Character recognizing device and its method

Description: TECHNICAL FIELD [Detailed description of the invention]

【0001】

[Technical field to which the invention belongs]

The present invention relates to a character recognition device such as an optical character recognition device (OCR) and a character recognition method, and in particular, inputs an image of an original image in which characters are represented, and cuts out and preprocesses characters from the input image. The present invention relates to a character recognition device and a character recognition method for performing the above.

【0002】

[Conventional technology]

Conventionally, as shown in FIG. 15A, there has been an optical character recognition device (OCR) that recognizes characters written on a form. The device according to the conventional example is an image input processing means 91 that inputs an image of an original image in which characters are represented and cuts out or preprocesses the characters from the input image, and six or more horizontal lines for erasing the characters. Character identification means 92 for discriminating characters with horizontal lines and 6 or more characters other than the discriminated characters with horizontal lines are extracted and compared with the standard pattern to identify the characters. It has means 93. As a result, characters with 6 or more horizontal lines are judged to be unrecognizable (hereinafter referred to as "reject").

【0003】

[Problems to be Solved by the Invention]

By the way, in the OCR according to the conventional example, as described above, there is a rule that it is necessary to enter 6 or more horizontal lines in order to erase the characters incorrectly entered in the form (JIS). Therefore, in order to erase the characters once entered, six or more horizontal lines must be entered. This would be a heavy burden for the writer, so even if he knew this rule, it was hardly followed. Therefore, there is a problem that even a character with a line that seems to be obvious to anyone with the intention of erasing the character is not rejected because it does not meet the above rule and is misread. ..

【0004】

For example, in the figure (b), the intention to erase "0" is not rejected and is recognized as "8", and in the figure (c), the intention to erase "1" is not rejected and " It is recognized as "8", and in the figure (d), the intention to erase "7" is recognized as "4" without being rejected, and in the figure (e), the intention to erase "6" is rejected. Here is an example of what was recognized as "6". Therefore, the present invention reliably erases what is recognized as a erased line indicating the erasure of characters by anyone to prevent misreading, but does not erase characters other than the erased characters. The purpose of the present invention is to provide a character recognition device and a character recognition method capable of recognizing characters with high reliability without imposing a burden on a person.

【0005】

[Means for solving problems]

In order to solve the above technical problems, the first invention comprises, as shown in FIG. 1, an image input processing means 11 for inputting an image, cutting out a character pattern from the input image, and performing preprocessing. Whether the character is a simple line segment consisting of only line segments in a certain direction such as a horizontal line or a vertical line, which is assumed to represent the erasure of a character, or a complex line segment having a complicated linear shape. , It has a lineless character discrimination processing means 12 for discriminating whether or not it is a normal character without a line segment, and an identification means 13 for discriminating characters.

【0006】

Here, "input of an image" means inputting an image of an original image in which characters are represented. For example, when recognizing characters on a form, the character pattern on the form is read by the photoelectric conversion unit. It is carried to the field of view, the character pattern is read, converted into an electric signal, and a digital signal is output by binarization or the like. "Character cutout" means to cut out only a character signal from the image data of a digital image including input noise. "Preprocessing" refers to removing noise and normalizing the position, size, inclination, line width, etc. of characters. It should be noted that the determination of the lined characters may also be included in the preprocessing. The "character with a line" means a character with a line indicating that the character should be erased, and in the present invention, the line has a simple line and a complex line. A "simple line segment" is a line segment consisting of only line segments in a certain direction such as a horizontal line or a vertical line, which is supposed to represent the erasure of characters. For example, a horizontal line, a vertical line, a diagonal line, or the like is a line segment. , 1 or 2 lines may be subtracted. Further, the "constant direction" is not necessarily limited to one direction such as horizontal, vertical or 45 °, and may be a vanishing line in which the deviation from such a direction is within a predetermined range, for example. When the assumed vanishing line is a horizontal line, the horizontal line in the range of, for example, ± 10 ° from the horizontal line (angle θ = 0) is also included. Further, the simple line disappearance processing is not necessarily limited to the case where one type of line disappearance such as a horizontal line is assumed, and it may be determined as a simple line disappearance by assuming a plurality of types of line disappearance. As a result, it is possible to erase what is clearly recognized as a line disappearance without missing it and prevent erroneous reading. In the embodiment described later, it is determined by performing adjacent projection.

