JP4534599B2 - Image processing method - Google Patents

Description

  The present invention relates to an image processing method for dividing a multivalued image in which pixel values of each pixel are assigned in multiple stages, such as a monochrome grayscale image, a color image, and a distance image, into a target area and a background area. .

  In recent years, image processing techniques have been used to identify, inspect, measure, and recognize various objects using images. In this type of image processing technology, since it is necessary to identify an object in the image, various techniques for extracting the object in the image have been proposed. For example, by performing illumination from the appropriate direction according to the positional relationship between the imaging device and the target object, the difference in density value between the target object and the background in the image is increased, and the pixel of each pixel in the grayscale image A technique for separating an object from a background by binarizing a grayscale image by applying an appropriate threshold to a density value that is a value is widely used.

  However, the technology that simply binarizes the grayscale image separates the object from the background when the difference in density value between the object and the background is insufficient or when there is a ground pattern in the background. Have the problem of not being able to. For example, in an optical character reader, if there is a pattern or ground pattern around a character, there is a problem that the character recognition accuracy is lowered. In addition, when the type of an article is determined by performing pattern matching in an image obtained by imaging an article conveyed on a conveyor in a production line, if the image includes a belt conveyor or the like, the part May not be identified.

Therefore, various techniques for extracting a target region from the image by dividing the region into parts having similar features in the image have been proposed. For example, regarding the processing of photographic images, in order to divide the image into “light”, “intermediate”, and “shadow”, clustering is performed by integrating the blocks by blocking the images and evaluating the similarity by pairing the blocks, There has been proposed a technique for re-evaluating an area formed by clustering while referring to an integration history to determine a contour (for example, see Patent Document 1).
Japanese Patent Laid-Open No. 8-167028

  In the technique described in Patent Document 1 described above, it is divided into three types of areas by clustering. However, this technique is not a technique that sufficiently removes noise when noise is present, and is particularly suitable for character recognition and component type recognition. Thus, it cannot be used for the purpose of separating the object area from the background.

  The present invention has been made in view of the above-described reasons, and an object thereof is an image processing method capable of accurately separating a background area and a target area even in a noisy image. Is to provide.

In the first aspect of the present invention, after dividing a multi-valued image into a plurality of rectangular blocks each including a plurality of pixels, the feature values relating to the distribution of pixel values in the block are compared between each pair of adjacent blocks. The multi-value is obtained by assigning the same label to each pair of blocks in which the similarity of the feature quantity is equal to or greater than the specified value and assigning a different label to each pair of blocks in which the similarity of the feature quantity is less than the specified value. Labels all blocks in the image, and further divides the block located at the boundary of the block connected area consisting of the blocks with the same label into a plurality of divided blocks. When the similarity between the feature quantity of the adjacent block and the feature quantity related to the distribution of pixel values of the divided blocks is equal to or higher than a specified value, the same label as the adjacent block is divided. By assigning to a lock, the divided block is integrated into the block connected area as a block constituting the block connected area, and divided into divided blocks, and the process of integrating the divided block into the block connected area is repeated to obtain a multi-valued image. The first process of integrating all the included pixels into any block connection area, and the boundary of the block connection area is searched by raster scanning, and a judgment area including a plurality of block connection areas is set around the boundary of the block connection area Then, one block connection region of the two block connection regions sandwiching the boundary is set as a block connection region to which attention is paid, and the average density of the other block connection region in the ratio of the area occupied by the other block connection region to the determination region a second process of obtaining a value obtained by multiplying the value as a density evaluation value of the block coupling region the noted A third process of obtaining an identification evaluation value for the difference between the average density value and the density evaluation value as a parameter, the interest and the attention block connected region by comparing the magnitude of the identification evaluation value specified threshold value And a fourth step of separating a block connection region around the block region.

According to this method, blocks that have similar pixel value distribution among adjacent blocks are assigned the same label to form a block connection area, so that such as blurred characters or shaded marks, etc. Even if the pixel value distribution is different from the background but the pixel value varies within the same region, the region can be correctly extracted. In addition, since the blocks are subdivided only at the boundary part of the block connection area and labels are assigned to all the pixels, the outline shape is accurately extracted at the boundary part of the block connection area, but the overall processing load is low, and noise is also generated. Can be easily extracted from the image, and high-precision and high-speed processing can be expected in character recognition and article type recognition. Further, the boundary of the block connection area is searched by raster scanning, a density evaluation value reflecting the density value is obtained for the block connection area around the boundary, and the difference between the density value and the density evaluation value of the block connection area of interest is calculated. since separated into the block coupling region of interest block coupling region near to the block area of interest (the object region to be described later) (background region to be described later) using an identification evaluation value with the parameter, it contains many noise separating accurately and block coupling region near to the block area of interest as a block coupling region of interest using the density evaluation value reflecting the density difference in an image is a by near the boundary of the even block coupling region, such as Is possible.

In addition, since the density evaluation value is determined using the average density value and the occupied area for two of the plurality of block connection areas included in the determination area set around the boundary, the density value near the boundary is determined. it is possible to obtain a density evaluation value that reflects the change in, it is possible to enhance the separation accuracy of the block coupling region near to the block area of interest and the focused block linking region. In particular, weighting is performed by the ratio of the occupied area of each block connection area within the judgment area, so that not only the difference in density value between the target block connection area and the surrounding block connection area with respect to the target block area , but also the area difference. By adding this information, the possibility of separating the two can be increased.

