JP4492258B2 - Character and figure recognition and inspection methods - Google Patents

Description

  The present invention mainly reads dot characters and figures printed or engraved on an object such as an industrial product (for example, processed food) for the purpose of product management or the like with an image input device such as a TV camera or a scanner. The present invention relates to a character / graphic recognition method and an inspection method for recognizing dot characters / graphics by performing image processing on an image obtained by an input device.

  Conventionally, a character / graphic (“character / graphic” means a character or graphic that can be generated by a character generator, and hereinafter simply referred to as “character”) is an image like a TV camera or a scanner. Various techniques for recognizing characters by reading with an input device and applying image processing to an image obtained with the image input device have been proposed. Since the technology for recognizing characters is a pattern recognition technology that associates with categories specified in advance, description of the structure of characters that makes it easy to associate categories, correction for deformation of characters, written on the object A technique such as segmentation that cuts out characters one by one is required.

  For example, in order to recognize a complicatedly deformed character such as a handwritten character, it has been proposed to describe the character structure in the following form (for example, refer to Patent Document 1 and Patent Document 2). In the techniques described in Patent Documents 1 and 2, a character image is thinned and a branch point or an intersection is extracted as a singular point, and the structure around the singular point is an intersection such as “X”, such as “K”. By dividing into three types of contact and branch such as “T”, a line connected at a singular point is divided into a plurality of strokes, and a primitive series that is a connected structure of monotone curves called primitives is used. Each stroke is represented, and a character structure is described by a combination of a primitive sequence and a singular point. A stroke corresponds to a part of a hand-written character (which almost coincides with a part written with a single stroke), and a primitive means a monotone curve in which x and y coordinates change monotonically in a two-dimensional image. Yes. The primitive connection structure is expressed by the convex direction formed by connecting two primitives and the arrangement order of primitives around the connection point in the right-handed system. It becomes a set of linked structure of the primitives that make up.

  In Patent Documents 1 and 2, a bounding box (the smallest upright rectangle surrounding the character) is used to correct the deformation of the character, and the longer side is set to the unit length without changing the aspect ratio of the bounding box. A technique is also disclosed in which the scale is normalized and correction is further performed by affine transformation.

  That is, in the techniques described in Patent Documents 1 and 2, after segmentation of characters is performed on the input image, the thinned line is analyzed to analyze the structure represented by the primitive sequence and the singular point, and then cut out from the input image Category candidates are extracted by matching the structure of the image and the model, and the partial images are normalized by calculating the distance to the model and evaluating the similarity, thereby associating the partial images with the model character categories. ing. Thus, robust pattern recognition is possible by using the structural information and quantitative information of characters.

  By the way, although the segmentation technique is not specifically described in Patent Documents 1 and 2, as a segmentation technique, a skeleton line figure obtained by thinning an input character string image is converted to a simple arc or a simple closed curve. After dividing into partial curves, a technique for estimating the possibility of touching a character using the presence or absence of a branch point or an intersection in the partial curve, the horizontal width of the partial curve, and the height of the character string has been proposed (for example, And Patent Document 3). Further, Patent Document 3 recognizes a character using a technique similar to Patent Documents 1 and 2 for a character string generated by combining partial curves obtained by dividing a skeleton line figure, and results of recognition. There is also disclosed a technique of forming a network having a partial figure recognized as a character among the combinations of nodes and outputting a combination of recognized characters of the node by selecting an optimum route from the network. Further, it is described that when the number of characters is known, an optimum route is selected from a route having the number of nodes that matches the number of characters from the network.

  By the way, in the techniques described in Patent Documents 1 and 2 described above, although the category of the input image is narrowed down using the structural information of characters, the structural information is used on the basis of branch points and intersections of line figures. It is assumed that the lines constituting the characters are continuous. The same applies to the technique described in Patent Document 3.

  Therefore, even for characters with relatively little deformation, such as characters printed or engraved on the object, dot characters widely used for displaying the expiration date or expiry date of food, defects in the printing device For characters that are faint due to the above (hereinafter referred to as “faint characters”), characters that are discontinuously separated from the lines that constitute the characters (hereinafter referred to as “divided characters”), etc. Information cannot be extracted accurately, and even if the techniques described in Patent Documents 1 to 3 are adopted, there is a possibility that characters cannot be recognized accurately.

In order to deal with this type of problem, the present inventor extracts the skeleton lines of characters and figures from the input image, approximates them with approximate lines that are simple figures, and approximate lines that are close to each other from these approximate lines. Has previously been proposed (Japanese Patent Application No. 2003-4258533), in which a set of characters is extracted as a target cluster and collated with a master cluster in which character and figure standards are represented in advance as a set of approximate lines similar to the target cluster. In this technology, when the target cluster and the master cluster are collated, the circumscribed rectangles of the target cluster and the master cluster are extracted in advance, and after aligning the circumscribed rectangles of both, The degree of proximity of the approximate lines included in both is evaluated as the shape similarity, and the master cluster having the highest shape similarity is associated with the target cluster category.
JP-A-7-21317 Japanese Patent No. 3183949 JP-A-6-76114

  As described above, in the technique disclosed in Japanese Patent Application No. 2003-4258533, even if a dot character, a blurred character, or a divided character is detected by collating with a master cluster represented by a line figure instead of using a character structure. Correct recognition is possible. However, when the dot spacing is wide compared to the dot size, such as a dot character consisting of extremely small dots, the target cluster has fewer features that can be matched with the approximate line included in the master cluster, and the shape similarity May decrease and the recognition rate may decrease.

  The present invention has been made in view of the above reasons, and its purpose is to recognize even dot characters / figures, blurred characters / figures, divided characters / figures, etc. To provide a character / graphic recognition method capable of accurately recognizing dot characters / graphics having a wide interval and a character / graphic inspection method capable of inspecting the legibility of characters / graphics. .

  The inventions of claims 1 to 6 relate to a method for recognizing characters and figures.

  The invention of claim 1 extracts a skeleton line of a character / figure from a character / figure image input as a recognition target, and then approximates the skeleton line as a set of approximate lines that are simple shapes, and continuously In addition, an approximate line in the vicinity of the approximate line group is added to the approximate line group by evaluating the distance between each pair of approximate lines that are not continuous. A target data extraction process that extracts the approximate line group treated as a unit as a target cluster, and a dot character that consists of dots arranged at arrangement points along the shape of the character / figure in a matrix consisting of multiple arrangement points vertically and horizontally A matching degree calculation process for evaluating the degree of coincidence between the dot character / figure master and the target cluster consisting of the dot arrangement in the figure and the size of the circumscribed rectangle of the dot character / figure. In the matching degree calculation process, after aligning the circumscribed rectangle of the dot character / figure master with the circumscribed rectangle set for the target cluster, a region of interest is set based on each arrangement point of the dot character / figure master. In addition, the degree of dot matching in each target area is calculated by calculating the degree of dot matching in the target cluster for each target area based on the positional relationship between the approximate line constituting the target cluster and each target area. The cluster matching degree is expressed as a numerical value of the similarity to the dot array contained in the character / graphic master, and when the cluster matching degree exceeds the threshold, the target cluster is recognized as the character / graphic represented by the dot character / graphic master. It is characterized by doing.