【0007】

The "complex line-breaking" is a character in which a complicated line is drawn on the character and the character is erased, and in order to determine whether or not the character is a complex line-breaking, for example, it is shown in the third invention. As described above, the complex amount is extracted and discriminated. "Character identification" means, for example, performing feature extraction or the like to compress information, inputting the obtained pattern, calculating the degree of matching with a standard pattern prepared in advance, and calculating the degree of matching with a standard prepared in advance, and the standard having the highest similarity. The pattern category is output as the recognition result. According to the present invention, discrimination processing is performed for simple lined characters and complex lined characters, and characters are identified for normal characters. Therefore, not only the line disappearance with the expected horizontal line or vertical line, but also the simple line disappearance that is slightly deviated from the line disappearance, but is not only the simple line disappearance that is clearly recognized as the line disappearance, but also the simple line disappearance. Even if it has a complicated linear shape completely different from that of the above, but it is clearly recognized as a derailment, it can be discriminated as a derailment without omission.

【0008】

In the second invention, as shown in FIG. 2, in the first invention, the lined character discriminating processing means 12 is a complex lined character discriminating processing means 14 that performs discriminating processing on the complex lined character. When it is determined by the complex lineless character discrimination processing means 14 that the character is not a complex lineless character, the linelessness is removed from the simple lined character candidate determined to have the simple lineless character. It has a simple lined character discrimination processing means 15 that performs a simple lined character discrimination process based on whether or not a character can be identified as a character. According to the present invention, the simple lineless character candidate is determined only when it is determined that the character candidate is not a complex lineless character candidate after first determining whether or not the character candidate has a complex lineless character. It is possible to prevent the misidentification that a character that cannot be identified as a character is not a lined character when the character is erased, and to determine a character candidate with a lined line reliably and reliably.

【0009】

In the third invention, as shown in FIG. 3, in the first invention, the lined character discrimination processing means 12 has a predetermined range adjacent to a predetermined direction with respect to a character pattern that has been cut out and preprocessed. Character information extraction means 16 for extracting character information consisting of at least one of the number of pixels counted by adjacent projection that projects each pixel in the predetermined direction or a complex amount representing the complexity of a figure, and the character shape. Based on the character shape information extracted by the information extracting means 16, the discriminating means 17 for discriminating the simple lined character candidate or the complex lined character candidate, and at least the lineless line was removed from the simple lined character candidate. In this case, if the character can be identified as a character by the identification means 13, the character has a lineless character determination means 18 that determines the character as a simple lineless character.

【0010】

Here, the "character shape information" is information including at least one of "the number of pixels counted by adjacent projection" and "complex amount". "Adjacent in a predetermined direction" means to be adjacent in a direction determined depending on the fixed direction such as a horizontal line or a vertical line, and is usually a direction perpendicular to the fixed direction, but is not necessarily vertical. However, the direction may be close to that. In the embodiment described later, an example of the character pattern in the vertical axis direction is shown when the vanishing line is a horizontal line. Further, within a predetermined range means a range of pixels adjacent to each other in the predetermined direction with the pixel of interest sandwiched between them, and is, for example, a 3-dot line sandwiching the pixel of interest. If the "predetermined range" is narrowed, only lines close to horizontal lines or vertical lines can be recognized as simple vanishing lines, but if it is widened, even lines with a large deviation from horizontal lines or vertical lines can be recognized. It can be recognized as a simple vanishing line. However, if it is made too wide, a line that is not a line may be judged as a line, and it is judged appropriately by experience and experiment. "Adjacent projection" means that each pixel within a predetermined range adjacent to the predetermined direction is projected in the predetermined direction. The "complex amount" includes, for example, the maximum linear density in a predetermined direction, the Euler number, the black pixel density, and the like, as will be described later.

【0011】

In the present invention, the simple lined characters are discriminated based on the number of pixels obtained by adjacent projection, and the complex lined characters are discriminated based on the complicated amount. Since the number of pixels obtained by these complex quantities or adjacent projections can be obtained objectively, easily, at high speed, and with high reliability, it is easy to discriminate characters with lines, and it is performed at high speed and with high reliability. be able to.

【0012】

In the fourth invention, as shown in FIG. 4, in the second invention, the complex lined character discrimination processing means 14 includes a complex amount extracting means 19 for extracting a complex amount representing the complexity of a figure and an extraction means. It has a complex lined character discriminating means 20 for discriminating whether or not it is a complex lined character based on the complicated amount, and the simple lined character discriminating processing means 15 is adjacent to a predetermined direction. Based on the pixel histogram calculation means 21 that calculates the number of pixels counted by adjacent projection that projects each pixel in the range in the predetermined direction and the calculated number of pixels, the simple lined character candidate is discriminated. When the simple lined character candidate determining means 22 and the determined simple lined character candidate are removed from the line and can be identified as characters by the identifying means 13, they are treated as simple lined characters. It has a character fixing means 23 with a simple line to be fixed. According to the fourth invention, the erased characters are discriminated based on both the number of pixels obtained by the adjacent projection and the complicated amount. Therefore, in addition to the effect produced by the second invention, the delineated characters can be discriminated more reliably.