According to the second aspect of the present invention, after dividing a multi-valued image into a plurality of rectangular blocks each including a plurality of pixels, the feature values relating to the distribution of pixel values in the block are compared between each pair of adjacent blocks. The multi-value is obtained by assigning the same label to each pair of blocks in which the similarity of the feature quantity is equal to or greater than the specified value and assigning a different label to each pair of blocks in which the similarity of the feature quantity is less than the specified value. Labels all blocks in the image, and further divides the block located at the boundary of the block connected area consisting of the blocks with the same label into a plurality of divided blocks. When the similarity between the feature quantity of the adjacent block and the feature quantity related to the distribution of pixel values of the divided blocks is equal to or higher than a specified value, the same label as the adjacent block is divided. By assigning to a lock, the divided block is integrated into the block connected area as a block constituting the block connected area, and divided into divided blocks, and the process of integrating the divided block into the block connected area is repeated to obtain a multi-valued image. a first step of integrating any of the blocks connected region all pixels included searches block boundaries connected region by raster scan, a determination region including a plurality of blocks connected region around the boundary of the block coupling region set, the average density value of each block coupling region weighting factor the ratio of the area occupied by the block coupling region determination area other than the block coupling region of interest among the blocks connecting region including at least part the decision region a second process of obtaining a weighted average value obtained by averaging by multiplying the weighting factor as the density evaluation value and the focus The third process of obtaining an identification evaluation value using the difference between the average density value and the density evaluation value of the lock connection area as a parameter, and the block connection area of interest by comparing the identification evaluation value with a prescribed threshold value And a fourth step of separating a block connection region around the block region of interest .

According to this method, blocks that have similar pixel value distribution among adjacent blocks are assigned the same label to form a block connection area, so that such as blurred characters or shaded marks, etc. Even if the pixel value distribution is different from the background but the pixel value varies within the same region, the region can be correctly extracted. In addition, since the blocks are subdivided only at the boundary part of the block connection area and labels are assigned to all the pixels, the outline shape is accurately extracted at the boundary part of the block connection area, but the overall processing load is low, and noise is also generated. Can be easily extracted from the image, and high-precision and high-speed processing can be expected in character recognition and article type recognition. Further, the boundary of the block connection area is searched by raster scanning, a density evaluation value reflecting the density value is obtained for the block connection area around the boundary, and the difference between the density value and the density evaluation value of the block connection area of interest is calculated. It contains a lot of noise because it separates into the block connection area of interest (object area described later) and the block connection area around the block area of interest (background area described later) using the identification evaluation value of the parameter. Even in such an image, the block connection region of interest and the block connection regions around the block region of interest are accurately separated using the density evaluation value reflecting the density difference near the boundary of the block connection region. Is possible. In addition, since the density evaluation value is determined using the overall average density value and the occupied area of the plurality of block connection areas included in the determination area set around the boundary, the change in density value near the boundary is determined. it is possible to obtain a density evaluation value that reflects, it is possible to enhance the separation accuracy of the block coupling region near to the block coupling region of interest and the focused block linking region. In particular, since the weighted average value of the density values using the ratio of the occupied area of each block connection area in the judgment area as a weighting factor is used, the block connection area of interest and the block connection areas around the block connection area of interest are By adding not only the density value difference but also the area difference information, it is possible to increase the possibility of separating the two.

According to a third aspect of the present invention, in the first or second aspect of the present invention, in the second step, evaluation image data is generated in which the obtained density evaluation value is used as a pixel value of a block connection area to which attention is focused, and density evaluation is performed. It is characterized in that the density evaluation value is not obtained again for the block connection area whose value has been obtained once.

  According to this method, since the density evaluation value is not obtained again for the block connection area in which the density evaluation value is once obtained in one image, the density evaluation value can be obtained with a small amount of calculation. High speed processing can be expected.

According to the method of the present invention, blocks having similar pixel value distribution among adjacent blocks are given the same label to form a block connection area, so that a blurred character, a shaded mark, etc. There is an advantage that a region can be correctly extracted even when the pixel value distribution is different from the background but the pixel value varies within the same region. In addition, since the blocks are subdivided only at the boundary part of the block connection area and labels are assigned to all the pixels, the outline shape is accurately extracted at the boundary part of the block connection area, but the overall processing load is low, and noise is also generated. Can be easily extracted from the image, and high-precision and high-speed processing can be expected in character recognition and article type recognition. Further, the boundary of the block connection area is searched by raster scanning, a density evaluation value reflecting the density value is obtained for the block connection area around the boundary, and the difference between the density value and the density evaluation value of the block connection area of interest is calculated. Since the block connection area (object area described later) and the block connection area around the block connection area (background area described later) with respect to the target block connection area are separated by using the identification evaluation value of the parameter, a lot of noise is included. separating accurately and block coupling region near to the block coupling region of interest and block coupling region the focused using the density evaluation value reflecting the density difference be an image of the block coupling region in the vicinity of the boundary of being There is an effect that it becomes possible to do.