  According to this method, as master data (a type of template) used for collation with a character / graphic image to be recognized, a dot character / graphic master composed of an array of dots is used. Evaluate the degree of overlap between the attention area set in the vicinity of the placement point that can be placed and the approximate line belonging to the target cluster. By using the target cluster, dot characters / figures, blurred characters / figures, split characters・ It becomes possible to recognize even figures. In addition, since dot characters / graphics are collated with an ideal arrangement of dots, the dot characters / graphics can be recognized regardless of the density of the dot spacing. That is, dot characters and figures can be recognized even when the dot interval is wider than the dot size.

  In the invention of claim 2, in the invention of claim 1, the cluster matching degree uses a cross-correlation value normalized between the dot matching degree in each region of interest and the arrangement of dots included in the dot character / graphic master. It is characterized by.

  According to this method, by using the normalized cross-correlation value between the dot matching degree and the dot arrangement included in the dot character / figure master as the cluster matching degree, when there is no dot that should be present, If there is noise in a place that should not exist, the cluster matching degree will decrease, and the degree of similarity between the dot character / graphic master and the target cluster can be easily evaluated by using the cluster matching degree. . In addition, since the cross-correlation value is normalized, the cross-correlation value is robust against changes in the size of dots due to blurring of printed dot characters / graphics.

  According to a third aspect of the present invention, in the first or second aspect of the present invention, the dot arrangement in the dot character / graphic master is determined by the dot line in the dot character / graphic and the extension direction of the virtual line connecting the dots. It is composed of the direction value at the position of each reflected dot, and in the matching degree calculation process, the corresponding point of the target cluster existing within the threshold defined by the distance from each dot in the direction intersecting the extension direction of the virtual line is calculated. After the search and corresponding points are detected, the dot character / figure master is transformed by affine transformation so that the degree of matching between each dot of the dot character / figure master and the corresponding point is increased, and then the dot matching degree is obtained. It is characterized by that.

  According to this method, the alignment error between the target cluster and the dot character / graphic master can be corrected, so that even if the dot character / graphic is misaligned, rotated, or changed in scale, it can be accurately recognized. Can do.

  According to a fourth aspect of the present invention, in the first to third aspects of the present invention, the character / graphic image is a grayscale image, the lightness is within a specified range, and the spatial secondary differential value is equal to or greater than a specified threshold value. This pixel is a target for extracting the skeleton line.

  According to this method, since the spatial quadratic differentiation method that obtains a large value for a linear or dot pattern and the binarization process that divides the area by lightness are used together, an edge whose lightness changes abruptly In addition, it is possible to extract only the characteristics of the lines or dots constituting the character / graphic to be recognized by removing the influence of the brightness unevenness in which the brightness changes smoothly.

  In the invention of claim 5, in the inventions of claim 100 to claim 3, the character / graphic image is a grayscale image, and the lightness distribution of the area corresponding to the background is approximated by a quadratic curved surface. Of the lightness of each pixel in the image, the pixel whose difference from the lightness represented by the quadratic curved surface is equal to or greater than a predetermined threshold is the target for extracting the skeleton line.

  This method approximates the brightness distribution obtained from the background with a quadratic surface when the background has relatively smooth and simple brightness unevenness, and the brightness in the character / graphic area is calculated from the brightness distribution. Since binarization is performed based on the difference from the predicted brightness, the influence of uneven brightness can be eliminated during binarization. Further, the least square method can be used to obtain the quadratic curved surface, and the number of operations is less than that of the spatial quadratic differentiation, so that the processing for removing unevenness of brightness can be performed with a small amount of calculation.

  According to a sixth aspect of the present invention, in the first to third aspects of the present invention, the character / graphic image is a grayscale image, and when selecting pixels for extracting the skeleton line by binarizing the brightness, Set the brightness range so that the area occupied by the pixels in the brightness range including the brightness of the characters and figures in the image of the letters and figures is equal to or greater than the known area occupied by the letters and figures. The pixels within the brightness range are targeted for extracting the skeleton line.

  According to this method, when the character / graphic to be recognized is known, the cumulative frequency value of the brightness histogram is referred to based on the area occupied by the character / graphic to be recognized in the screen. Since the threshold value for the digitization is set, it is possible to accurately extract the character / graphic area by setting an appropriate threshold value even when the area of the character / graphic object to be recognized in the image is small. .

  The inventions of claims 7 to 10 relate to a method for inspecting characters and figures.

  According to a seventh aspect of the present invention, there is provided a dot character / graphic master when the cluster collation degree is equal to or greater than a threshold in the collation degree calculation process in the character / graphic recognition method according to any one of the first to sixth aspects. Using the dot matching degree obtained at the position and the dot arrangement in the dot character / graphic master, it is determined whether the dot presence / absence matches the character / graphic image for each placement point of the dot character / graphic master. And having a legibility evaluation process for determining the quality of a character / figure that is a recognition target based on the number of disagreement arrangement points.

  According to this method, it is possible to evaluate the degree of difference between a character / figure to be recognized and a dot character / figure master in dot units, and even if the recognition target is a character / figure in which dots are arranged, it is legible. Can be evaluated.

  In the invention of claim 8, in the invention of claim 7, in the legibility evaluation process, in the dot character / figure master, the arrangement point where the presence / absence of dots in the dot character / figure master does not match the character / figure image is determined. A dot missing point when the dot is located, a noise point when the dot is not located, and a legibility when one of the number of dot missing points or noise points exceeds the specified threshold Is judged to be defective.

  According to this method, it is possible to detect missing dots and noise in a character / graphic to be recognized as defects.

  In the invention of claim 9, in the invention of claim 7 or claim 8, the dot character / figure master has pre-registered arrangement points of important parts related to legibility, and the legibility evaluation process comprises -When the arrangement point where the presence or absence of dots in the graphic master does not match the character / graphic image is an important part, it is determined that the legibility is poor.

  According to this method, because missing parts and noise are judged as important parts of characters / graphics that are likely to be misidentified with other characters / graphics, it is not only possible to judge legibility but also misidentification of characters / graphics. Can be reduced.

  In the invention of claim 10, in the inventions of claims 7 to 9, in the legibility evaluation process, there are a plurality of characters / figures to be recognized and characters composed of a series of dots. The number of placement points when the placement point where the presence or absence of a dot in the dot character / graphic master does not match the character / graphic image is the placement point where the dot exists in the dot character / graphic master.・ For figures and dot characters and figure masters, find each line of dots aligned along the direction in which the characters and figures to be recognized are arranged, and dot the number of arrangement points obtained from the characters and figures to be recognized. When the result of dividing by the number of placement points obtained for the character / graphic master is less than or equal to the specified threshold, the dot for which the number of characters / figures to be recognized was obtained Characterized by determining that a row is missing.

