JPH05342414A - Template preparing device and pattern recognizing device - Google Patents

Abstract

PURPOSE:To prepare a template which can exactly discriminate a character even when the character has degraded printing quality in the case of performing pattern recognition by performing template matching. CONSTITUTION:A prescribed area is set through a character segmenting part 14 to a reference character image, and the character image is sent to a character part template preparation part 15 and a background part template preparation part 16. At the respective preparation parts, distance conversion is respectively performed to a character part or a background part, and templates whose template value becomes larger, as they are separated farther from the edges of characters are prepared. Then, both templates are added (a code is added to the background part before the addition) by a synthesizing part 17, and the template is prepared and stored in a memory 7 for template. The prepared template can be read at need and the desired template value can be corrected based on the inputs of a cursor key 10a and ten keys 10b by a correction part 18.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognition apparatus for recognizing a pattern of an input unknown character image or the like by performing template matching, and a template creation for creating a template for performing such pattern recognition. It relates to the device.

[0002]

2. Description of the Related Art An unknown character image captured from an image pickup device is compared with a reference pattern (template) created in advance, and a character recognition device for recognizing (determining) the character by discriminating the degree of matching is conventionally used. Kaihei 3-24
There was something shown in the 0892 public information.

The character recognition device disclosed in this publication has a first mask pattern for a portion where a character should not be missing, that is, the inside of the character, and a second mask pattern for a portion which should not be projected, that is, a second mask pattern. The unknown character image data is created and compared with both patterns, and the character portion of the unknown character image and the first mask pattern are (improper)
Whether or not the character image is a character of the mask pattern is determined based on the number of matching pixels and the number of (non) matching pixels of the background portion of the unknown character image and the second mask pattern.

[0004]

However, in the above-mentioned conventional character recognition device, each pattern is "1.0".
Therefore, it is not possible to make a determination in consideration of the degree of missing or bleeding of characters (especially the depth in the thin or thick direction). That is, for example, a printed character may be partially missing or bleeding, and the periphery may be blurred.
Then, the recognition rate for such blurred characters may be lowered, and there is a possibility that the characters may be erroneously recognized or may not be discriminated. As a result, for example, if it is determined whether the characters printed on the label such as a sticker are correct, and if it is determined whether the product is a good product or a defective product, the characters that are slightly blurred due to fluctuations in the print quality (however, are recognized as characters However, there is a problem in that the correct character recognition cannot be performed with respect to a non-defective product and accurate discrimination processing cannot be performed.

The present invention has been made in view of the above background, and an object of the present invention is to enable accurate discrimination of a character even if the character has poor print quality, and to improve the recognition rate. It is an object of the present invention to provide a template creation device and a pattern recognition device for improving the quality inspection accuracy.

[0006]

In order to achieve the above-mentioned object, the template creating apparatus according to the present invention creates a first template by converting the distance to the inner part of a pattern such as a character as an input reference. First converting means for
Second conversion means for converting the distance to the background portion of the pattern to create a second template; and the first and second conversion means.
And a means for synthesizing the first and second templates created by the converting means.

Further, preferably, means for specifying a predetermined position of the template created by combining by the above-mentioned combining means, and means for correcting the template value at the predetermined position specified by the means It is to further provide.

Further, in the pattern recognition device, a template storage means for storing the template created by the template creation device, a template stored in the template storage means, and a pattern such as an input unknown character image And means for calculating the degree of coincidence, and a judgment means for receiving the output of the calculation means and making a predetermined judgment.

[0009]

In the template creating apparatus, the first conversion means is used for the reference pattern and the distance conversion is performed for the inside of the pattern. As a result, the first template is created inside the pattern such that the template value (absolute value) increases away from the edge, that is, toward the inside. Similarly, the second conversion means is used to perform distance conversion on the background portion. As a result, a second template is created in which the template value (absolute value) increases as the distance from the edge of the pattern increases to the outside. Then, by synthesizing the two templates by the synthesizing means, a template value is created in which the template value (absolute value) is small near the edge of the pattern, and the template value increases toward the inner and outer sides.