In the present embodiment, as shown in FIG. 2, an object 1 on which characters are written is imaged using an imaging device 2 such as a CCD camera, and other character regions are displayed in the image obtained by the imaging device 2. As an example, a case of separation from the region will be described. The imaging device 2 is assumed to output a monochrome signal. The output of the imaging device 2 is input to an image processing device 10 that performs processing described later. The output of the imaging device 2 is an analog signal that is converted into a digital signal by the A / D converter 11 provided in the image processing device 10 and then stored in the storage means 12 provided in the image processing device 10. The storage unit 12 temporarily stores the grayscale image output from the A / D converter 11 and temporarily stores data generated during various processes described later on the grayscale image. A memory having a storage area and a storage device such as a hard disk device for storing various data for a long period of time are included. The grayscale image stored in the temporary storage area in the storage unit 12 is a multivalued image in which pixels are arranged in a matrix and multi-stage pixel values are set for each pixel. In this embodiment, since a grayscale image (gray image) is used, a density value is used as a pixel value in the following. However, a color image or a distance image can be used as a multivalued image, and a hue value is used as a pixel value in a color image. Alternatively, saturation may be used, and the distance is used as the pixel value in the distance image. Further, in a color image, one pixel value is associated with each pixel using a color difference from an appropriate reference color, or the lightness L ^* chromaticity a ^* , b ^* of L ^* a ^* b ^* color system is 3 One value may correspond as a pixel value of one pixel.

  The image processing apparatus 10 includes a microcomputer as a main component, and performs the processing shown in FIG. 3 on the image stored in the storage unit 12, thereby separating the character area, which is the object of the present embodiment, from other areas. . Note that the image processing apparatus 10 is provided with a monitor 3 using a CRT or a liquid crystal display, so that the image stored in the storage means 12 and the image being processed can be monitored by the monitor 3.

  Assume that the image processing apparatus 10 extracts a character area from the image stored in the storage unit 12 (that is, divides the image into a character area and another area). In addition, in the character area of the image stored in the storage unit 12, there may be a discontinuous portion (a gap between the left and right portions) or a portion with large unevenness in brightness. In such an image, if binarization is simply performed with pixel values (density values), characters with discontinuous portions cannot be handled as a whole (as one character), and characters with large unevenness of light and darkness have a shape. It becomes impossible to extract clearly. Furthermore, noise generated by the image pickup means 2 may be included in the image. If noise is included, pixels having the same pixel value as characters may be irregularly generated in the binarized image. is there. In other words, when information other than characters is included, information that cannot be handled as a whole although it is originally a single character, or the shape of characters is unclear, optical When trying to recognize a character in a character reader or the like, if the density value of the original image is only binarized, the information obtained will be excessive or insufficient, and the character recognition rate will decrease.

  Therefore, unnecessary information other than characters (hereinafter referred to as “background region”) can be removed from the image, and can be handled as a single region for each character, and the density value varies among characters. Even if there is, it is required to be able to extract the shape of the character almost accurately. Hereinafter, a target area such as a character is referred to as an “object area”. In an optical character reader, in order to improve the character recognition rate, it is required as a pre-process to separate an object region, which is a character region, and a background region. Even when the type of article is recognized by pattern matching, separation of the object area and the background area, which is the article area, is required as preprocessing in order to increase the accuracy of pattern matching.

In order to perform such pre-processing, in the present embodiment, the procedure shown in FIG. 3 is performed on the grayscale image stored in the storage unit 12 to separate the object region and the background region. That is, first, in the image division processing unit 13, the grayscale image stored in the storage means 12 is a block B (see FIG. 4) that is a square region having one side of a power of 2 (2 ⁿ pixels ^where n is a positive integer). (S1). The size of the block B is set so that one side of the block B is about the line width of the character if the character is an object region. The size of such a block B can be automatically set based on the spatial frequency in the image.

  Next, each block B obtained by the image division processing unit 13 is transferred to the feature amount calculation unit 14, and the feature amount relating to the distribution of the pixel values of each block B is obtained (S2). In the present embodiment, the average value and standard deviation of the density values of each pixel included in the block B are used as the feature amount. The feature amount obtained for each block B is compared in the tracking processing unit 15, and the similarity between the feature amount of the block B adjacent to the target block B and the target block B is equal to or greater than a specified value. Block B is tracked (S3). When the similarity between each pair of adjacent blocks B is equal to or greater than a specified value, the same label is assigned to both blocks B, thereby forming a block connection region composed of a plurality of blocks B to which the same label is assigned. That is, adjacent blocks B whose feature amount similarity is equal to or greater than a predetermined value with respect to the target block B are sequentially tracked, and block connected regions are formed by the tracked blocks B in the form shown in FIG. (In FIG. 5, a group of different shades is a block connection area). For the block B that can be traced, a label is assigned to each block connection area and stored in the storage unit 12. A processing procedure for comparing block feature areas by comparing feature quantities of adjacent blocks B will be described later.

  On the other hand, when the block B is tracked as described above and a plurality of block connection areas are formed in the grayscale image, the block B at the boundary of the block connection areas to which different labels are assigned is transferred to the block division processing unit 16. Deliver (S4). The block division processing unit 16 forms divided blocks Bn (= B1, n is a positive integer, see FIG. 6) obtained by dividing each block B, which is a boundary portion of the block connection area, into two vertically and horizontally (S5, S6). That is, the number of pixels included in the divided block B1 is a quarter of the number of pixels in the block B. The divided block B1 generated in the block division processing unit 16 is delivered to the feature amount calculation unit 14 to obtain the feature amount, and the feature amount obtained for the divided block B1 is obtained from the feature amount of the surrounding block B in the tracking processing unit 15. The similarity to is evaluated. When the similarity between the feature amount of the divided block B1 and the feature amount of the surrounding block B is equal to or greater than the specified value, the same label as the surrounding block B is given to the divided block B1, and the divided block B1 is the surrounding block B1. (S7). By such labeling processing, a label is given to each divided block B1, and the shape of the block connection area is refined in units of the division block B1 at the boundary portion of the block connection area. In this way, since the divided blocks B1 are arranged at the boundary portion of the block connection area, the divided block B1 is treated as equivalent to the block B, the divided block B1 is further divided into divided blocks B2, and the above processing is repeated. Specifically, it is divided into units of one pixel (S8, S9). Here, it goes without saying that the block connection area formed using the divided blocks Bn is stored in the storage means 12. If all the blocks B and the divided blocks Bn (n is a positive integer) are labeled as described above, each block connection region is considered to be an object region or a background region. Can be separated.