  According to this method, when a character / graphic is formed by an array of dots, it is possible to detect so-called line omission where a single line of dots is lost due to an abnormality in a device that prints or stamps the character / graphic, It is possible to automatically and quickly detect an abnormality in a device for printing or engraving characters / graphics.

  According to the character / figure recognition method of the present invention, a dot character / figure master consisting of an array of dots is used as master data used for collation with a character / figure image to be recognized. Evaluate the degree of overlap between the attention area set in the vicinity of the placement point where dots can be placed and the approximate line belonging to the target cluster. By using the target cluster, dot characters, fading characters, split characters, etc. There is an advantage that it becomes possible to recognize. In addition, since dot characters and figures are matched with the ideal arrangement of dots, dot characters and figures can be recognized regardless of the density of the dot spacing, and the dot size compared to the dot size. There is an advantage that dot characters and figures can be recognized even when the interval is wide.

  According to the character / figure inspection method of the present invention, the degree of difference between a character / figure to be recognized and a dot character / figure master can be evaluated in dot units. Even if it is a figure, there is an advantage that legibility can be evaluated.

  An example of an apparatus used in this embodiment is shown in FIG. In the illustrated example, a TV camera 12 is provided as an image input device for inputting an image of a sheet-like or three-dimensional object 11. Characters (meaning characters / graphics as in the conventional configuration) are printed or stamped on the object 11 (the surface of the object 11 is depressed or raised in an embossed shape), and the TV camera 12 captures an image including the characters. To do. The present embodiment mainly recognizes characters by printing or engraving, and does not assume recognition of handwritten characters. An analog signal image captured by the TV camera 12 is input to the digital image generation device 13 and converted into a digital signal. The digital image generation device 13 generates a grayscale image whose pixel value is lightness. The grayscale image is stored in the storage device 14, and the grayscale image stored in the storage device 14 is subjected to image processing described below in the image processing device 15. The storage device 14 includes a semiconductor memory and a hard disk device. The storage device 14 stores a dot character master necessary for character recognition in addition to an image obtained by imaging the object 11. The following description will be made on the assumption that characters are printed on the object 11, but the same technique can be adopted even when characters are stamped on the object 11.

  By the way, the dot character master is a kind of template for collating with characters printed on the object 11, and, as shown in FIG. 3, m vertical × n horizontal (in this embodiment, 7 × A matrix composed of five (5) arrangement points qi (i = 1, 2,..., 35) is set as one character unit, and dots di (i = 1, 1) are arranged at arrangement points qi along the character shape within the character unit. 2, ...) are arranged. In general, when a person recognizes an arrangement of dots as characters, a virtual line ri (i = 1, 2,...) Is complemented between a pair of adjacent dots, and the virtual line ri is connected along the character shape. Thus, it is considered that the arrangement of dots di is recognized as a character composed of a series of virtual lines ri.

  Therefore, as a dot character master, a position value (a value indicating a direction indicated by a double-ended arrow in FIG. 3) that can indicate the position of the dot di within the character unit and the extending direction of the virtual line ri connecting the dots di. Data using a circumscribed rectangle surrounding the character unit is used. In the dot character master, a value different from the arrangement point qi where the dot di is not arranged is assigned to the arrangement point qi where the dot di is arranged. For example, “1” is assigned to the placement point qi where the dot di is placed, and “0” is assigned to the placement point qi where the dot di is not placed. The direction value is the direction perpendicular to the virtual line ri when only one virtual line ri is connected to the dot di, and the angle formed by both virtual lines ri when two virtual lines ri are connected to the dot di. Is the direction of bisecting. Therefore, the direction value has information corresponding to the normal direction at the position of the dot dot of the character represented by connecting the virtual line ri, and two virtual lines ri are connected to one dot di. When using the direction value, the data amount can be reduced as compared with the case where each virtual line ri is individually described. When the character expands and contracts, it is considered that the position of the dot di moves in the direction represented by the direction value. Therefore, when the size of the character in the image obtained from the object 11 and the dot character master are matched. The dot di may be displaced in the direction represented by the direction value.

FIG. 3 shows an example of a dot character master representing the character “2”, and represents the arrangement of the dots di by combining the presence / absence of the dots di at each placement point qi and the direction value for each dot di, The position and size of the circumscribed rectangle rc are also used as a dot character master in combination with the dot di array. In Figure 3, for example no dots in the arrangement point _{q 1,} the constellation points _{q 2} exist dot _{d 2,} the direction value of the dot _{d 2} is in the direction of 112.5 degrees with respect to the horizontal direction.

  The basic processing procedure of the image processing performed in the image processing device 15 is as shown in FIG. 1. First, in the target data extraction process (S1), the object 11 for character recognition is imaged including characters. From this image, the attribute of the character to be compared with the dot character master is extracted as target data. In the conventional technique, segmentation is performed as a pre-processing for extracting target data, and each character is segmented by segmentation. However, in this embodiment, segmentation is not performed in the target data extraction process, and dot characters, divided characters, By handling characters including a continuous part as a whole in the target data extraction process, division into units smaller than one character is suppressed. That is, if the distance between the separated parts is within a prescribed threshold value, the separated parts are handled as one unit. Such units are hereinafter referred to as clusters. A cluster extracted from the target data is called a target cluster, and the target cluster usually includes one or more characters, and may include a plurality of characters.

  When target data divided into target clusters is obtained in the target data extraction process, a matching degree calculation process (S2) for matching the dot character master with the target cluster is performed. In the collation degree calculation process, the dot character master position is aligned with the target cluster, the dot collation degree indicating the presence or absence of the dot di corresponding to each arrangement point qi of the dot character master is obtained, and further based on the dot collation degree array. Thus, a cluster matching degree representing the similarity between the target cluster and the entire dot character master is calculated. In other words, in the matching degree calculation process, by evaluating the degree of matching between the dot character master and the target cluster, it is possible to determine which character of the dot character master corresponds to the character represented by the target cluster. Recognize

  Further, after the character recognition, the dot collation degree is obtained again at the collation position with the position where the cluster collation degree obtained in the collation degree calculation process is maximized as the collation position, and the dot position and the object 11 in the dot character master are obtained. Interpretation for verifying consistency with the image of the character obtained from the above, determining whether the character printed on the object 11 is legible, and finally determining whether the character maintained on the object 11 is good or bad An evaluation process (S3) is performed.

  Hereinafter, the processing procedure will be specifically described for the target data extraction process, the matching degree calculation process, and the legibility evaluation process.

  The processing procedure of the target extraction process is shown in FIG. In the target data extraction process, the input grayscale image is binarized (S11), and skeleton lines are extracted from the binary image (S12). For extraction of the skeleton line, a well-known technique such as the Holditch method may be used, but it is necessary to adopt a technique that leaves no extra features other than characters in the binarization.

