JPH0769932B2 - Character recognition device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a character recognition device, and more particularly to an improvement of a character recognition device that is used in an image pickup means to read a character and recognize the character accurately and at high speed.

[Prior Art] Stamped characters are resistant to aging, and their readability is not lost even in an environment where they are easily soiled. Therefore, it is widely used as a character that gives important information for production instructions and management in various production processes.

In particular, in the production process of automobiles, it is widely used to write a unique number to each component constituting the vehicle body, engine, and the like.

For this reason, various types of character recognition devices have been proposed in the past, and such a conventional device outputs a grayscale image by capturing an image of a stamped character portion using an ITV camera or the like, and outputs each grayscale image. Convert the pixel table to binary data,
It is formed to detect each character from the binary image data.

Then, the unknown character pattern detected in this way is compared with the previously registered standard character pattern by sequentially superimposing and comparing them, and the characters are identified based on the degree of coincidence or similarity between the two.

However, it is difficult to control so that the stamping conditions are always the same with a stamping machine that is widely used at present, and the groove depth varies for each stamped character. Therefore, in a so-called engraved character image, the character line width often varies as a whole even for the same kind of character.

Further, when the number of digits of the character to be recognized is large, for example, when the number of digits of the character is large, such as the vehicle body number, the angle and orientation of the groove of each character with respect to the proof become uneven. Therefore, in a so-called engraved character image, the character line width is partially uneven.

Furthermore, if the engraved characters are painted after being stamped, unevenness will occur around the characters due to uneven coating or uneven adhesion of paint around the engraving grooves, and the unevenness will cause the characters themselves to deform. . For this reason, when the engraved characters on the painted surface are imaged, the entire line width of the character part may change or the line width may partially differ, resulting in a character outline with a lot of unevenness. Often exhibits a pattern.

Therefore, as in the conventional device, even if the detected unknown character pattern and the standard character pattern registered in advance are simply pattern-matched, the imprinted character may not be accurately recognized for the various reasons described above. There was a problem that there were many.

In order to solve such a problem, JP-A-57-103582
The technology disclosed in Japanese Patent Laid-Open No. 57-103583 is known.

In this conventional technique, a plurality of standard character patterns corresponding to the variations are prepared and pattern matching is performed in order to cope with variations in the character image at the time of engraving a stamped character or at the time of imaging.

However, this makes it necessary to prepare a large number of standard character patterns in response to various deformed character patterns such as those on a painted surface.

Therefore, this conventional technique has a problem that the memory capacity required for registering the standard character pattern increases and the processing time for pattern matching increases.

Further, as another conventional technique, a technique is proposed in which the threshold used when converting a grayscale image of a stamped character into binary image data is changed to bring the character line width of the unknown character pattern closer to the character line width of the standard character pattern. Is being done.

For example, the technology disclosed in JP-A-61-208178,
Proceedings of 6th General Conference of the Institute of Electronics and Communication Engineers, 1985, 6-15
Such a technique is disclosed on page 3.

However, even if such a conventional technique is used, when recognizing the engraved characters on the painted surface, each character should be prepared in case the character line width is partially different or the character contour has irregularities. It is necessary to prepare multiple standard character patterns for each character.

Therefore, there is a problem in that the memory capacity required for registering the standard character pattern increases, and the marking character cannot be recognized quickly.

[Problems to be Solved by the Invention] The present invention has been made in view of such a conventional problem, and an object thereof is to provide a character whose line width varies wholly or partially and a contour shape. Characters that are not slippery and have many irregularities,
In particular, it is an object of the present invention to provide a character recognition device capable of accurately and quickly recognizing an engraved character on a painted surface.