Through the processing described above, block connection areas are extracted from the image, and labels are assigned to the respective block connection areas Di (i = 1 to 17) as shown in FIG. By the way, the original image generated from the image shown in FIG. 7 is a gray image including the character “8”, and the character is darker than the background (the density value is small), but the character cannot be easily identified in the original image. It is assumed that there is a lot of noise (such as when there are water droplets or bubbles around the character). At the stage of extracting the block connection area Di, the area that should be one character is divided into a plurality of block connection areas Di, and cannot be recognized as characters as it is. Therefore, in the present embodiment, each block connection region Di is used by using an evaluation value (hereinafter referred to as “identification index value”) that reflects a difference in density value of pixels included in adjacent block connection regions Di. Is provided for separating the image into a background area and an object area. As will be described later, the identification index value rvi includes a parameter corresponding to the difference in density value of the pixels included in each block connection area Di, and represents the ease of distinguishing the block connection area Di like the contrast. . If separating the background region and the object region, it is possible to determine the character from the positional relationship of the object body region.

  A processing procedure for separating each block connection area Di into a background area and an object area after the generation of the block connection area Di to which the label is attached is as shown in FIG. That is, the boundary of the block connection area Di is first searched by performing raster scanning on the entire image (S1). When the boundary of the block connection area Di is found, the determination area Dd within a predetermined range is set around the found position (S2). The determination area Dd includes a plurality of determination points Pd distributed at equal intervals, and the number of determination points Pd in each block connection area Di at least partially included in the determination area Dd is obtained (S3). In short, an index of the area occupied by each block connection region Di in the determination region Dd is obtained as the number of determination points Pd.

  Next, in order to evaluate the contrast with the periphery of each block connection area Di included in the determination area Dd, first, a standard of the density value of the periphery excluding the block connection area Di of interest in the determination area Dd is obtained. (S4). Below, the standard of the density value around each block connection area Di is referred to as the density evaluation value SaIi for the block connection area Di to which attention is paid. The density evaluation value SaIi is an area (using the number of determination points Pd) occupied by the peripheral block connection area Di excluding the block connection area Di of interest in the determination area Dd, and the average value of the density values of the block connection areas Di. (Hereinafter referred to as “average density value”) Iai. Therefore, the difference between the average density value Iai and the density evaluation value SaIi of the block connection area Di of interest reflects the contrast between the block connection area Di of interest and the surrounding area.

  As described above, since the difference between the average density value Iai and the density evaluation value SaIi of the block connection area Di of interest reflects the contrast, it can be used as the identification index value rvi. However, since the identification index value rvi is used as an index for separating the object region and the background region, it is easier to separate the object region and the background region by including not only the density value but also the area of the block connection region Di as a parameter. It is thought that it becomes.

  Therefore, in the present embodiment, as the identification index value rvi, a value obtained by multiplying the difference between the average density value Iai and the density evaluation value SaIi of the block connection area Di by the total number of pixels ndi of the block connection area Di of interest is used. (S5). That is, since the difference in density value between the background area and the object area is large, the absolute value of the difference between the average density value Iai of the block connection area Di serving as the background area and the density evaluation value SaIi for the block connection area Di is relatively Since the background area is generally larger than the object area, the number of pixels ndi of the block connection area Di serving as the background area is multiplied by the difference between the average density value Iai and the density evaluation value SaIi. It is considered that the absolute value of the obtained identification index value rvi increases with a significant difference compared to the absolute value of the identification index value rvi obtained for the object region.

  In the present embodiment, the identification index value rvi of each block connection area Di is obtained based on such knowledge, and each block connection area is binarized by binarizing the identification index value rvi using an appropriate threshold Th. Di is separated into a background area and an object area. In setting the threshold value Th, an average value arv of the identification index value rvi is obtained in the entire image (S6), and the threshold value is determined between the average value arv of the identification index value rvi and the maximum value or the minimum value of the identification index value rvi. Th is set (S7). A method for obtaining the threshold Th will be described later.

  In the following description, the case shown in FIG. 7 is taken as an example. In the illustrated example, 17 block connection areas Di are provided, the two block connection areas D1 and D2 belong to the background area, and the remaining block connection areas D3 to D17 belong to the object area. Further, it is assumed that the block connection area D1 is brighter (the density value is higher) than the object area, and the block connection area D2 is darker than the object area.

  As described above, for each block connection area Di, the average density value Iai of each block connection area Di is required to obtain the density evaluation value SaIi, and the number of pixels in the block connection area Di is required to obtain the identification index value rvi. ndi is required. Therefore, the average density value Iai and the number of pixels ndi are obtained for each block connection area Di. In the following description, the values shown in Table 1 are used as the average density value Iai and the number of pixels ndi of each block connection area Di.

  As described in step S1 in FIG. 1, the boundary of the block connection area Di is obtained by raster scanning. That is, a pixel on the boundary between two adjacent block connection areas Di is searched by raster scanning in increments of one pixel, and the position where the label of the block connection area Di is changed is located on the boundary between the two block connection areas Di. Pixels. Further, as in step S2, the determination area Dd is set around the obtained pixel, and the density evaluation value SaIi is obtained using the block connection area Di including at least a part of the determination area Dd.