  In order to explain the problem at the time of binarization, a gray-scale image with uneven brightness and noise is shown in FIG. 5A, and the change in brightness on the line AA ′ in FIG. Shown in b). In FIG. 5A, the lightness of the character “R” is lower than that of the background, and there is only one portion where the AA ′ line intersects the character. Therefore, in FIG. The brightness should be lowered only at the location, but in actuality, due to the unevenness of brightness (in FIG. 5 (a), the difference in brightness is represented by the difference in the hatched portion) and the presence of noise ns The brightness is low at the location. That is, when binarization is performed with the threshold value Th1 that is a constant value shown in FIG. 5B, as shown in FIG. 5C, portions other than characters remain in the binary image. Although FIG. 5 is not a dot character formed by arranging dots, since it is necessary to evaluate the dot arrangement if the dot character is to be recognized, if the noise ns remains in the binary image, the dot character is changed. This makes it difficult to distinguish between the constituent dots and obstructs the recognition of dot characters. Therefore, any of the three types of techniques described below is employed so that no features other than characters remain in the binarization. In the example described below, it is assumed that the brightness of the character area is lower than the background. However, when the brightness of the character area is higher than the background, the light area and the dark area with respect to the lightness of the character area. The same technique can be applied even when and exist in the background.

  As a first method of binarization, there is a method of extracting pixels that are extracted by binarizing lightness with a threshold value and whose spatial second-order differential value obtained from a grayscale image is equal to or greater than the threshold value. . In order to obtain the spatial second-order differential value, a striking filter such as a Laplacian-Gaussian filter may be applied to the grayscale image. FIG. 6A is a spatial secondary differential image in which a Laplacian-Gaussian filter is applied to the grayscale image shown in FIG. 5A and the pixel value of each pixel is a spatial secondary differential value. FIG. 6B shows a change in the spatial second derivative value on the line corresponding to the AA ′ line in FIG. 5A in the spatial second derivative image in FIG. 6A. In the spatial second-order differential image, a linear feature having a specific thickness can be extracted. Therefore, on the line A-A ′ intersecting with the character, the space second at the portion corresponding to the character as shown in FIG. Whereas the second derivative value becomes larger, the spatial second derivative value becomes smaller at the part where the brightness unevenness or noise occurs. That is, by setting an appropriate threshold value Th2 for the spatial second-order differential value, it is possible to separate a part considered to correspond to a character from a lightness unevenness or a noise part.

  However, since binarization of the spatial second-order differential value by the threshold Th2 may include a part other than the character, in order to extract only the character, the threshold is applied to a normal gray image. In addition, it is desirable to employ pixels extracted from both the binary image and the spatial second-order differential image as the pixels of the character portion. By such processing, only the region corresponding to the character can be extracted as shown in FIG. As the threshold Th2 for the spatial secondary differential value, an average value of the spatial secondary differential values in the pixels whose lightness is equal to or less than the threshold Th1 is adopted, and in the spatial secondary differential image, pixels whose spatial secondary differential value is equal to or greater than the threshold Th2 are used. A white pixel (background) may be used.

As a second method of binarization, a background that is not a character is approximated by a quadratic expression, and a pixel whose difference from the approximated quadratic expression is equal to or greater than a threshold is extracted as a character to remove unevenness in brightness. There is. That is, first, a region corresponding to a character is roughly cut out by binarizing the lightness with a threshold value, and the relationship between the lightness and coordinates of the pixel that has become the background by binarization is approximated by a quadratic expression. That is, the lightness distribution is approximated by a quadric surface. As a quadratic expression to be approximated, it is desirable to use a quadratic expression representing a quadratic hyperboloid such as the following expression, but other quadratic expressions can also be used.
b (x, y) = a1 · x ² + a2 · y ² + a3 · x + a4 · y + a5
However, b (x, y) is the brightness at the coordinates (x, y), and a1 to a5 are coefficients. If the coefficients a1 to a5 are determined using the above equation as a regression equation, an approximate expression of the background brightness b (x, y) excluding characters can be obtained. If the difference between the lightness b (x, y) obtained from this approximate expression and the actual lightness is obtained, and pixels whose difference is equal to or greater than a predetermined threshold are extracted as pixels corresponding to the character, the character and the background are separated. Can be binarized. This technique is suitable for removing the effect of uneven brightness, but when noise with a large difference in brightness from the approximate expression occurs or when there is a pattern in the background that cannot be expressed by a quadratic expression. The background cannot be removed, but the amount of calculation is smaller compared to the case of obtaining the spatial second-order differential value. Therefore, when smooth brightness unevenness that can be approximated by a quadratic expression occurs, high-speed computation can be expected with a small amount of computation.

  The third method of binarization is a method of determining a threshold for binarization using the number of pixels in an area corresponding to a character. Now, when the area corresponding to the character is small with respect to the size of the image as shown in FIG. 7A, the brightness histogram of the area corresponding to the character as shown in FIG. The frequency near the peak p1) is significantly less than the brightness level of the background region. In general, when a threshold value for binarization is determined using a histogram, lightness between two peaks having a high frequency among peaks appearing in the histogram is often used as the threshold value. However, in the histogram as shown in FIG. 7B, four peaks p1 to p4 are generated, and it is not possible to automatically determine which peak p1 to p4 is preferably used as the threshold value. In particular, in the example shown in FIG. 7, since the number of pixels in the region corresponding to the character is small, the frequency of the peak p1 that is considered to be the region corresponding to the character is small, and the lightness between the two peaks having the high frequency is used as the threshold value. An appropriate threshold cannot be determined.

  By the way, since the attribute of the character is known at the time of recognition or inspection of the character printed or stamped for the purpose of product management, the area occupied by the character in the image can be estimated. The threshold value is set so that the area (number of pixels) of the following region is equal to or larger than the estimated area. However, if the character is a faint character or there is noise, adopting a threshold value (assuming 9A in FIG. 7) that obtains the estimated area will cause noise in the region corresponding to the character. The number of pixels in the corresponding area is relatively large, and there may be cases where it is not possible to obtain a sufficient number of pixels to extract character features. Therefore, a larger number of pixels than the estimated area is extracted. Set the threshold value to be slightly higher. For example, a threshold (9C is assumed in FIG. 7) that is larger than the threshold 9A at which the estimated area is obtained, and the rate of change of the frequency with respect to the brightness is greater than or equal to a specified value is estimated and estimated. An appropriate threshold value 9B between the threshold value 9A and the threshold value 9C from which the obtained area is obtained is used for binarization. The threshold value 9B can be appropriately set between the two threshold values 9A and 9C. For example, the threshold value 9B is set so that an average number of pixels of the number of pixels equal to or less than the threshold value 9A and the number of pixels equal to or less than the threshold value 9C is obtained. Set. In this technique, it may be difficult to determine the threshold when the background includes many regions having a small brightness difference from the character, but the threshold is determined by estimating the area of the character. Even if there is a change in lightness due to the lighting condition or the like, an appropriate threshold value can be set, and the amount of calculation is reduced compared with the case of obtaining a spatial second derivative value.