[Means for Solving Problems] In order to achieve the above-mentioned object, the character recognition device of the present invention, as shown in FIG. 1, is an image pickup means for reading characters and character image data from the read image data. The character detection unit includes a character detection unit for cutting out, and a character identification unit for extracting the outer shape of the character from each of the cut out character image data to determine the type of the character, wherein the character detection unit converts the read image data into binary image data. A binary image conversion means for converting and outputting the image data, a character detection cutout means for cutting out image data of each character from the binary image data in a predetermined circumscribing frame, and a position and size of the cut out character image data are normalized and output. And a normalizing means for performing, wherein the character identification unit creates a circumscribing frame of characters from the normalized character image data,
As a coordinate series that represents the distance from the circumscribing frame to the character outline, an outline shape extraction means that extracts a character pattern that represents the outline shape of an unknown character, and a standard coordinate series that represents the distance from the standard circumscribing frame to the character outline of various characters. The standard pattern memory stores the character pattern as a character pattern, and stores the range representing the characteristic portion of the standard character pattern and its load value as a calculation condition for character recognition paired with the standard character pattern, and is output from the outer shape extracting means. Distance difference calculation means for sequentially comparing the unknown character pattern and each of the standard character patterns while applying a load to the characteristic portion according to the calculation condition, and calculating the weighted average value of the difference in the distance from the circumscribing frame to the character contour. And a character type determining means for determining the type of character of the unknown character pattern from the weighted average value of the calculated distance differences. It is characterized by

Here, it is preferable that the character detection unit is provided with an expansion / contraction circuit that performs expansion / contraction processing on the binary image data, removes isolated pixel data that becomes noise, and outputs the data to the character detection / cutout unit.

Further, in the present invention, the circumscribing frame may have various shapes if necessary, but it is preferable to use a rectangular frame for convenience of data processing.

In this case, the character detection / cutout means detects each character one by one from the binary image data, finds a circumscribed rectangular frame for each detected character, and circumscribes a circumscribed rectangular frame corresponding to each character image data from the binary image data. The standard pattern memory stores a pair of left and right coordinate series representing the distance from the left and right sides of the standard circumscribing rectangular frame of various characters to the character contour as a standard character pattern, and the characteristic part of the standard character pattern. It is preferable that the range and the load value thereof are recorded as a calculation condition for character recognition that forms a pair with the standard character pattern.

Further, in the present invention, the standard pattern memory stores a coordinate series obtained by smoothing the distances from the standard circumscribing frame of various characters to the character contour as the standard character pattern. The coordinate series obtained by smoothing the distance to is formed so as to be extracted as a character pattern representing the outer shape of an unknown character. By doing so, even when recognizing a character having a contour shape with many irregularities, only the main contour of the character can be extracted without being affected by the irregularity, and the character recognition can be performed more accurately.

Further, in the present invention, the distance difference calculating means shifts the unknown character pattern sent from the outer shape extracting means by one pixel in the vertical and horizontal directions to create five types of unknown character patterns for the same character. The creation circuit, the contour shape difference circuit that superimposes the contour shapes of the unknown character pattern and the standard character pattern, and creates and outputs the coordinate series corresponding to the difference between them, and the feature part according to the calculation conditions that pair with the standard character pattern. A weighted average calculation circuit that calculates the total value of the coordinate series corresponding to the calculated distance difference while adding the load, and outputs the value obtained by weighting the total value by the number of coordinate series as the difference between the distance and the unknown character pattern. , The minimum distance calculation with respect to all five standard character patterns, and outputs the minimum value to the character type determination means. It is preferable that the circuit is formed so as to recognize the character without being affected by the positional deviation of the character.

By doing so, the character recognition can be performed more accurately even for the character whose line width is wholly or partially changed.

[Operation] The present invention is configured as described above, and the operation will be described below.

The apparatus of the present invention first reads a character by the image pickup means to obtain a grayscale image thereof, and then converts the grayscale image into binary image data by the binary image forming means.

This binary image data is expanded / contracted using an expansion / contraction circuit.
By performing the contraction processing, it is possible to remove small noises scattered in the background in the binary image, compensate for breaks and missing lines in the character portion of the binary image, and improve character recognition accuracy.

Next, the character detection / cutout means detects only the character portion in the binary image, forms a rectangular frame circumscribing the character, and cuts out the character image in the rectangular frame.

After that, the normalizing means normalizes the size by enlarging or reducing the character image in the circumscribing rectangular frame so that the size of the circumscribing rectangular frame of the cut out character becomes a standard size. In order to compare the character image with the standard character pattern, the position where the character image is aligned with the determined position is normalized.

Next, the contour shape extraction means extracts a coordinate series representing the distance from each side of the normalized circumscribing rectangular frame of the character to the contour of the character as a character pattern representing the contour shape of the unknown character.

In the present invention, the contour shape thus obtained is not used as it is as an unknown character pattern, but the contour shape obtained is smoothed to obtain a smooth contour shape, and the smoothed contour shape is used as an unknown character pattern. Used as.