  In the example shown in FIG. 8, the determination area Dd is a 31 × 31 square area, and in the determination area Dd, seven determination points Pd each having a size of one pixel are arranged at equal intervals in the vertical and horizontal directions. . That is, a total of 49 determination points Pd are set in the determination area Dd, and the distance between adjacent determination points Pd is set to 5 pixels (4 pixels are included between the determination points Pd). However, the size of the determination area Dd, the interval between the determination points Pd, and the number of the determination points Pd shown in FIG. 8 are merely examples, and a setting input unit is provided so that the user can set as appropriate. The determination area Dd does not necessarily have a square shape as long as it includes a plurality of block connection areas Di. For example, the determination area Dd having a circular shape or the like can be set. However, the square-shaped determination area Dd is easy to set and the calculation described later is easy.

  In the example shown in FIG. 8, a pixel at the boundary between the block connection area D1 and the block connection area D3 is detected by raster scanning, and a determination area Dd is set with this pixel as a central pixel (a pixel indicated by reference numeral pc in the drawing). ing. The determination area Dd includes block connection areas D2 and D4 in addition to the two block connection areas D1 and D3 that sandwich the boundary. That is, four block connection areas D1 to D4 existing around the boundary position obtained by searching by raster scanning are included in the determination area Dd. The density evaluation value SaIi is obtained using the pixels included in each of these four block connection areas D1 to D4.

  In the example shown in FIG. 8, the determination area Dd includes four block connection areas D1 to D4, and the number of determination points Pd included in each of the block connection areas D1 to D4 is 31, 12, and 4, respectively. There are two. When the number of determination points Pd included in each connected region Di is represented by ni, n1 = 31, n2 = 12, n3 = 4, and n4 = 2. Further, as is apparent from the figure, the pair of adjacent block connection regions D1 to D4 is (D1, D3) (D2, D3) (D2, D4). The density evaluation value SaIi is obtained by one of the following methods.

  As described above, the density evaluation value SaIi for each block connection area Di is the density value of the block connection areas Di around the block connection area Di excluding the target block connection area Di among the block connection areas Di at least partially included in the determination area Dd. Therefore, as the simplest density evaluation value SaIi, one of two adjacent block connection areas Di is set as a block connection area Di of interest, and the average density value Iai of the other block connection area Di is set as Can be used.

  In the example shown in FIG. 8, the adjacent block connection area Di in the determination area Dd is a combination of (D1, D3) (D2, D3) (D2, D4), but the boundary where the determination area Dd is set is a block. Since this is the boundary between the connection region D1 and the block connection region D3, the density evaluation values SaI1 and SaI3 are obtained for the combination of (D1, D3). That is, if attention is focused on the block connection area D3, the density evaluation value SaI3 is the average density value Ia1 of the block connection area D1, which is 220. If attention is paid to the block connection area D1, the density evaluation value SaI1 is the average of the block connection area D3. The density value Ia3 is 35.

  When the density evaluation value SaIi is obtained by the above-described method, the determination area Dd is not necessary because it is only necessary to know which block connection area Di is the boundary where the label of the block connection area Di is changed. However, instead of the average density value Iai of each block connection area Di, a density value is obtained for each judgment point Pd included in each block connection area Di in the judgment area Dd, and obtained for each block connection area Di. It is also possible to use the average value of the density values as the density evaluation value SaIi. In this case, although the amount of calculation increases, the average value of the density values is obtained in the determination area Dd. It becomes easier to reflect changes in density values. Specifically, the density value of the determination point Pd at the position of m rows and n columns in the determination region Dd is I (m, n), and one of the two block connection regions Di and Dj across the boundary is set. The density evaluation value SaIi for the block connection area Di is obtained by Equation 1. That is, the density evaluation value SaIi for the block connection area Di is an average value of the density values of the other block connection area Dj. In Equation 1, (m, n) represents the position of the determination point Pd included in the block connection region Dj in the determination region Dd, and nj is the number of determination points Pd included in the block determination region Dj.

  Although the method of obtaining the average value of the density values of each block connection area Di in the determination area Dd by Expression 1, the calculation amount increases although the change of the density value near the boundary is reflected. Therefore, instead of the average value of the density values of each block determination area Di, a value obtained by weighting the average density value Iai of each block determination area Di as a weighting factor is the ratio of the area occupied by each block connection area Di to the determination area Dd. You may make it use. Since this value is a simple calculation by simply multiplying the average density value by a weighting factor, the amount of calculation can be reduced as compared with the case of using Equation 1.

That is, the weighting coefficient is ni / n0, where ni is the number of determination points Pd included in each block connection area Di and n0 (= 49) is the total number of determination points Pd included in the determination area Dd. Therefore, when obtaining the density evaluation value SaIi for each block connection area Di, the value obtained for each block connection area Di is Iai · ni / n0 using the average density value Iai. The values obtained for each of the block connection areas D1 to D4 are as follows.
D1: 220 × 31/49 = 139.18
D2: 130 × 12/49 = 31.84
D3: 35 × 4/49 = 2.86
D4: 30 × 2/49 = 1.22
In the example shown in FIG. 8, since the boundary where the determination area Dd is set is the boundary between the block connection area D1 and the block connection area D3, the density evaluation value SaI1 for the block connection area D1 is a value obtained from the block connection area D3. Therefore, the density evaluation value SaI3 for the block connection area D3 is a value obtained from the block connection area D1 and is 139.18.