  When a skeleton line image obtained by extracting a skeleton line from a binary image obtained by using an appropriate binarization technique according to the object 11 among the three types of binarization techniques described above, The line element is extracted, and the line element is approximated to an approximate line having a simple shape such as a straight line or an elliptical arc. A node is represented by the coordinates of the node and the node name. This process will be described below with a specific example.

  Here, a case where the skeleton line of the character “R” is extracted from the skeleton line image as shown in FIG. 8A will be described as an example. In this case, when a binary skeletal line image is raster scanned and end points and branch points (including intersections) are extracted from the skeletal line as nodes (S13 in FIG. 4), the end point t1 as shown in FIG. 8B. , T2 and branch points j1, j2 are extracted. After the nodes are extracted, line elements connecting the nodes are extracted (S14). That is, the lines are traced from the end points t1 and t2 and the branch points j1 and j2 to the other end points t1 and t2 or the other branch points j1 and j2 (S14b). e4 is extracted and recorded (S14c). When one end of the line elements e1 to e4 is open, the tracking path to reach the other end points t1 and t2 or the branch points j1 and j2 is one line element (for example, line elements e3 and e4). When both ends are closed, the tracking path to reach the end points t1 and t2 or the branch points j1 and j2 that are the start points is set as one line element (for example, line elements e1 and e2). If the node that has finished tracking is the end points t1 and t2, the tracking is terminated, and the tracking is performed from the end points t1 and t2 that are not being tracked. Further tracking is performed from the branch points t1 and t2 (S14a, S14d).

  When the line elements e1 to e4 are extracted as described above, the line elements e1 to e4 are approximated as a set of “approximate lines” which are simple shapes (geometric shapes such as straight lines and ellipses). Here, an approximate line represented by a simple mathematical expression is used as a simple shape. Two types of approximate lines are used: a straight line and a part of an ellipse (hereinafter referred to as an “elliptical arc”). By using straight lines and elliptical arcs as approximate lines in this way, parameters for expressing approximate lines can be reduced as compared to the case of using Bezier curves and spline curves, and the processing load for various operations is also reduced. . In the shape shown in FIG. 8B, the line element e1 is approximated by two straight lines forming the upper left corner of the character “R” and one elliptical arc from the right end of the upper straight line to the branch point j2. can do. That is, as shown in FIG. 8C, the line element e1 is approximated by the approximate lines 11 and 12 that are two straight lines and the approximate line l3 that is one elliptical arc, and two approximate lines 11 that are straight lines. , L2 and a line connection point d1 connected to one approximate line l2 that is a straight line and the other approximate line l3 that is an elliptical arc are introduced as nodes. Since the line elements e2 to e4 are straight lines, they are used as they are for the approximate lines l4 to l6. In short, a line element is approximated by an approximate line that can be expressed by a mathematical expression, and corner points and line connection points are extracted (S15 in FIG. 4).

  The process of extracting skeletal line nodes and line elements from the skeleton line image is performed for the entire surface of the image or a specific range specified in the image, and after extracting the nodes and approximate lines, the nodes in the range handled as one unit In addition, node data and approximate line data related to the approximate line are collected as a target cluster. In node data, the coordinates of the node and the approximate line name connected to the node are associated with the node name. In the approximate line data, the parameter of the line element and the node name of the node connected to the line element correspond to the approximate line name. Attached. For example, when a binary image obtained by binarizing a grayscale image is as shown in FIG. 9A, the figure represented by the target cluster is as shown in FIG. 9B.

  The process of generating the target cluster is performed according to the procedure represented by step S16 and step S17 in FIG. First, an appropriately selected node is set as a starting point node (S16b), and a node at the other end of the approximate line having the starting point node as one end is detected (S16d). The type of the node at the other end is classified into an end point, a branch point, and others (S16e). If it is an end point, the same processing is performed for another approximate line connected to the origin node (S16c). The branch point is set as a new starting node (S16g), and the same processing is performed. If the node at the other end is neither an end point nor a branch point, the node is set as a new starting node (S16f), and the type is classified for the node at the other end of the approximate line in which the node is connected to one end. When this processing is performed, all nodes connected to the first selected starting node via the approximate line can be extracted. However, any number of approximate lines may be interposed between the starting node selected first and another node. Therefore, if there is no approximate line that has not been processed (S16c), a set of node data and approximate line data including the first selected origin node and regarded as one character is generated. The set of approximate line data is temporarily stored as cluster candidates (S16h). The cluster candidate generation and storage operations are repeated until all nodes in the image are included in any cluster candidate (S16a). Through the above processing, cluster candidates pc1 to pc33 are extracted as shown in FIG. In FIG. 9C, the cluster candidates pc1 to pc33 are surrounded by a circumscribed rectangle.

  As can be seen from FIG. 9C, the cluster candidates pc1 to pc33 obtained as described above do not necessarily form a cluster for each character. Therefore, in step S17 shown in FIG. 4, processing for combining cluster candidates corresponding to one character is performed. Whether or not the cluster candidates pc1 to pc33 are to be combined is evaluated based on the distance between the nodes and approximate lines included in each pair of cluster candidates pc1 to pc33. That is, two cluster candidates are extracted from each cluster candidate pc1 to pc33, and the shortest distance between the distances between all the nodes included in the other cluster candidate and the distance between the approximate lines is obtained (S17b). When the shortest distance is equal to or less than the prescribed threshold value (S17c), it is determined that both cluster candidates should be handled as one unit, and both cluster candidates are handled as one cluster and registered as target data (S17e). If the shortest distance is larger than the prescribed threshold value, both cluster candidates are determined as different characters, and are treated as different clusters and target data is registered (S17d). For example, by setting a threshold value so that the shortest distance between nodes in the cluster candidates pc1 to pc7 in FIG. 9C is smaller than the threshold value, both cluster candidates pc1 to pc7 are changed as shown in FIG. It will be handled as one unit. However, when the distance between the node and the approximate line is short like cluster candidates pc2 and pc3, a plurality of characters may be treated as one character, and evaluation of cluster candidates pc1 to pc33 as shown in FIG. In some cases, the target clusters tc1 to tc4 obtained by the above include a plurality of characters.

  By combining the cluster candidates by the above-described processing, two or more portions where one character is adjacent, such as a character or a dot character including discontinuously separated parts such as “i” and “ha” Even in the case of being composed of, it is possible to combine them into a one-character cluster. This is the same even if it is a faint character or a divided character.

  Next, in the collation degree calculation process, the dot character master and the target data described above are aligned, and in the subsequent legibility evaluation process, the dot collation degree for determining the presence / absence of each dot and the alignment are performed. Processing for obtaining a cluster matching degree for evaluating the overall similarity of necessary character shapes is performed. The processing procedure of the matching degree calculation process for obtaining the dot matching degree and the cluster matching degree will be described in detail with reference to FIG.

  For the alignment of the target cluster and the dot character master, the positions of the circumscribed rectangles set in the target cluster and the dot character master are aligned (S22). However, since the target cluster includes target data for one character or more and may include a target for a plurality of characters, the dimensional ratio between the height dimension and the width dimension of the dot character master and the target cluster is used. Estimate the number of characters included in the target cluster, move the dot character master dm within the target cluster tc as shown in FIG. 11 according to the estimated number of characters, obtain the cluster matching degree described later at each position, Using the position where the degree is the maximum as a collation position, a dot collation degree to be described later is obtained at the collation position, and this dot collation degree is used in a legibility evaluation process to be described later.