By doing so, even for a character having a contour shape with a large unevenness, only the main contour of the character can be extracted as an unknown character pattern without being affected by the unevenness. Therefore, for example, even for a character having an irregular contour such as a character engraved on the painted surface, the character recognition can be performed with higher accuracy.

Next, the distance difference calculating means smoothes the distance from the circumscribing frame registered in the standard character pattern memory to the unknown character pattern thus obtained and the character outline, and represents each coordinate sequence. According to the calculation condition, the standard pattern is sequentially compared while applying a load to the characteristic portion, and pattern matching is performed to calculate the weighted average value of the differences in the distance from the circumscribed frame to the character contour.

Here, the calculated weighted average value becomes the minimum value when the unknown character pattern and the standard character pattern match.

Therefore, the character type determination means searches the standard character pattern having the smallest weighted average value of the calculated distance differences to determine the type of character of the unknown character pattern.

The calculation of the weighted average value of the distance difference is preferably performed as follows.

That is, the unknown character pattern and the standard character pattern are added while applying a predetermined weight to the coordinate series in the range representing the characteristic part of the standard character pattern in accordance with the operation conditions registered in the standard pattern memory as a pair with the standard character pattern. The difference from the character pattern is summed over all coordinate series, and the value is weighted averaged by the number of coordinate series to obtain the difference α of the distance.

At this time, if necessary, only the coordinate series of the range representing the characteristic part of the calculation condition is used, and the difference β of only the specific range is calculated in the same manner as the difference α.

The difference β thus obtained is larger than the difference α when the unknown character pattern and the standard character pattern do not match. For this reason, the character type determination means, only the difference α,
Alternatively, the character type can be determined with higher accuracy using either of the differences α and β.

Further, in the present invention, the shift pattern creating circuit is used to shift the unknown character pattern sent from the outer shape extracting means by one pixel in the vertical and horizontal directions to create five types of unknown character patterns for the same character. In addition, it is preferable to use the minimum distance calculation circuit to obtain the minimum distance values of the five types of unknown character patterns for all standard character patterns.

By doing this, even if the character line width of the unknown character pattern fluctuates over the whole character or a part of the character, or even if there is a positional deviation in the character pattern,
The unknown character can be accurately recognized based on the standard character pattern without being affected by these factors.

[Effects of the Invention] As described above, according to the present invention, a circumscribing frame of a character image cut out from read image data is created, and a coordinate series representing the distance from the circumscribing frame to the character contour is defined as the contour of an unknown character. Since it is extracted as an unknown character pattern that represents the shape,
The amount of data of the extracted unknown character pattern is extremely small as compared with the conventional technique in which the character pattern is extracted as it is from the cut out character image. Therefore, the data amount of the standard character pattern registered for pattern matching with the unknown character pattern can be greatly reduced. As a result,
According to the present invention, the capacity of the standard pattern memory used for registering the standard character pattern can be greatly reduced, and further, the data processing time for recognizing the character can be shortened and the character can be recognized quickly. be able to.

Further, according to the present invention, the range representing the characteristic portion of the standard character pattern and the load value thereof are stored as the calculation condition for character recognition forming a pair with the standard character pattern, and pattern matching with the unknown character pattern is performed. The feature is emphasized. Therefore, even if there is a similar character, the character can be accurately recognized.

Further, in the present invention, the unknown character pattern output from the outer shape extracting means is shifted in four directions of up, down, left and right to create five unknown character patterns for the same character type. By performing pattern matching with a character pattern, it is possible to perform more accurate character recognition even for characters whose line width varies in whole or in part,
Moreover, it is possible to obtain a character recognition device that is resistant to positional deviation of character patterns.

Further, in the present invention, since the outer shape of the character is smoothed to create the unknown character pattern, even when recognizing a character having a contour shape having many irregularities, only the main contour shape of the character is not affected by the irregularity. It is possible to obtain a character recognition device that can extract characters and that can more accurately recognize characters that have a smooth contour shape and that have many irregularities, especially engraved characters on a painted surface. it can.

[Embodiment] Next, a preferred embodiment of the character recognition device according to the present invention will be described with reference to the drawings.

FIG. 2 shows a preferred embodiment of the character recognition device according to the present invention. In the figure, a frame 12 which is a component of an automobile body has an engraved character A to be recognized. There is. The engraved character A is a 12-digit character string including, for example, alphanumeric characters and symbols.