【0013】

In the fifth invention, as shown in FIG. 5, in the fourth invention, the simple lined character determination means 23 removes the line from the determined simple lined character candidate and becomes the identification means 13. The line-removing means 24 to be sent, the character candidate storage means 26 for storing the simple line-lined character candidates before the line-breaking is removed, and the simple line-lined character candidates from which the line-cutting has been removed can be identified as characters. In that case, each of the simple lined character candidates is determined as a simple lined character, and if it cannot be identified as a character, the simple lined character candidate stored in the character candidate storage means 26 is determined as a normal character. It has a character determination means 25 with a line to be sent to the identification means 13. According to the fifth invention, the unlined character candidates whose lines have been removed by the line removing means 24 before removal are temporarily stored in the character candidate storage means 26. Therefore, when the lined character candidate from which the line has been removed cannot be identified as a character, the lined character candidate is immediately read from the storage means 26 and sent to the identification means 13 to display the character. Since the identification is performed, the characters can be identified at high speed with a simple configuration.

【0014】

In the sixth invention, as shown in FIG. 6, an image is input, a character pattern is cut out and preprocessed from the input image (S1), and the character pattern cut out or preprocessed is subjected to. A character with a line that has either a simple line segment consisting of only line segments in a certain direction such as a horizontal line or a vertical line, which is supposed to represent the erasure of a character, or a complex line segment having a complicated line shape. It is to determine whether it is a normal character without a line segment (S2) and to identify a character for a normal character without a line segment (S3).

【0015】

In the seventh invention, as shown in FIG. 7, an image is input, a character pattern is cut out and preprocessed from the input image (S4), and the character pattern cut out or preprocessed is subjected to. When a complex lined character with a complicated line shape is discriminated (S5) and it is determined that the character is not a complex lined character, it is assumed to indicate the erasure of the character. Performs a process to determine whether a character has a simple line with a simple line consisting of only lines in a certain direction such as a horizontal line or a vertical line, or a normal character without a line (S6). , Character identification is performed for normal characters that are not lined (S7).

【0016】

In the eighth invention, as shown in FIG. 8, an image is input, a character pattern is cut out and preprocessed from the input image (S11), and the character cut out or preprocessed is complicated. Performs the process of discriminating a character with a complex line with a complex line (S12), and if it is determined that the character is not a character with a complex line, the character is cut out and preprocessed. (S13), a pixel histogram is calculated by adjacent projection that projects in the predetermined direction for each pixel within a predetermined range adjacent to the predetermined direction, and based on the pixel histogram, simple lined character candidates are determined. In (S14), if the line is removed from the simple lined character candidates, it is confirmed as a simple lined character if it can be identified as a character, and if it cannot be identified as a character, it is determined as a normal character. Then (S15), the characters are identified (S16) for the confirmed normal characters.

【0017】

As shown in FIG. 9, the ninth invention is determined as a simple lined character when the line is removed from the simple lined character candidate in the eighth invention and can be identified as a character. However, if it cannot be identified as a character, it is determined whether it is a simple lined character candidate (S151) when it is confirmed as a normal character (S151), and if it is a lined character candidate, it is lineless. (S152), character identification is performed for the lined character candidate whose line has been removed, and if it can be identified as a character, it is confirmed as a simple lined character and the character cannot be identified. Is confirmed as a normal character (S153, S154), and in the process of identifying the normal character (S16), the character is identified for the confirmed normal character (S161, S162), and if it is not identified. Outputs an error signal (S163), outputs the recognition result when it is identified (S164), and erases it in the case of a lined character (S165).

【0018】

BEST MODE FOR CARRYING OUT THE INVENTION

The character recognition device and the method thereof according to the embodiment of the present invention will be described below with reference to FIGS. 10 to 12. As shown in FIG. 10, the character reader according to the present embodiment optically reads the character pattern conveyed and written on the form, converts the character pattern into an electric signal, and digitizes the character pattern by binarization or the like. An optical reading unit 30 that outputs a signal, a conveying unit 31 that conveys a form to the optical reading field of the optical reading unit 30, a dictionary 32 that stores a standard pattern of characters, and a display that displays on the screen. It has an output unit 33 that prints out on a unit or a recording paper, an operation unit 34 that the user performs various operations, and a CPU and a memory 35 having various functions for character recognition.