On the other hand, the pattern recognition apparatus stores the template thus created in the storage means, compares the read unknown image data with the stored template, and obtains the degree of coincidence. At this time, even if the edge periphery of the unknown image data is blurred or missing due to, for example, blurring of printing, since the template value of the edge periphery of the reference pattern is small, the pattern due to such blurring or the like is generated. Blur has little effect on the calculated degree of coincidence. However, when the pattern of the template and the input unknown image data are different, there is no pattern in the central part of the original pattern (the template value (absolute value is large)), or
Where it should be the background (template value (absolute value)
However, in this case, the influence on the degree of coincidence becomes large and becomes very small. As a result, the determination is performed in consideration of the degree of missing or bleeding of characters (particularly the depth in the thin or thick direction).

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of a template creating apparatus and a pattern recognition apparatus according to the present invention will be described in detail below with reference to the accompanying drawings. FIG. 1 shows an example of an apparatus in which both the template creating apparatus according to the present invention and a character recognition apparatus which is an example of a pattern recognition apparatus are integrally configured. First, the hardware configuration will be mainly described. As shown in the figure, an image pickup device 1 such as an industrial video camera captures a character image 2 which is directly opposed to a video signal representing a grayscale image of the character image 2. It is designed to output. And
The output is sent to a binarization circuit 3, where it is binarized at an appropriate level and then stored in a binary image memory 4 composed of a RAM or the like. The character image 2 used in this example is composed of a plurality of characters printed on the label.

Further, the binary image stored in the binary image memory 4 is a product-sum calculation circuit 5 for character recognition (inspection) or a template creating unit in the CPU 6 for creating a template serving as a reference pattern. (Specific configuration will be described later). Which of the two is to be output is determined by the type of work currently being performed. The template created by the template creating unit is stored in the template memory 7 including a DRAM or the like.

Further, the binary image memory 4 and the template memory 7 are adapted to send the respective stored data to the product-sum calculation circuit 5 in accordance with the address obtained by the address calculation circuit 8. Then, the product-sum operation circuit 5 receives the template data W stored in the template memory 7 and the unknown input character data I stored in the binary image memory 4, and the following formula S = ΣI * W The sum-of-products calculation is performed to calculate the sum-of-products value S. The higher the sum of products S, the higher the quality. Then, the product sum value S is sent to the CPU 6.

Further, in this example, there is provided a work memory 9 for storing a table (a specific configuration will be described later) required for the inspection, and further the cursor key 10a and the ten-key pad 10.
Operation device 10 such as b and CRT 11 as a display device
Is equipped with. Then, each circuit and the like are driven by a control signal from the CPU 6.

Next, the template creating section in the CPU 6 will be described with reference to FIG. First, as a premise, a character image 2 that serves as a reference for creating a template in this example.
Is composed of a plurality of characters as described above, and the characters of the template creation target have the same size (point number). Then, as shown in FIG. 2, the template creation unit receives the binary image as shown in FIG. 3 captured through the imaging device 1 and stored in the binary image memory 4, and receives an arbitrary image in the character image 2. A character cutout unit 14 for extracting one character is provided. In this example, since the characters have the same size, the character cutout unit 14 automatically determines a predetermined area (width w, height) based on the starting point (xn, yn) input using the cursor key 10a. h) is set and cut out. In addition, serial numbers are assigned to all characters, and code numbers are assigned for each character type. (In this example, A is 11, B is 12, C is 13 ...). Then, the serial number and code No.
At the same time, a table (see FIG. 4) in which the above starting points (“xn, yn” (n is the same as the serial number)) are associated is created,
It is adapted to be stored in the work memory 9.

On the other hand, the image data in the area cut out by the character cutting section 14 is sent to the character part template creating section 15 which is the first converting means and the background part template creating section 16 which is the second converting means. It is designed to be used. Then, in the character portion template creation unit 15, the image data (reference character) is contracted, and then the distance conversion (the template value is increased as the distance from the edge of the character becomes longer) with respect to the character portion (inside the edge). To set). The template value outside the edge, that is, the background portion is set to 0. Since the distance conversion is performed after contraction in this way, the template value in the vicinity of the inside of the edge of the character (before contraction) that is the original reference becomes 0.

In the background portion template creating unit 16, the image data (reference character) is expanded, and then the distance conversion is performed with respect to the background portion (outside the edge) (the template is separated as the distance from the edge of the character increases). Set a large value). Then, the background portion template creating unit 16 adds a negative sign to the obtained template value. It should be noted that the template value of the inside of the edge, that is, the character portion is also set to 0 in the creating unit 16. Since the distance conversion is performed after the expansion as described above, the template value near the outside of the edge of the original character that serves as a reference also becomes zero.