The character recognition device 10 of the embodiment is for automatically reading the marking character A, and the marking character A is illuminated by the fluorescent lamp 14,
The image is taken by the TV camera 16. The engraved character A is perceived as a gray image in which the character part in the groove is dark and the surface part of the frame 12 is light.

Then, the character detection unit 100 cuts out the character image data from the read image data output from the TV camera 16, and the standard character pattern automatic generation circuit 200 through the line switching circuit 34.
And selectively output to the character identification unit 300.

The standard character pattern automatic creation circuit 200 has a line switching circuit 3
Based on the character image data input via 4, create a coordinate series that represents the distance from the standard circumscribing frame of various characters to the character outline as a standard character pattern, and the range that represents the characteristic part of the standard character pattern and its load. The value is obtained as a calculation condition for character recognition that forms a pair with the standard character pattern.

Therefore, before recognizing characters, the line switching circuit 34 is set on the standard character pattern automatic creation circuit 200 side, and various standard characters are photographed using the TV camera. The standard character pattern of the standard character and the calculation condition for character recognition that is paired with the standard character pattern are created, and the standard pattern memory 4 of the character identification unit 300 is created.
It will be registered in 2.

Further, the character identification unit 300 extracts the outer shape of the character as an unknown character pattern from the character image data input through the line switching circuit 34, and calculates the unknown character pattern and the registered standard character pattern as the calculation condition. And the character type is determined.

Therefore, after the standard character patterns of various standard characters and the calculation conditions for character recognition corresponding to the standard character patterns are registered in the standard pattern memory 42, the line switching circuit 34 is connected to the character identifying section 300.
When the character marking unit 300 is set on the side and the various characters A are imaged using a TV camera, the character identification unit 300 automatically determines the type of the character A based on the character image data input via the line switching circuit 34. It will be judged.

Character Detection Unit In the present embodiment, the output signal of the TV camera 16 input to the character detection unit 100 is converted into a digital value by the A / D conversion circuit 18 and stored in the first image memory 20.

The grayscale image data stored in the first image memory 20 is 2
The binarization circuit 22 binarizes the binarized image data based on a predetermined binarization threshold value and stores the binarized image data in the second image memory 24. At this time, in the binary image data, for example, the dark portion (character portion) of the grayscale image data is “1” and the light portion (background portion).
Is stored as “0”.

The binary image data stored in the second image memory 24 is subjected to contraction processing and expansion processing once by the expansion / contraction circuit 26, and then output to the character detection / cutout circuit 28.

As a result, it is possible to remove isolated pixel data that becomes noise (isolated "1" corresponding to noise other than "1" that constitutes a character), and pixel data that constitutes a character (of "1" Pixel data) Discontinuity can be corrected.

Next, the character detection / cutout circuit 28 detects each character one by one from the binary image data of, for example, a 12-digit character string, and obtains a rectangular frame circumscribing the detected character to each character. Then, each detected character is cut out by the circumscribing rectangular frame, and binary image data for each character is output.

Next, the size normalization circuit 30 enlarges or reduces the character image in the circumscribed rectangular frame for each character so that the circumscribed rectangular frame of each cut out character has a predetermined standard size. And make the size of the character image constant.

Next, the position normalization circuit 32 calculates the center coordinates of the circumscribing rectangular frame, aligns the center coordinates with a predetermined alignment point, and aligns the aligned character image data via the line switching circuit 34 to automatically generate a standard character pattern. The data is selectively output to the creation circuit 200 or the character identification unit 300.

Standard Character Pattern Automatic Creation Circuit FIG. 3 shows a preferred example of the standard character pattern automatic creation circuit 200 according to this embodiment, and FIG. 4 shows the standard character pattern automatic creation circuit shown in FIG. Is used to create a standard character pattern of character type "5".

When operating the standard character pattern automatic generation circuit 200 of the embodiment, the line switching circuit 34 shown in FIG. 2 is set to the side of the broken line, and various conditions such as an engraving position, an engraving strength and illumination are set. With some changes, a plurality of character image data with different conditions are obtained for the same character type. Then, it is input to the cumulative pattern creating circuit 50.

The cumulative pattern creation circuit 50 calculates the frequency of whether it becomes a character part or a background part for each pixel of character image data of the same character type, creates a cumulative pattern representing the frequency distribution, and creates a third image memory 52. Remember.