  The calculation using the weighting factor ni / n0 reflects the proportion of the area occupied by each block connection area Di in the determination area Dd. When the object area is based on characters, the background area in the determination area Dd is more than the object area. Since the occupied area also increases, the background region has a larger weight coefficient. The weighting factor described above is a value used when the density value of the background area is higher than the density value of the object area (when the background area is brighter than the object area), and the density evaluation value SaIi for the background area and the object area The ratio with the density evaluation value SaIi is increased, and the density evaluation value SaIi for the background area is significantly smaller than the density evaluation value SaIi for the object area. For example, when the average density value Iai is used, the density evaluation values SaI1 and SaI3 are 35, 220 and 6.29 times, respectively, whereas when the weight coefficient ni / n0 is multiplied, the density evaluation value is obtained. SaI1 and SaI3 are 2.86 and 139.18, respectively, which is 48.66 times. If the background area is darker than the object area, (1-ni / n0) may be used as a weighting factor.

  The density evaluation value SaIi obtained by each of the methods described above is obtained only for the two block connection regions Di sandwiching the boundary, but cannot be said to reflect the entire density value around the boundary. . Therefore, as the density evaluation value SaIi reflecting the entire density value around the boundary, the block connection area Di other than the target block connection area Di among the block connection areas Di at least partially included in the determination area Dd. A method of using a weighted average value of the average density value Iai will be described. In the weight coefficient for obtaining the weighted average value, among the block connection areas Di at least partially included in the determination area Dd, each block connection area Di excluding the block connection area Di of interest occupies the determination area Dd. Use the area ratio. In other words, if the number of determination points Pd included in each block connection region Di is represented by ni, the weighting coefficient uses the number ni of determination points Pd included in the block connection region Di excluding the block connection region Di of interest. Are represented as ni / Σni, respectively.

For example, in FIG. 8, if the density evaluation value SaI3 for the block connection area D3 is to be obtained, the number of determination points Pd included in the block connection areas D1, D2, and D4 is 31, 12, 2, respectively. The weighting factors for the connected regions D1, D2, and D4 are 31 / (31 + 12 + 2), 12 / (31 + 12 + 2), and 2 / (31 + 12 + 2), respectively, and the average density values Ia1, Ia2, and Ia4 are 220, 130, Since it is 30, SaI3 = (220 × 31 + 130 × 12 + 30 × 2) / (31 + 12 + 2) = 187.56. Similarly, SaI1, SaI2, and SaI4 can be obtained as follows.
SaI1 = (130 × 12 + 35 × 4 + 30 × 2) / (12 + 4 + 2) = 97.78
SaI2 = (220 × 31 + 35 × 4 + 30 × 2) / (31 + 4 + 2) = 189.73
SaI4 = (220 × 31 + 130 × 12 + 35 × 4) / (31 + 12 + 4) = 181.28
The density evaluation value SaIi obtained by the above calculation becomes larger as the average density value Iai around the block connection area Di of interest is higher, and the weight is increased as the ratio of the block connection area Di around the determination area Dd is larger. The coefficient ni / Σni becomes small. In general, since the background area has a larger area than the object area, when the average density value Iai of the background area is higher than the average density value Iai of the object area, the density evaluation value SaIi of the object area is calculated by the above calculation. The density evaluation value SaIi is larger.

After the density evaluation value SaIi is obtained by any of the methods described above (step S4 in FIG. 1), the identification index value rvi can be obtained using the density evaluation value SaIi. The identification index value rvi is a value that reflects the difference in density value of the pixels included in each block connection area Di, and reflects the difference in density value between the block connection area Di of interest and the surrounding block connection area Di. The identification index value rvi can be expressed by the following equation, where Iai is the average density value of the block connection area Di of interest, ndi is the total number of pixels of the block connection area Di of interest, and SaIi is the density evaluation value.
rvi = (Slai-Iai) × ndi
For example, with respect to the block connection areas D1 to D4, the identification index values rv1 to rv4 are obtained by using, as an example, the weighted average values of the average density values Ia1 to Ia4 related to the block connection areas D1 to D4 as the density evaluation values SaI1 to SaI4. And the following.
rv1 = (97.78−220) × 5300 = −647766
rv2 = (189.73−130) × 3100 = 185163
rv3 = (187.56-35) × 250 = 38140
rv4 = (181.28-30) × 22 = 3328.16
The identification index value rvi obtained as described above is a negative value in the block connection area Di having a higher density value than the surrounding area, and the background area has a larger number of pixels than the object area. The value increases. That is, the identification index value rvi of the background area (for example, the block connection area D1) brighter than the object area has a large absolute value and a negative sign. The density evaluation value SaIi used for obtaining the identification index value rvi does not necessarily need to be a weighted average value of the average density values in each block connection region D. The density evaluation value SaIi obtained by any of the methods described above is used. It is possible to use.

  In order to separate the object area from the background area in the entire image, it is necessary to determine the identification index value rvi for all the block connection areas Di. However, if the calculation for obtaining the identification index value rvi for the boundary is found every time a boundary where the label of the block connection area Di is changed by raster scanning, the boundary of the block connection area Di of the same combination is detected many times. As a result, the amount of calculation increases, and the identification index value rvi for each block connection area Di cannot be uniquely determined. Therefore, one value is determined from a plurality of identification index values rvi for each block connection area Di. The calculation for this is also needed.