Then, both template creating sections 15 and 16
Is output to the synthesizing unit 17, and the two templates given there are synthesized to create a template of a reference character. Specifically, the templates created by both creators 15 and 16 are added. Thereby, for example, the template for "C" becomes as shown in FIG. The inner and outer peripheral edges of the original character edge are 0 (Don't Ca
re). That is, the character on the template formed by the positive template value is slightly smaller than the original character, and the boundary between 0 and -1, which is the boundary between the backgrounds, is slightly larger than the original character. This allows
The permissible amount for positional deviation becomes large.

Further, in this example, the synthesizing unit 17 is provided with a limiter function so that the absolute value of each template value of the obtained template does not become larger than a predetermined value (for example, 5). That is, if there is a template value whose absolute value is 6 or more, all are set to 5. It should be noted that this predetermined value can be set arbitrarily, and further, in this example, the synthesizing unit 17
It was provided in the above, but each template creation unit 15 described above,
16 may be provided respectively, and a predetermined limit process may be performed when the character part template and the background part template are created so that the absolute value of the template value forming each template is set to a predetermined value or less. Then, the template created in this way is code No. Along with this, it is stored in the template memory 7 (see FIG. 6).

Further, in this example, the template memory 7 is used.
There is provided a correction unit 18 for reading out the template stored in and correcting the value of each template value forming the automatically created template to be large or small. That is, the template to be corrected by the operation device 10 is read into the correction unit 18 and displayed on the CRT 11.
Then, the template value to be corrected is specified by the input from the cursor key 10a, and then the specific correction value is input by the input from the ten key 10b (see FIG. 7). That is, in this example, the cursor key 10a is a means for specifying, and the ten-key 10b is a means for correcting.

Accordingly, in order to distinguish "C" from "O", for example, the presence or absence of the blank portion on the right side of C occupies a large weight, so that the template value located at the relevant portion of the background portion is determined, for example. The detection accuracy is improved by setting a large value such as -4 or -5.

Further, when the viewpoint is changed, it can be discriminated as "C" even if the peripheral portion of the left side portion is bleeded or chipped when "C" is printed. If the distance between the two becomes short due to blurring, there is a high possibility of misidentifying "O". That is, even if the degree of bleeding is the same, in the case of the former, it falls within the range of good products, but in the latter case, it may be a defective product, and in the latter part, it is judged as a defective product. Even if the bleeding is larger than the degree of bleeding, the bleeding spot may fall within the range of non-defective products in the former case.
Therefore, by correcting the template value as in this example, it becomes possible to more correctly determine whether the product is a good product or a defective product.

In this example, the template value is specified one by one with the cursor key 10a and the correction process is performed. However, for example, a predetermined area is specified with the cursor key 10a or the mouse, and the area is specified. Various means such as simultaneously correcting a plurality of template values by multiplying a template value existing therein by a predetermined coefficient or adding or subtracting a predetermined value can be adopted.

The process flow for creating a template for the cut-out character in the template creating unit (apparatus) described above is as shown in the flowchart of FIG. Then, steps 161 and 162 in the figure are a processing flow in the character portion template creating unit 15,
Steps 163 and 164 in the figure are processing flows in the background portion template creation unit 16, and steps 165 and 165 in the figure are shown.
166 is a processing flow in the composition unit 17.

Further, the CPU 6 is provided with a judging section 20 which is a judging means which constitutes a part of the character recognition device for performing the inspection process. The determination unit 20 receives the product sum value S calculated by the product sum operation circuit 5 and the threshold value Tn stored in the table in the work memory 9, compares the two, and calculates the product sum value S Is larger than the threshold value T, the character is judged to be non-defective.
The threshold value Tn stored in the work memory 9 is
Although it is set individually for each character, for example, after creating a template, take a picture of multiple good and bad samples,
On the other hand, the product sum value S is obtained, and the boundary value between the product sum value of good products and the product sum value of defective products is obtained.
Can be set.

Next, the operation for creating a template of a plurality of types of characters from one character image 2 using the above-mentioned device will be described with reference to FIG. At the same time, the binary image data of the reference character is stored in the binary image memory 4 via the binarization circuit 3 (S110).

Next, with respect to the first (serial number n = 1) character ("C") in the image data, the starting point (x1, y1)
) Is cut out. As a result, the serial number n and
Character code No. A table is created by associating the start point with the start point and stored in the work memory 9. The serial number n and the character code No. Since it is known in advance, the set starting point (xn, yn) is actually stored in the area of the work memory 9 for storing the starting point of the serial number n (S120 to S140).