In FIG. 4A, for example, a cumulative pattern when the character type is “5” is simulated. In the figure, the area surrounded by the broken line is the area of 10% or more of the probability of being the character portion, and the area filled in black is the probability of being the character portion.
% Or more area.

Next, the average character pattern creating circuit 54
From the cumulative pattern stored in 52
An area of 50% or more pixels is extracted to create an average character pattern, which is stored in the fourth image memory 56.

FIG. 4b schematically shows an average character pattern created for the character type “5”, for example.

Next, the outer shape creating circuit 58 forms a rectangular frame circumscribing the average character pattern stored in the fourth image memory 56. Then, horizontal scanning is performed from the left and right sides of the rectangular frame to the contour line of the character, the distances from the right and left sides to the contour line are obtained along each horizontal scanning line, and the coordinate series corresponding to the position of each side is created. To do. Then, the coordinate series of the right side and the left side thus obtained are stored in the fifth image memory 60 as the outer shape for the left side and the outer shape for the right side, respectively.

FIG. 4 shows a state in which a circumscribing rectangular frame i for the average character pattern b is created and a contour shape c for the left side and a contour shape d for the right side are created by taking the character type “5” as an example.

In this way, according to the apparatus of this embodiment, the coordinate series representing the outer shape of each character type is created and stored in the fifth image memory 60.

The similar character type selection circuit 62 compares the outer shape of the character type of interest and the outer shape other than the character type among the outer shape of each character type stored in the image memory 60 of FIG. 5, and selects a similar character type. .

That is, when the sum of the differences between the coordinate series indicating the outer shapes of the two over all coordinate series is smaller than a predetermined value, it is determined that the two are similar, and the similar character type is stored in the similar character type data memory 64. Remember.

At this time, a plurality of similar character types for the focused character type may be selected, or no similar character type may be selected. For example,
FIG. 4 shows a contour e on the left side of the character type “S” that is similar to the contour c on the left side of the character type “5”. Actually, the foreign currency shapes on both the left and right sides are used, and similar character types are selected for all character types.
Store in memory 64.

Next, the outline feature portion extraction circuit 66 determines the left and right outline shapes (coordinate series) of the character type of interest stored in the fifth image memory 60 and the left and right outlines of the character type stored in the similar character type data memory 64. The outer shape (coordinate series) of is compared with the outer shape of the target character type to obtain a range that well represents the characteristics of the outer shape of the character type.

That is, the difference between the coordinate series indicating the outer shapes of the two is calculated over all coordinate series, and then the sum of the differences over the coordinate series for each fixed small section is calculated. Then, all small sections whose total sum exceeds a predetermined value are picked up, ranked in descending order of value, and stored in the range storage memory 68 as a range indicating a characteristic part.

When there are a plurality of similar character types, the range indicating the characteristic part ranked for each character type is re-ranked and stored according to the frequency of each character type.

In FIG. 4, for example, for the character type “5”, the outer shape c
Is compared with the outer shape e of the similar character type “S” to show a range in which the outer shape c is particularly different from the outer shape e, that is, an inverse ratio f indicating a characteristic portion is obtained.

On the other hand, the outer shape smoothing circuit 70 smoothes the left / right outer shape stored in the fifth image memory 60, and stores the smoothed left / right outer shape as a standard character pattern in the storage memory.
The range is stored in the sixth image memory 72 in combination with the range indicating the characteristic part f stored in 68.

The smoothing method is performed by calculating an average value of a total of 5 points including two points before and after each point of the coordinate series indicating the outer shape and using the average value as the coordinate value of each point. For example, as for the character type “5”, FIG. 4 shows an outline shape g obtained by smoothing the outline shape c with respect to the left side.

Next, the distance calculation condition determining circuit 74 uses the sixth image memory 72.
Among the smoothed contour shapes stored in the above, between the contour shape of the character type of interest and the contour shape other than the character type, both character types are sequentially compared while applying a load to the range f indicating the feature of the contour shape. Pattern matching is performed, and the difference between the smoothed outer shape coordinate series is summed over all coordinate series. Then, the summed values are weighted averaged by the number of coordinate series, and the value is set as the distance difference α.

In this way, α is calculated for all the other character types with respect to the character type of interest, and the minimum distance difference α is obtained.