  As described above, the dot character master dm is moved in the target cluster tc, and the position where the cluster matching degree is maximized is obtained, so that the characters in the target cluster tc including a plurality of characters are separated. Here, the cluster matching degree represents the similarity between the cluster candidate pc in the target cluster tc and the entire dot character master dm, and the fact that the cluster matching degree is high is represented by the dot character master dm. Represents a high probability of matching a character. In this way, by comparing the positions of the circumscribed rectangles of the dot character master dm and the target cluster tc to obtain the cluster matching degree, the template is shifted pixel by pixel, and compared with the case where the dot matching degree is obtained, the area to be matched Can be greatly reduced, and the processing load can be reduced.

  After the alignment between the target cluster and the dot character master, first, correction is performed so that the dot character master is deformed so as to be aligned with the dots of the target cluster (S23 to S25). That is, as shown in FIG. 12 (a), dots d1 to d14 of the dot character master are added to the nodes and line elements of the target cluster (parts where the nodes are represented by white circles and the line segments connecting the nodes are line elements). When the target cluster is detected from the position of each dot d1 to d14 in the direction indicated by the direction value (see FIG. 3) and the target cluster is detected, the position detected in the target cluster is The dots d1 to d14 are associated with each other. That is, when a corresponding point of a target cluster that is within a threshold value defined by the distance from each of the dots d1 to d14 in a direction intersecting with the extending direction of the virtual line is searched for and the corresponding point is detected in the target cluster, Are associated with dots d1 to d14. Dots that cannot detect the target cluster even if searching in the direction indicated by the direction value set for each dot d1 to d14 of the dot character master (in the example shown, d4, d10, d13, and d14 are indicated by the direction value). The node having the closest distance to the direction within the specified angle range is associated (S23).

Further, based on the positional relationship between the dots d1 to d14 in the dot character master and the corresponding points of the target cluster, parameters a1 to a6 for deforming the dot character master are obtained by the least square method based on the following equation (S24). .
X = a1 · x + a2 · y + a3
Y = a4 · x + a5 · y + a6
(X, Y) is a coordinate on the target cluster, and (x, y) is a table on the dot character master. Since this deformation equation is an affine transformation equation that compensates for deformation of the oblique angle of position, rotation, scale, and coordinates, the width of the target cluster is slightly narrower than the width of the dot character master as shown in FIG. 9B. Even if the character is deformed, the dot character master can be superimposed (S25). However, even if the affine transformation of the dot character master is performed, it is difficult to make the dot character master completely coincide with the target cluster. Therefore, as shown in FIG. The attention area ai (i = 1, 2,..., 35) centered on the arrangement point qi is set in (S26). In the illustrated example, the attention area ai is a square area having the same size centered on the position of the arrangement point qi after deformation of the dot character master, and the attention areas ai set for the adjacent arrangement points qi overlap each other. It is set to a size that does not. Note that the attention area ai may be circular instead of square.

  Next, a dot matching degree for evaluating whether or not the target cluster overlaps the attention area ai set for each arrangement point qi of the dot character master is obtained (S27), and the target cluster and the dot character master are determined based on the dot matching degree. The cluster matching degree indicating the overall similarity to is obtained (S28).

  First, a method for obtaining the dot matching degree will be described. For the dot matching degree, the length of the line element li (i = 1 to 7) of the target cluster included in each attention area ai is used. For example, since the attention area a16 overlaps with part of the line elements l2 and l5 included in the target cluster, the dot matching degree with respect to the attention area a16 is the portion of the line element l2 and l5 passing through the attention area a16. Calculated as the sum of lengths. The dot matching degree becomes larger as more line elements are present in the attention areas a1 to a35. That is, it is a numerical value indicating the presence or absence of dots in the target cluster. An example of the dot matching degree obtained for FIG. 13A is shown in FIG.

  Since the dot matching degree is obtained for each arrangement point qi constituting the dot character master, a dot matching degree matrix in which the dot matching degree is associated with each arrangement point qi for one character of the dot character master is obtained. Further, the dot matching degree of each of the attention areas a1 to a35 of the dot matching degree matrix is a measure of the degree of matching between the character shape from which the target cluster is extracted and the dot character master. Therefore, when the dot matching degree is obtained, the cluster matching degree for evaluating the similarity between the dot matching degree matrix and the dot character master is obtained (S28). In order to obtain the cluster matching degree, first, as shown in FIG. 14, “1” is given to the arrangement points qi where the dots d1 to d14 exist in the dot character master, and “0” is given to the remaining arrangement points qi. A binarized matrix is formed, a normalized cross-correlation value (cross-correlation coefficient) between the dot matching degree matrix and the binarizing matrix as shown in FIG. 13B is obtained, and this value is used as the cluster matching degree. .

  The normalized cross-correlation coefficient between the dot matching degree matrix of FIG. 13B and the binarization matrix of FIG. 14 is 0.919, which means that the similarity between them is 91.9%. To do. When the target data does not exist at the position where the dot of the dot character master exists or when the target data exists at the position where the dot does not exist in the dot character master, the cross-correlation coefficient is a small value. Therefore, by using this value as the cluster matching degree, it is possible to obtain a feature quantity that well represents the similarity of the whole character.

  Therefore, the position of the dot character master is aligned with a plurality of positions of the target cluster and the cluster matching degree is obtained, and the position where the cluster matching degree is maximized is determined as the position where the character similar to the dot character master exists in the target cluster, The dot matching degree matrix obtained at this time is stored in a buffer (not shown). Specifically, when the cluster matching degree is obtained, it is compared with the maximum value of the cluster matching degree obtained so far (S29), and the newly obtained cluster matching degree is larger than the maximum value of the cluster matching degree. When it is larger, the newly obtained cluster matching degree is replaced with the maximum value (S30). Further, the position of the dot character master at this time is set as a collation position (S31), and a dot matching degree matrix is registered in the buffer as a dot matching degree matrix for inspection (S32). The processes in steps S22 to S32 are repeated until the cluster matching degree is obtained at all positions of the target cluster (S33), and then the process proceeds to a legibility evaluation process in which the quality of printed characters is determined.

  By the way, legibility is the property of whether or not characters can be read and is based on qualitative judgment. The reason why the character cannot be read is that, for example, the character is largely missing as shown in FIG. 15 (a), noise other than the character is mixed as shown in FIG. 15 (b), or the character is added as shown in FIG. 15 (c). Possible causes such as significant deformation. In the legibility assessment process, the degree of legibility due to this type of cause can be judged on a quantitative basis.