Next, it is determined whether or not the character is the first type (character code No) (S150). That is, it is determined whether or not a template has already been created for that character, and if no template has been created, a predetermined process is performed on that character to create a template (S160). The specific processing of this creation is as in steps 161 to 166 shown in FIG. Then, the created template is code No. It is stored together with the template memory 7 (S170),
Character image 2 with processed characters (serial numbers) as a reference
It is judged whether or not it is the last number N of the characters included in (S18).
If it is smaller than 0) and N, there are still unprocessed characters, so the process returns to step 130 and the process is performed on the next serial number character. If it is determined in step 150 that the character has already been created, the template creation process is not performed and the process jumps to step 180 to determine whether "n = N". .. By doing so, templates corresponding to the number of types of characters (six in this example) are stored in the template memory 7. Then, if necessary, the correction unit 18
Modify the template value of the template created in.

In this example, one template is created for characters of the same type and the characters are shared. However, even if the characters are the same, predetermined processing is performed for each character if the printing location is different. Alternatively, a template may be created.

Next, the operation of the character recognition apparatus according to the present invention for inspecting an unknown character image using the template created as described above will be described with reference to the flow chart shown in FIG. First, the imaging device 1
The unknown character image 2 to be inspected is captured, binarized at an appropriate level, and then stored in the binarized image memory 4 (S201 to 203).

Then, the CPU 6 is operated to send a reset signal to the product-sum calculation circuit 5 to clear the previously calculated product-sum value to zero. Further, the work memory 9 is accessed to read the table in ascending order of the serial number, and the code No.
And the starting point (xn, yn) are given to the address arithmetic circuit 8. Then, in the address arithmetic circuit 8, the given code No. An address for scanning the template of the character designated by is obtained, and the address signal is output to the template memory 7. At the same time, an address for scanning the area of the character portion to be inspected from the given starting point is obtained by the same method as in the case of the template, and the address signal is output to the binary image memory 4. As a result, the character data having the serial number 1 is sequentially input from both memories 5 and 7 to the product-sum calculation circuit 5 (S204 to 207).

Next, the product-sum operation circuit 5 performs matching processing between the unknown binary image of the input n-th character and the distance-converted template. Ie 1
Alternatively, the binary image data represented by 0 and the data of each template value of the template are sum-of-products calculated, and the sum-of-products value S
Calculate n. As a result, for example, when the character portion in the unknown binary image data is missing, the binary data in that portion becomes 0, so the template value at that portion is not added, and due to bleeding, etc. When printing is performed by protruding from the character portion of, the negative template value of the background portion is added. That is,
The higher the quality, the higher the calculated sum of products. In addition,
When different characters are printed, the degree of coincidence between the character part of the binary image data and the character part of the template (the part of the positive template value) is low, and the character part of the binary image data and the template The degree of coincidence with the background portion (the portion of the negative template value) increases, and the calculated product sum value becomes extremely small (S208).

The calculated product sum value Sn is sent to the judgment unit 20 in the CPU 6, where it is read out from the table of the work memory 9 and compared with the threshold value Tn at the serial number n to give a given product sum value Sn threshold value. If it is larger, it is judged as a good product, and if it is smaller, it is judged as a defective product (S209-2).
11).

Then, it is judged whether or not all characters have been judged to be non-defective or defective (whether the serial number n is the final value N) (S212). Return and perform the above process for the next character. And when the processing for all characters is completed,
It is determined whether or not there is a defective character. If there is even one, the inspection object is a defective product, and if there is no character, it is determined to be a good product (S213 to 215).

In this example, in order to extract the data such as the generation status of the non-defective product / defective product and the degree of quality for each character as a material, the defective character is judged by the judgment of good / defective for each character in steps 209 to 211. Even if it is determined that the next character is processed, if the defective product is simply excluded, the process may be terminated when the defective product is determined in step 211,
Needless to say, various usage methods can be used.

In the above-described embodiment, the image pickup apparatus 1 is described as an example using an industrial video camera, but the image pickup apparatus 1 is not limited to this, and an image scanner or any other apparatus can be used.