Repeat this calculation while changing the value of the load,
Calculate the minimum distance α for each load value. Then, the value of the load at which the minimum value of the distance difference α is maximized is calculated,
The range indicating the load value and the characteristic portion is used as a calculation condition for calculating the distance.

If the minimum value α thus obtained does not exceed the predetermined threshold value, only the coordinate series of the range representing the characteristic part of the calculation condition is used, and the difference β of only the specific range is calculated in the same manner as the difference α. Calculate Then, the value of the load at which the minimum value of the difference β in the distance is the largest is obtained, and this load value and the range indicating the characteristic portion are used as the calculation conditions for calculating the distance.

That is, for the difference α, all the character types are taken into consideration, and the load value is obtained using the range other than the characteristic portion. Therefore, it may be difficult to determine the similar character type only by the difference α.

On the other hand, the difference β is effective for distinguishing the similar character type because the load value is obtained by using only the feature part in consideration of only the similar character type, and the load value is the load of the difference α. It is often different from the value.

Therefore, the difference α is sufficient for the dissimilar character type, but when there is a similar character type, it is preferable to add the difference β to the difference α and make the determination based on both values.

Therefore, the difference β may be obtained for similar character types as necessary.

Here, the load value is added to the range showing the characteristic part in the coordinate series showing the smoothed outer shape, but changes with respect to the coordinate series of the small sections ranked within the range showing the characteristic part. You may. For example, in FIG. 4, the range f indicating the characteristic portion of the smoothed outer shape g for the character type “5” is shown.
The load value h given to is illustrated.

In this way, the calculation condition of each character is obtained, stored in the image memory 76 of FIG. 7, and then stored in the sixth image memory 72 and the calculation condition stored in the seventh image memory 76. A coordinate series indicating the smoothed outer shape and a set for each character type are combined and output to the standard pattern memory 42 shown in FIG.

In this way, a pair of standard character patterns and their calculation conditions are registered in the standard pattern memory 42 for each character. For example, for the character type “5”, FIG.
f and h are registered as standard character patterns and calculation conditions.

Character Recognition Unit The character recognition unit 300 of this embodiment includes an outer shape extraction circuit 40, a standard pattern memory 42, a distance difference calculation circuit 44, and a character type determination circuit 46, and as shown in FIG. It is formed to recognize an unknown character pattern by side-setting.

Then, as in the case of creating a standard character pattern, the engraved character A stamped on the frame 12 is photographed using the TV camera 16 and detected / cut out for each character using the character detection unit 100. After the size and position are normalized, the unknown character image pattern is sequentially input to the outer shape extracting circuit 40.

The outer shape extracting circuit 40 creates a circumscribed rectangular frame of an unknown character image pattern, scans the contour line of the character pattern from the left side and the right side of the circumscribed rectangular frame, and determines the value of the distance from each side to the contour line of each side. A coordinate series is created corresponding to the position, and each is obtained as an outer shape for the left side and an outer shape for the right side.

Further, with respect to the contour shapes for the left side and the right side, similarly to the contour shape smoothing circuit 70 of FIG. 3, a coordinate series in which the contour shape is smoothed is obtained, and this is directed to the distance difference calculation circuit 44 as an unknown character pattern. Output.

FIG. 6 shows a state in which the contour shape extraction circuit 40 and the distance difference calculation circuit 44 of this embodiment extract the contour shape for the character type "5" and calculate the distance to the standard character pattern "3". ing. For example, FIG. 6A shows a state in which a circumscribing rectangular frame b is created for a normalized pattern a of the character type “5”, and an outer shape c for the left side and an outer shape d for the right side are obtained. Has been done,
FIG. 6B shows how the outer shape c with respect to the left side is smoothed to obtain the outer shape e.

FIG. 5 shows a specific configuration of the distance difference calculation circuit 44 of this embodiment. The smoothed contour shape of the unknown character pattern sent from the contour shape extraction circuit 40 is a shift pattern. It is input to the creating circuit 44a.

Then, the shift pattern creation circuit 44a shifts the position of the input contour shape by one pixel to the top, bottom, left, and right.
A new contour shape is newly created, and for the same character type, five types of contour shapes are created for each of the top, bottom, left, and right together with the reference contour shape.

On the other hand, from the standard pattern memory 42, the left and right smoothed contour shapes indicating the standard pattern of each character type are sequentially sent and stored in the standard character pattern setting memory 42a, and the calculation conditions thereof are the calculation condition setting memory. It is stored in 42b.