  Therefore, in this embodiment, after obtaining the identification index value rvi for one block connection area Di, evaluation image data is generated with the obtained identification index value rvi as the pixel values of all the pixels in the block connection area Di, For the block connection area Di whose pixel value is determined in the evaluation image data, the pixel value is not changed until the processing of the image is completed. That is, in the evaluation image data, the identification index value rvi serving as an image value is obtained only once for each block connection area Di, and the determination of the identification index value rvi for each block connection area Di is facilitated and the amount of calculation is large. Therefore, the processing speed can be increased.

  Further, for the block connection area Di for which the identification index value rvi is set in the evaluation image data, the identification index value rvi is set even if a boundary between the block connection area Di and another block connection area Di is found by raster scanning. The determination area Dd is not set when the boundary is between the same block connection areas Di as when the determination is made, and the determination area Dd is set only when the boundary is between different block connection areas Di. Further, when pixel values have already been set in the evaluation image data for all the block connection areas Di included in the newly set determination area Dd, the calculation of the identification index value rvi by the determination area Dd is not performed. To. As described above, the block connection area Di once the identification index value rvi is determined is prevented from being overlapped as much as possible, so that the amount of calculation can be reduced and the processing speed can be increased.

  Hereinafter, a technique for separating the object region from the background region using the identification index value rvi obtained for each block connection region Di will be described. In order to separate the object region from the background region, a threshold value Th for the identification index value rvi is set, and the magnitudes of the identification index value rvi and the threshold value Th are compared.

  In order to set the threshold Th, first, an average value arv of the identification index value rvi in the entire image is obtained by Equation 2. This average value arv is a measure of contrast in the entire image. In Equation 2, k is the number of block connection regions Di in the image.

The threshold value Th that separates the object region and the background region is set using the following equation based on the average value arv obtained by Equation 2. The coefficient α in the following expression is appropriately set depending on the density value relationship between the object area and the background area. That is, the user can input the coefficient α as appropriate. However, 0.2 is set as the specified value of the coefficient α.
Th = arv + α · (Imax−arv)
Here, | I | max uses the maximum value of the identification index value rvi when the background area is brighter than the object area (the density value is higher), and the identification index value when the background area is darker than the object area. The minimum value of rvi is used. The above formula uses a value obtained by adding α times the difference between | I | max of the identification index value rvi and the average value arv to the average value arv as the threshold Th, and if the coefficient α is 0.2, The threshold value Th is set slightly closer to | I | max than the average value arv.

  If the identification index value rvi is separated based on the magnitude relationship with the threshold Th obtained by the above equation, the object region and the background region can be easily separated. That is, when the background area is brighter than the object area, the block connection area Di whose identification index value rvi is larger than the threshold value Th is the object area, and when the background area is darker than the object area, the identification index value rvi is the threshold value. A block connection area Di smaller than Th becomes an object area. In an actual image, a place where the background area is light and dark may be mixed with respect to the object area. In this case, the threshold Th is set on both sides of the average value arv. That is, three steps of a range between two threshold values Th that determine the upper and lower limits of the object region, a range that exceeds the upper limit threshold Th and reaches a maximum value, and a range that falls below the lower limit threshold Th and reaches a minimum value. Can be divided into

  By the way, in general, the density value of the background varies depending on the lighting and the positional relationship of the TV camera. In particular, when the imaging target is illuminated from one direction, a substantially constant gradient occurs in the background density value in a specific direction in the image, and the density value gradually increases or decreases. In the processing procedure described above, the entire image is a processing target. However, when a gradient occurs in the density value, the average value arv of the identification index value rvi obtained by Equation 2 is the object region, the background region, May be an intermediate value with respect to the background, and even if the threshold value Th is set by such an average value arv, it is difficult to accurately separate the background region and the object region.

  Therefore, as shown in FIG. 9, if the image is divided into a plurality of relatively small inspection areas Fi (i = 1 to 9 in the illustrated example) and the above-described processing is performed for each inspection area Fi, The influence of the density value change in the background can be reduced, and the background area and the object area can be accurately separated. The inspection area Fi is set by dividing the target area Ti including the target object area into a plurality of parts in the image. Each inspection area Fi is set to have the same size according to the size of a known inspection object to be extracted as an object area. For example, if the inspection target is a character, the size of the inspection area Fi is set to be about the size of the character. When a density gradient is generated in the background, the inspection areas Fi are arranged along the direction of the density gradient. Alternatively, if the inspection target is a character string, the target area Ti is set so as to include the entire character string, and the inspection area Fi is arranged in the direction in which the character strings are arranged.

  As described above, since the inspection area Fi is set by dividing the target area Ti, one inspection target may be divided into a plurality of inspection areas Fi. Therefore, an inspection region Fj (j = 11 to 19 in the illustrated example) is set by shifting the inspection region Fi set as shown in FIG. 9 by the half width of the inspection region Fi in the arrangement direction of the inspection region Fi. (See FIG. 10). If this inspection area Fj is set, the inspection area Fi and the inspection area Fj partially overlap each other, so that the probability that most of one inspection object is included in any of the inspection areas Fi and Fj increases. It becomes possible to accurately separate the region and the object region.

  In order to determine which of the inspection areas Fi and Fj set as described above contains most of the inspection object, the following processing procedure is used. That is, as shown in FIG. 11A, after obtaining the object region Eb in the inspection region Fi, the object region Eb is separated into the object regions Eb that can be regarded as a group in the target region Ti. The object region Eb that can be regarded as a single unit can be extracted by extracting a region with high similarity by pattern matching between the known information and the object region Eb when using known information such as a character string stamped on a product. If it cannot be used, the interval between the extracted object regions Eb is used.