  In the legibility evaluation process, as shown in FIG. 17, a dot missing point f1 in which there is no corresponding portion in the character printed on the object 11 with respect to the arrangement points where the dots d1 to d14 exist in the dot character master. Alternatively, a noise point f2 where noise corresponding to a dot exists in the character of the object 11 is extracted from the arrangement points where the dots d1 to d14 do not exist in the dot character master, and the dot missing point f1 and the noise point are extracted. The number of f2 is obtained (S41 to S47). That is, by applying an appropriate threshold value to the dot matching degree matrix obtained in the matching degree extraction process, a corresponding dot is printed on the character printed on the object 11 for each dot d1 to d14 of the dot character master. It is determined whether or not it exists, and if the dot matching degree is less than or equal to the threshold for the dots d1 to d14, the dot missing point f1 is set, and an arrangement point having a dot matching degree equal to or more than the threshold other than the dots d1 to d14 is referred to as the noise point f2. To do.

  After obtaining the numbers of dot missing points f1 and noise points f2 as described above, each is compared with an appropriate threshold value (S52, S53). Since both the dot missing point f1 and the noise point f2 are portions that do not coincide with the dot character master, a large number of these means low legibility. Therefore, when any one of the dot missing point f1 and the noise point f2 is abnormal in threshold value, it is evaluated that the legibility is low, and the character printed on the object 11 is determined to be defective (S57). If any number is less than the threshold, it is determined that the character printed on the object 11 is good (S56). Here, the threshold values for the number of missing dot points f1 and noise points f2 may be, for example, 2 and 3, respectively. That is, if the dot missing point f1 is within one and the noise point f2 is within two, it is determined to be good.

  For evaluation of legibility, in addition to the number of dot missing points f1 and noise points f2, the number of adjacent dot missing points f1 may be used together (S48 to S50, S54). Since the presence of the adjacent dot missing point f1 means that there are two or more dot missing points f1, when the number of adjacent dot missing points f1 is used for evaluation, the dot missing point f1. It is assumed that the threshold for the number of is three.

  The processing procedure of the legibility evaluation process shown in FIG. 16 includes determination of missing lines (S51, S55). Line missing means that when characters are printed (or engraved), dot missing points are continuously generated on a straight line due to a malfunction of an apparatus for printing or engraving characters. The cause of the line omission is that the character is moved by running a print head in which a plurality of dot-like print portions that print each dot constituting the character are arranged in a row in a direction perpendicular to the direction in which the print portions are arranged. When a printing apparatus for printing is used, missing dots continuously occur on a straight line along the traveling direction of the print head due to a defect in one of the printing units. For example, in an ink jet printing apparatus, line omission may occur due to clogging of a print head.

  When line omission occurs, depending on the shape of the character, there may be only one or more dot omission points, so it may be considered good in the legibility determination based on the number of dot omission points f1 and noise points f2. . However, when there is an abnormality in the printing apparatus such as the print head being clogged, it is necessary to deal with it at an early stage, and in particular, it is necessary to detect missing lines as a serious abnormality. Therefore, when a plurality of characters are arranged, the number of dots on each line in the direction in which the characters are arranged (generally on a straight line at each height position for horizontally written characters) is obtained in advance and printed. The line missing is determined by comparing with the number of dots on the same line obtained from the characters. That is, the number of dots on each line obtained from the printed character is divided by the number of dots on each line obtained from the dot character master, and it is determined that the line is missing when the division result is equal to or less than a prescribed threshold value. For example, the example shown in FIG. 18 is an example in which characters are printed using a print head that forms seven vertical dots. If the print head is normal, the printed characters are as shown in FIG. Shall be. The number of each side-by-side dot is shown at the right end. Here, as shown in FIG. 18 (b), if the dots in the third row from the top are lost due to an abnormality in the print head, the number of dots is 0 while 11 dots originally exist. From this, it can be determined that a line omission has occurred in the third row.

  By the way, as a dot collation degree, you may use not only the value mentioned above but any of six types of values demonstrated below.

  (1) FIG. 19 shows an expansion process performed so that the line elements included in the target cluster are expanded to a certain width, and the number of pixels occupied by the expanded line element in the attention area ai is used as the dot matching degree. Can do. When the line element is expanded and used, when the line element exists near the boundary of the attention area ai, the dot matching degree is obtained in the plurality of attention areas ai sandwiching the boundary, and the attention area ai where the dot matching degree is not zero. Therefore, the shape of the target cluster is easily reflected in the dot matching degree, and the recognition can be performed with higher stability than when the line element expansion processing is not performed.

  (2) As another dot matching degree, a probability density distribution is given to the target cluster such that the probability decreases as the distance from the approximate line constituting the target cluster increases, and the probability is integrated in each attention area ai. A value may be used for the dot matching degree of each attention area ai. FIG. 20 shows the concept of probability density distribution, where the probability is expressed in shades (the darker the color, the higher the probability). The greater the distance from the approximate line, the lower the probability, so the degree of dot matching with respect to the approximate line near the boundary of the attention area ai decreases, and rational dots corresponding to the distance from the approximate line for each attention area ai. A matching degree can be given.

  (3) The attention area ai may be superimposed on the grayscale image that is the original image, and the average brightness for each attention area ai may be used as the dot matching degree. When the dot matching degree is obtained in this way, the expansion process is not required, and it is not necessary to obtain the probability density distribution. Therefore, the dot matching degree can be easily obtained.

  (4) When a skeleton point that is not a skeleton line but a point sequence is obtained when extracting a skeleton line, as shown in FIG. 21, the skeleton points in each attention area ai (the sequence of skeleton points is Since this corresponds to the skeleton line li, the number of skeleton lines li in the figure may be used for the dot matching degree. The number of skeleton points corresponds to the length of the skeleton line li. However, when the skeleton point is obtained, the processing becomes simpler than obtaining the length of the skeleton line li.

  (5) When a skeleton point is used as the dot matching degree, a rectangle having a specified width and height centered on the skeleton point is set as shown in FIG. 22, and the rectangle overlaps with each attention area ai. It is also possible to use the total area as the dot matching degree. This method corresponds to the skeleton line expansion process, and the dot matching degree can be obtained by a simple method.

  (6) As a method for obtaining the degree of dot matching using the skeleton point, as shown in FIG. 23, a probability density distribution is obtained in which the probability decreases as the distance from the center of the skeleton point increases with respect to the skeleton point. You may use the integrated value of the probability in ai for the dot collation degree of each attention area ai. FIG. 23 shows the concept of probability density distribution, and the probability is expressed in shades (the darker the color, the higher the probability). In this method, the same effect as the case of obtaining the dot matching degree using the probability density distribution obtained for the skeleton line can be expected, and moreover, since the skeleton point is used, the processing amount is smaller than when the skeleton line is used.

  Generally, characters include feature parts that are key points for recognition by humans or devices. Therefore, a standard for feature parts (important parts) is set for each character, and the feature parts are used as dot missing points or noise points. When it becomes, it may be determined to be defective regardless of the number of missing dots or noise points.