Further, in the above-described embodiment, when the reference character image data is input to the template creating unit (device) provided in the CPU 6, the image data is temporarily stored in the binary image memory 4. However, this is because a plurality of characters are cut out from one image data and a template is created for each character.For example, one character is imaged, or the entire plurality of imaged characters are treated as one image. If one template is created, the clipping action is not necessary, so it may be input directly (without being stored in the binary image memory) (or in any case, it may be stored as well). Good). Further, in such a case, the character cutting section 14 is basically unnecessary, but in the case where it is better to specify the starting point (x, y) from the convenience of the subsequent character recognition (inspection) processing, , May be provided.

Furthermore, the object in the present invention is not limited to characters as in the above-mentioned embodiment, but may be pictures such as logos and other marks, and a pattern recognition device by general template matching. And a device for creating a template for pattern recognition thereof.

Furthermore, in the above-described embodiment, an example in which a character image composed of characters whose arrangement has been determined in advance is recognized has been described, but the present invention is not limited to this, and for example, unknown input is performed. Matching processing is performed on a plurality of templates created in advance for character image data (character type is also unknown), and the matching degree (sum of products value in the above example)
Can be used as a character recognition device for determining the character type. And, as mentioned above,
When matched with a template of different characters,
Since the value is extremely small, it is possible to easily judge the pass / fail.

Further, in the above-described embodiment, the template creating device and the character (pattern) are the most suitable embodiment.
Although the integrated device with the recognition device has been described, it goes without saying that it may be separately formed.

Further, in the above-mentioned embodiment, the template value created by performing the distance conversion in each of the template creating sections 15 and 16 is increased by 1 for each pixel, but this increase amount is set arbitrarily. Yes, in some cases, a steep rise may be provided by making it larger by 2 or more (for example, 1 to 3), or conversely, a gentle rise may be made by continuing the same numerical value. It can be implemented.

Further, as the synthesizing process in the synthesizing unit 17, the adding process is performed in the above-mentioned embodiment. However, for example, the synthesizing unit without adding a negative sign to the template value created by the background part template creating unit 16 The template of the background portion may be subtracted from the template of the character portion by inputting into 17, and various other modifications can be implemented.

[0043]

As described above, in the template creating apparatus and the pattern recognizing apparatus according to the present invention, the distance conversion is performed also on the template value of the background portion, and the larger the template value (absolute value) becomes, the farther from the edge of the reference pattern. Therefore, even if a character having a poor print quality, such as a chip or a bleed, can be accurately discriminated, the recognition rate and the quality inspection accuracy can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a preferred embodiment of a pattern (character) recognition device and a template creation device according to the present invention.

FIG. 2 is a diagram showing a part of a specific configuration in the CPU shown in FIG.

FIG. 3 is a diagram showing an example of a binary image obtained by capturing a character image.

FIG. 4 is a diagram showing an example of a table stored in a work memory.

FIG. 5 is a diagram showing an example of a template created by the template creating device of the present invention.

FIG. 6 is a diagram showing a state of a template stored in a template memory.

FIG. 7 is a flowchart illustrating a function of a correction unit.

FIG. 8 is a flowchart illustrating a procedure for creating one template.

FIG. 9 is a flowchart illustrating the operation of the template creation device.

FIG. 10 is a flowchart illustrating the operation of the character recognition device.

[Explanation of symbols]

 1 image pickup device 2 character image 3 binarization circuit 4 binarization memory 5 product-sum operation circuit (means for obtaining coincidence) 6 CPU 7 template memory (template storage means) 8 address operation circuit 9 work memory 10a cursor key ( 10b Numeric keypad (correction means) 11 CRT 14 Character cutout part 15 Character part template creation part (first conversion part) 16 Background part template creation part (second conversion part) 17 Composition part (composition) Means) 18 Correction unit (means for specifying / means for correction) 20 Judgment unit (judgment means)

Claims (3)

[Claims] 1. A first conversion means for distance-converting an inner portion of a pattern such as a character as an input reference to create a first template; and a distance-converting second portion for a background portion of the pattern. A template creating apparatus comprising: a second converting unit that creates a template; and a unit that combines the first and second templates created by the first and second converting units. 2. The template creating apparatus according to claim 1, further comprising: a unit that specifies a predetermined position of the created template, and a unit that corrects a template value of the predetermined position specified by the unit. 3. A template storage means for storing a template created by the template creation device according to claim 1, a template stored in the template storage means, and a pattern of an input unknown character image or the like. A pattern recognition device comprising: a means for comparing the calculated values with each other to calculate the degree of coincidence;