Then, the outer contour shape difference circuit 44b first selects one of the five outer contour shapes that are shifted by the shift pattern creating circuit 44a and stores the outer contour shape in the standard character pattern setting memory 42a. The coordinate shape corresponding to the difference between the two is created, and the coordinate series is created and then output to the weighted average calculation circuit 44c.

For example, in FIG. 6B, the left outer shape e of the character type “5” and the left outer shape f of the character type “3” of the standard character pattern are overlapped with each other, and the coordinate series h indicating the difference shape thereof is shown. Is shown to be created.

Then, the weighted average calculation circuit 44c uses the inter-outer shape difference circuit 4
For the coordinate series indicating the differential shape sent from 4b, according to the operation conditions sent from the operation condition setting memory 42b, while adding a predetermined load to the range indicating the characteristic part of the standard character pattern, all values of the coordinate series are Of the coordinate series, the weighted average is obtained by the number of coordinate series, and this value is output as the distance difference α to the minimum distance selection circuit 44d.

In FIG. 6B, the difference shape h for the character type “5” is shown.
The distance is calculated by adding a load to the characteristic range g
An example is shown in which a value i is obtained as the distance difference α.

In this manner, the same calculation is performed for all of the five types of position-shifted outline shapes created by the shift pattern creation circuit 44a to obtain the distance difference α, which is output to the minimum distance selection circuit 44d.

The minimum distance selection circuit 44d selects the minimum value among the five types of distance differences α sent from the weighted average calculation circuit 44c and outputs it to the character type determination circuit 46 of FIG. That is, the minimum value of the distance α is selected and output for each of the left and right contour shapes.

Then, as shown in FIG. 2, the character type determination circuit 46 stores the distance differences α for the left and right outline shapes sent from the distance difference calculation circuit 44, respectively, and stores the distance differences α for all standard character patterns. After that, the distance differences α are rearranged with respect to the left outer shape and the right outer shape.

That is, the sum of the distance differences α for the left and right outline shapes of a certain standard character pattern is rearranged in ascending order,
When the character type with the smallest sum is the first candidate and the character type with the second smallest sum is the second candidate, the value of the first candidate is smaller than a predetermined value and the value of the difference between the first candidate and the second candidate. Is greater than a predetermined value, the character type of the unknown character pattern is determined to be the first candidate character type, and in other cases, it is determined that there is a reading error and cannot be determined.

At this time, if necessary, only the coordinate series of the range representing the characteristic part of the calculation condition is used, and the difference β of only the specific range is calculated in the same manner as the difference α.

The difference β thus obtained is larger than the difference α when the unknown character pattern and the standard character turn do not match. In particular, in the case of a similar character type, the difference β obtained by focusing only on the feature portion is larger than the difference α in order to identify the original similar character type, and thus the identification is easy.

Therefore, the character type determination circuit 46 can determine the character type with higher accuracy using only the difference α or using the differences α and β.

In this way, the character recognition device 10 according to the present invention makes it possible to quickly and accurately read the imprinted character.

If a standard pattern memory that stores the existing standard character patterns is used, the line switching circuit 34 can omit the standard character pattern automatic generation circuit 200.

Further, in the above-described embodiment, the case where the rectangular frame is used as the circumscribing frame of the character has been described as an example, but the present invention is not limited to this, and a circumscribing frame having another shape may be used as necessary. .

Further, in the above-described embodiment, the case of recognizing a stamped character has been described as an example, but the present invention is not limited to this and may be used for recognizing another character as necessary.

Further, in the above embodiment, the case where the optical image pickup means is used has been described as an example, but the present invention is not limited to this.
Other types of imaging means may be used if desired.

