure of the defect inspection 2. FIG. It is a figure which shows the example of a setting of the measurement object area | region and window for a position measurement process. It is a flowchart which shows the procedure of a position measurement process. It is a figure which shows the specific example of the process of ST207-211 of FIG. It is a flowchart which shows the procedure of number inspection. It is a figure which shows the range in which the same value is set in the conversion process based on a color difference.

Explanation of symbols

1 Body 2 Camera 10 CPU
11 Memory 12, 13, 14 Image memory 14, 15, 17 Reference image memory

Claims (8)

In a method of processing a color image composed of density data of red, green and blue color components,
A process of setting two areas having the same size on a color image to be measured, a process of obtaining a color difference between areas based on an average color in each area, and a process of comparing the color difference with a predetermined threshold value The color image processing method is characterized in that it is determined whether or not there is a portion having a different color on the image by executing a plurality of cycles while changing the setting position of at least one region. In a method of processing a color image composed of density data of red, green and blue color components,
While scanning the color image along a predetermined direction, repeatedly performing a process for obtaining an average color along a direction orthogonal to the scanning direction at predetermined intervals,
In the three-axis virtual space corresponding to each color component, a point corresponding to the color boundary is extracted from these points based on the positional relationship of the points corresponding to the plurality of average colors obtained when scanning the color image. And a color image processing method, wherein the scanning position when the average color corresponding to the point is obtained is determined as the position where the color has changed. In a method of processing a color image composed of density data of red, green and blue color components,
By executing a process of obtaining a color difference for each set of corresponding pixels between a color image to be measured and a reference color image registered in advance, and a process of comparing the color difference with a predetermined threshold value A method of processing a color image, wherein a portion where a color different from a reference color image appears in the measurement target image is extracted. In a method of processing a color image composed of density data of red, green and blue color components,
On the measurement target color image, a measurement target area of the same size as the reference color image registered in advance is provided.
For each of the image in the measurement target area and the reference color image, the average color is calculated for each point corresponding to the color of each component pixel in the virtual space with three axes corresponding to each color component. A process for obtaining a distance and a direction with respect to a point corresponding to
Using the distance and direction obtained for each of the constituent pixels, the similarity between the image in the measurement target region and the reference color image is obtained, and when the similarity exceeds a predetermined threshold, the measurement target A method of processing a color image, characterized in that it is determined that an area corresponds to a reference color image. An image input means for inputting a color image composed of density data of each color component of red, green and blue;
After setting a measurement target area of a predetermined size on the color image input by the image input means, area setting means for dividing the measurement target area and setting a plurality of small areas;
After calculating the average color for each small area, calculating means for determining a color difference based on the average color for each set of adjacent small areas;
The color difference obtained by the calculating means is compared with a predetermined threshold value, and when the color difference exceeds the threshold value, it is determined that a color difference is generated between two small areas corresponding to the color difference. An image processing apparatus comprising: a determination unit that performs the determination. An image input means for inputting a color image composed of density data of each color component of red, green and blue;
Average color calculation means for obtaining an average color along a direction orthogonal to the scanning direction while scanning a color image input from the image input means along a predetermined direction;
Based on the positional relationship of points corresponding to a plurality of average colors obtained by the average color calculation means in a virtual space with three axes corresponding to the color components, points corresponding to color boundaries are selected from these points. Extracting means for extracting;
An image processing apparatus comprising: discrimination means for discriminating a scanning position when an average color corresponding to the point extracted by the extraction means is obtained as a color changed position. An image input means for inputting a color image composed of density data of each color component of red, green and blue;
A memory for registering a reference color image;
An arithmetic means for obtaining a color difference for each set of corresponding pixels between a color image input from the image input means and a reference color image in the memory;
The color difference obtained by the computing means is compared with a predetermined threshold value, and when the color difference exceeds the threshold value, a color different from the reference color image appears in the pixel from which the color difference is extracted. An image processing apparatus comprising: discrimination means for discriminating that it is present. An image input means for inputting a color image composed of density data of each color component of red, green and blue;
A memory for registering a reference color image;
Area setting means for setting a measurement target area having the same size as the reference color image on the color image input from the image input means;
For each of the image in the measurement target region and the reference color image, a process for calculating an average color in the image, and for each point corresponding to the color of each constituent pixel of the virtual space by three axes corresponding to each color component, The degree of similarity between the image in the measurement target region and the reference color image using the obtained distance and direction for each constituent pixel after executing the processing for obtaining the distance and direction for the point corresponding to the average color Computing means for obtaining
An image processing apparatus comprising: a determination unit that compares the similarity with a predetermined threshold and determines that the measurement target area corresponds to a reference color image when the similarity exceeds the threshold. .

Images