  For example, the character “8” as shown in FIG. 24 (a) and the character “B” as shown in FIG. 24 (b) have many common points in shape, and the difference between them is the dots d21 to d21 shown in FIG. 24 (b). The presence or absence of d23. Therefore, the dots d21 to d23 are important parts related to legibility, and if the dots d21 and d22 are missing as shown in FIG. 24C, it becomes difficult to identify “8” or “B”. Misreading is likely to occur. Therefore, when the noise point occurs at the positions of the dots d21 to d23 which are important parts related to the legibility with respect to “8” of the dot character master, it is judged as defective, and the dot missing point with respect to the dot master “B”. Is determined to be defective when the dot occurs at the positions of dots d21 to d23.

  Alternatively, as shown in FIGS. 25 (a) and 25 (b), for characters “N” and “H”, the dots d24 and d25 in FIG. 25 (a) and the dots d26 and d27 in FIG. 25 (b) are for identification. It becomes the characteristic part. Therefore, when dots are detected at both the positions of the dots d24 and d25 and the positions of the dots d26 and d27 as shown in FIG. to decide.

It is operation | movement explanatory drawing of embodiment of this invention. It is a schematic block diagram of the apparatus used for the same as the above. It is a figure which shows an example of the dot character master used for the same as the above. It is operation | movement explanatory drawing of the target data extraction process 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 which shows the production | generation process of the target cluster in the same as the above. It is operation | movement explanatory drawing which shows the production | generation process of the cluster candidate in the same as the above. It is operation | movement explanatory drawing which shows the collation degree calculation process 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 which shows the legibility evaluation process 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 which shows the concept which calculates the dot collation degree in the same as the above. It is operation | movement explanatory drawing which shows the concept which calculates the dot collation degree in the same as the above. It is operation | movement explanatory drawing which shows the concept which calculates the dot collation degree in the same as the above. It is operation | movement explanatory drawing which shows the concept which calculates the dot collation degree in the same as the above. It is operation | movement explanatory drawing which shows the concept which calculates the dot collation degree 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.

Explanation of symbols

11 Object 12 TV Camera 13 Digital Image Generating Device 14 Storage Device 15 Image Processing Device

Claims (10)

  Extracts the skeleton line of the character / graphic from the image of the character / graphic input as the recognition target, then approximates the skeleton line as a set of approximate lines that are simple shapes, and approximates the continuous approximate line as one approximation In addition to the line group, by evaluating the distance between each pair of approximate lines that are not continuous, the approximate line group in the vicinity of the approximate line group is also added to the approximate line group. Target data extraction process to extract as target clusters, and dot arrangement and dots in dot characters / figures configured by arranging dots at placement points along the shape of characters / figures in a matrix composed of multiple vertical and horizontal placement points A collation degree calculation process for evaluating the degree of coincidence between the dot character / graphic master composed of the dimensions of the circumscribed rectangle of the character / graphic and the target cluster. After aligning the circumscribed rectangle of the dot character / graphic master with respect to the circumscribed rectangle set for the target cluster, the target cluster is set by setting the attention area based on the arrangement points of the dot character / graphic master. In addition to calculating the dot matching degree by quantifying the degree of dot presence / absence in the target cluster for each area of interest based on the positional relationship between the approximate line and each area of interest, the dot matching degree and dot character / graphic master in each area of interest The cluster matching degree that represents the similarity with the array of included dots is expressed as a numerical value, and when the cluster matching degree exceeds a threshold value, the target cluster is recognized as a character / figure represented by a dot character / figure master. To recognize characters and figures to be used.   2. The character / graphic recognition according to claim 1, wherein the cluster matching degree uses a normalized cross-correlation value between a dot matching degree in each region of interest and a dot arrangement included in a dot character / graphic master. Method.   The dot arrangement in the dot character / figure master is composed of the dot position in the dot character / figure and the direction value at the position of each dot reflecting the extension direction of the virtual line connecting the dots. In the process, the corresponding point of the target cluster that is within the threshold specified by the distance from each dot in the direction intersecting the extension direction of the virtual line is searched, and when the corresponding point is detected, each dot of the dot character / graphic master is detected. 3. The character / graphic figure according to claim 1 or 2, wherein the dot matching degree is obtained after the dot character / graphic master is transformed by affine transformation so that the degree of coincidence between the character and the corresponding point is high. Recognition method.   Character / graphic images are grayscale images, and pixels whose lightness is within a specified range and whose spatial second-order differential value is equal to or greater than a specified threshold are targets for extracting the skeleton line. The method for recognizing characters and figures according to any one of claims 1 to 3.   The character / graphic image is a grayscale image that approximates the lightness distribution of the area corresponding to the background with a quadric surface, and the lightness represented by the quadric surface among the lightness of each pixel in the character / graphic image 4. The character / graphic recognition method according to claim 1, wherein a pixel whose difference from a threshold value is equal to or more than a predetermined threshold is a target for extracting the skeleton line. 5.   The character / graphic image is a grayscale image, and in selecting the pixel from which the skeleton line is extracted by binarizing the lightness, the light within the lightness range including the lightness of the character / graphic is included in the character / graphic image. A brightness range is set so that the area occupied by the pixel is equal to or greater than the known area occupied by the character / graphic and within the specified range, and the pixels within the brightness range are to be extracted from the skeleton line. The character / graphic recognition method according to claim 1, wherein:   7. The dot collation obtained at the position of the dot character / graphic master when the cluster collation degree is equal to or greater than a threshold in the collation degree calculation process in the character / graphic recognition method according to claim 1. Using the degree and the arrangement of dots in the dot character / graphic master, it is determined whether or not the dot presence / absence matches the character / graphic image for each arrangement point of the dot character / graphic master. A method for inspecting characters / graphics, comprising a legibility evaluation process for determining the quality of characters / graphics to be recognized based on the number of characters.   The legibility evaluation process determines a dot missing point when an arrangement point where the presence or absence of a dot in the dot character / graphic master does not match the character / graphic image is an arrangement point where a dot exists in the dot character / graphic master. 8. The readability is judged to be poor when one of the number of dot missing points and noise points exceeds a prescribed threshold as a noise point when the dot is an arrangement point where no dot exists. Character / graphic inspection method.   In the dot character / graphic master, arrangement points of important parts related to legibility are registered in advance, and the legibility evaluation process is an arrangement in which the presence or absence of dots in the dot character / graphic master does not match the character / graphic image. 9. The character / graphic inspection method according to claim 7, wherein when the point is an important part, the legibility is judged to be poor.   In the legibility evaluation process, if there are a plurality of characters / figures to be recognized and the character is composed of a series of dots, the presence / absence of dots in the dot character / figuration master does not match the character / graphic image. The number of placement points when the placement point is a placement point where dots are present in the dot character / graphic master, the recognition target character / figure and the character / figure master to be recognized The result is obtained by dividing the number of placement points obtained from the character / figure to be recognized by the number of placement points obtained for the dot character / figure master. 8. The method according to claim 7, wherein when it is equal to or less than the threshold value, it is determined that a line of dots whose number has been obtained is missing in a character / graphic to be recognized. Inspection method characters and graphics of any one of claim 9.