[Brief description of drawings]

1 is a block diagram of a character recognition device according to the present invention, FIG. 2 is a block diagram of a preferred embodiment of the character recognition device according to the present invention, and FIG. 3 is a standard character pattern in the device shown in FIG. A detailed block diagram of the automatic generation circuit, FIG. 4 is an explanatory view showing a state of generating a standard character pattern for a character type “5” in the standard character pattern automatic generation circuit shown in FIG. 3, and FIG. 5 is shown in FIG. FIG. 6 is a detailed block diagram of the distance difference calculation circuit in the device shown in FIG. 6, and FIG. 6 shows the outline shape extraction circuit and the distance difference calculation circuit of the embodiment device shown in FIGS. It is explanatory drawing which shows the state which calculates the distance with the character pattern "3". 10 …… Character recognition device 16 …… TV camera as image capturing means 22 …… Binarization circuit 26 …… Expansion / contraction circuit 28 …… Character detection / cutout circuit 30 …… Size normalization circuit 32 …… Position normalization Circuit 40 …… Outer shape extraction circuit 42 …… Standard pattern memory 44 …… Distance difference calculation circuit 44a …… Shift pattern creation circuit 44b …… Outer shape difference circuit 44c …… Weight average calculation circuit 44d …… Minimum distance calculation Circuit 46 …… Character type judgment circuit 100 …… Character detection unit 300 …… Character identification unit A …… Stamped characters

Claims (5)

[Claims] 1. An image pickup device for reading a character, a character detection unit for cutting out the character image data from the read image data, an external shape of the character is extracted from each of the cut out character image data, and the type of the character is determined. A character identifying section, the character detecting section converting the read image data into binary image data and outputting the binary image data; and extracting the image data of each character from the binary image data with a predetermined circumscribing frame. And a normalization unit for normalizing and outputting the position and size of the extracted character image data, the character identification unit including a character circumscribing frame from the normalized character image data. Make,
A contour shape extraction unit that extracts the coordinate series that represents the distance from the circumscribed frame to the character contour as a character pattern that represents the contour shape of the unknown character, and the standard coordinate series that represents the distance from the standard circumscribed frame to the character contour of various characters. The standard pattern memory stores the character pattern as a character pattern, and stores the range representing the characteristic portion of the standard character pattern and its load value as a calculation condition for character recognition paired with the standard character pattern, and is output from the outer shape extracting means. Distance difference calculation means for sequentially comparing the unknown character pattern and each of the standard character patterns while applying a load to the characteristic portion according to the calculation condition, and calculating the weighted average value of the difference in the distance from the circumscribing frame to the character contour. And a character type determination means for determining the character type of the unknown character pattern from the weighted average value of the calculated distance differences, The standard pattern memory stores the coordinate series obtained by smoothing the distance from the standard circumscribing frame of various characters to the character contour as a standard character pattern, and the contour shape extraction means smooths the distance from the circumscribing frame to the character contour. A character recognition device characterized in that it is formed so as to extract the coordinate series obtained in this way as a character pattern representing the outline shape of an unknown character. 2. The apparatus according to claim 1, wherein the character detection unit performs expansion / contraction processing on the binary image data, removes isolated pixel data that becomes noise, and functions as character detection / cutout means. A character recognition device having an expansion / contraction circuit for directing output. 3. The apparatus according to any one of claims (1) and (2), wherein the character detection / cutout means detects each character from the binary image data one by one and circumscribes each detected character. The standard pattern memory is formed so as to obtain a rectangular frame and cut out each character image data from the binary image data with a corresponding circumscribing rectangular frame. The standard pattern memory is a left and right representing the distance from the left and right sides of the standard circumscribing rectangular frame of various characters to the character contour. Character recognition characterized by storing a pair of coordinate series as a standard character pattern, and storing a range representing the characteristic part of the standard character pattern and its load value as a calculation condition for character recognition forming a pair with the standard character pattern apparatus. 4. The apparatus according to any one of claims (1) to (3), wherein the distance difference calculation means has one pixel in the vertical and horizontal directions for the unknown character pattern sent from the outer shape extraction means. Shift pattern creating circuit for creating five types of unknown character patterns for the same character, and the outline shapes of the unknown character pattern and the standard character pattern are overlapped to create and output a coordinate series corresponding to the difference between the two. While adding a load to the feature part according to the calculation circuit that forms a pair with the standard character pattern and the shape difference circuit, the total value of the coordinate series corresponding to the calculated distance difference is obtained, and the total value is weighted average by the number of coordinate series. The weighted average calculation circuit that outputs the calculated value as the difference of the distance from the unknown character pattern, and the minimum value of the distance from the five types of unknown character patterns to all the standard character patterns. Te calculated, wherein the minimum distance computing circuit for outputting toward character type determining means, the character recognition device, characterized in that the character recognition without being affected by the positional deviation of a character. 5. A character recognition device according to any one of claims (1) to (4), characterized by recognizing a stamped character.