  When the object region Eb that can be regarded as a unit is extracted, the center of gravity position Jn (n = 1, 2,...) Is obtained for each unit, and the distance between the center positions Ci and Cj of the inspection regions Fi and Fj and the center of gravity position Jn. Ask for. The calculation of the distance is not performed for all the inspection areas Fi and Fj, but only for the inspection areas Fi and Fj including the object area Eb that can be regarded as a unit. In the illustrated example, the object area Eb that can be regarded as the character “A” is included in the inspection areas F1, F2, and F11. Therefore, the center position C1 of the object area Eb and the center positions C1, C1 of the inspection areas F1, F2, and F11 are included. The distances from C2 and C11 are obtained. In the inspection area (F11 in the illustrated example) in which the obtained distance is the minimum, there is a high possibility that most of one inspection object is included. Therefore, in this inspection area F11, the object area and the background area are separated. If it is performed again, the object region and the background region can be accurately separated.

  For example, in the example shown in FIG. 11A, the object areas Eb corresponding to the characters “A” and “P” are to be inspected, but in the inspection area Fi, the boundary lines are “A” and “P”. Is set so as to be divided almost in half, and the inspection area F2 includes a part of each of the characters “A” and “P”. On the other hand, in the inspection area Fj shown in FIG. 11B, since the inspection areas F11 and F12 include almost the entire characters “A” and “P”, the objects in the inspection areas F11 and F12 are included. If the region and the background region are separated, there is a high possibility that the object region corresponding to the inspection target can be accurately separated from the background region.

It is operation | movement explanatory drawing which shows embodiment of this invention. It is a block diagram same as the above. It is operation | movement explanatory drawing which shows the outline | summary of operation | movement same as the above. It is operation | movement explanatory drawing in the state which set the block in the same as the above. It is operation | movement explanatory drawing in the state which formed the block connection area | region in the same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above. It is operation | movement explanatory drawing same as the above.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Object 2 Imaging device 3 Monitor 10 Image processing device 11 A / D converter 12 Storage means 13 Image division processing device 14 Feature amount calculation unit 15 Tracking processing unit 16 Block division processing unit

Claims (3)

After dividing the multivalued image into a plurality of rectangular blocks each including a plurality of pixels, the feature values related to the distribution of pixel values in the block are compared between adjacent pairs of blocks, and the similarity of the feature values is determined. The same label is assigned to each pair of blocks that are equal to or greater than the specified value, and a different label is assigned to each pair of blocks that have a feature value similarity less than the specified value to all the blocks in the multi-valued image. The feature value of the block adjacent to the block before dividing the block is divided into a plurality of divided blocks, and the block located at the boundary portion of the block connection area composed of the blocks to which the label is attached is assigned. When the similarity between the divided block and the feature value relating to the pixel value distribution of the divided block exceeds a specified value, the same label as the adjacent block is given to the divided block All the pixels included in the multi-valued image by repeating the process of integrating the divided block into the block connected area as a block constituting the block connected area and dividing the divided block into the block connected area. The boundary of the block connection area is searched by raster scanning , a determination area including a plurality of block connection areas is set around the boundary of the block connection area, and the boundary One block connection area of two block connection areas sandwiching the block is defined as a block connection area of interest, and the ratio of the area occupied by the other block connection area to the determination area is multiplied by the average density value of the other block connection area. a second process of obtaining a value as the density evaluation value, the average density value and the density Review of blocks connected region the noted Third step and said identification evaluation value and the focus is near to the block region the noted block coupling region block by comparing the magnitude of the specified threshold value to determine the identity evaluation value for the difference between the value and the parameter An image processing method comprising: a fourth step of separating the connection area. After dividing the multivalued image into a plurality of rectangular blocks each including a plurality of pixels, the feature values related to the distribution of pixel values in the block are compared between adjacent pairs of blocks, and the similarity of the feature values is determined. The same label is assigned to each pair of blocks that are equal to or greater than the specified value, and a different label is assigned to each pair of blocks that have a feature value similarity less than the specified value to all the blocks in the multi-valued image. The feature value of the block adjacent to the block before dividing the block is divided into a plurality of divided blocks, and the block located at the boundary portion of the block connection area composed of the blocks to which the label is attached is assigned. When the similarity between the divided block and the feature value relating to the pixel value distribution of the divided block exceeds a specified value, the same label as the adjacent block is given to the divided block All the pixels included in the multi-valued image by repeating the process of integrating the divided block into the block connected area as a block constituting the block connected area and dividing the divided block into the block connected area. The boundary of the block connection area is searched by raster scanning, and a determination area including a plurality of block connection areas is set around the boundary of the block connection area. The ratio of the area occupied by the block connection area other than the target block connection area to the judgment area among the block connection areas that are at least partially included in is determined by multiplying the average density value of each block connection area by the weight coefficient and averaging A second process of obtaining the weighted average value as a density evaluation value, and the block connection region of interest The third process of obtaining the identification evaluation value using the difference between the average density value and the density evaluation value as a parameter, and comparing the size of the identification evaluation value with a specified threshold value, and the block connection region of interest and the focus of attention. fourth images processing how to; and a step of separating the block coupling region near to the block area. In the second process, evaluation image data is generated with the obtained density evaluation value as a pixel value of the block connection area to which attention is paid, and the density evaluation value is not obtained again for the block connection area for which the density evaluation value has been obtained once. image processing how according to claim 1 or claim 2, characterized in that.

Images