【0019】

Further, as shown in the figure, the CPU and the memory 35 have a character cutting / preprocessing unit 36 that cuts out and preprocesses characters from the input image, and a complicated line disappearing having a complicated linear shape. If the complex lined character discrimination processing unit 37 that performs discrimination processing on the complicated lined characters and the complex lined character discrimination processing unit 37 determine that the characters are not complex lined characters, the characters are displayed. With the simple lined character discrimination processing unit 38 that performs discrimination processing for simple lined character candidates with simple lined lines consisting only of lines in a certain direction such as horizontal lines or vertical lines that are supposed to represent erasure. , Identification unit 39 that identifies characters for erased characters or normal characters, erases the characters if they are determined to be erased characters, and identifies the characters if they are determined to be normal characters. It has a result output instruction unit 40 for instructing an output to the effect of an error when the character is not identified even though the result is determined to be a normal character.

【0020】

Further, as shown in FIG. 10, the character cutting / preprocessing unit 36 is cut out as a character cutting unit 41 that cuts out only a character signal (character pattern) from the image data of the input digital image including noise. It has a preprocessing unit 42 that removes noise from the character signal and performs regularity such as the position and size of the character. Here, the optical reading unit 30, the transport unit 31, and the character cutting / preprocessing unit 36 correspond to the image input processing unit.

【0021】

Further, the complex lined character discrimination processing unit 37 determines whether or not the character is a complex lined character based on the complex quantity extraction unit 43 that extracts the complex amount from the character pattern and the extracted complex amount. It has a character discriminating unit 44 with a line. Here, the "complex amount" includes, for example, a linear density such as a maximum linear density or an average linear density in a predetermined direction, a Euler number, a black pixel density, or the like. The "linear density in a predetermined direction" is a value when an image in a rectangle is scanned along a predetermined fixed direction and a portion where a white pixel changes to a black pixel (or a black pixel to a white pixel) is counted. Say. As an example, FIG. 11B shows an example in which the maximum density in the vertical direction is 3. Here, the "predetermined direction" is usually a direction perpendicular to a certain direction of a line segment assumed as a simple vanishing line.

【0022】

"Euler number" E is the number of connected components H subtracted from the number of connected components C, where C is the number of connected components connected to each other in the image and H is the number of holes in the image. , E = CH. As an example, FIG. 11 (c) shows an example of the Euler number E = 1. "Black (white) pixel density" D is the ratio of the area B (number of black (white) pixels) of the image of interest to the area S (total number of black (white) pixels) of the circumscribing rectangle of the image of interest. D = S / B.

【0023】

As an example, FIG. 11 (d) shows an example of the contents. Note that FIG. 11A shows an example of a character with a complicated line. The complex lined character discriminating unit 44 determines whether or not the character is a complex lined character by appropriately combining individual complex quantities such as the maximum linear density, the Euler number, or the black pixel density whose features have been extracted, or a combination thereof. It is done based on things. The general tendency of the extracted complex amount and the normal character or the lined character is as follows.

【0024】

[table 1]

<img file="JPP3345246B2_D0001.tif" />【0025】

For these complicated quantities, a dictionary or a discriminant function is obtained in advance by an existing method, and it is discriminated whether or not the characters are lined. As shown in FIG. 10, the simple lined character discrimination processing unit 38 calculates the number of pixels counted by the adjacent projection that projects in the predetermined direction for each pixel within the adjacent predetermined range with respect to the predetermined direction. The pixel histogram calculation unit 45, the simple lined character candidate determination unit 46 that discriminates the simple lined character candidate based on the calculated number of pixels, and the discriminated simple lined character candidate are lineless. If the character can be identified as a character by the identification unit 39 when the above is removed, the character determination unit 47 with a simple line disappearing is provided. Here, "adjacent projection" means that each pixel in a predetermined range (n lines) adjacent to a predetermined direction is projected in the predetermined direction (the number of pixels is added). In the present embodiment, when the simple vanishing line is a horizontal line, the predetermined direction is the vertical direction, and the predetermined range is three lines.

【0026】

Figures 12 (a) and 12 (b) show a description of the adjacent projection. In the figure, an n-line black pixel projection histogram that counts the number of pixels in the adjacent n-lines (3 lines in FIG. 12) is created for the simple disappearance of horizontal lines. That is, the sum of the number of pixels in the direction along the horizontal line, which is a simple disappearance for every three lines in FIG. 12 (a), is shown in FIG. 12 (b). When the histogram value exceeds a predetermined threshold value, the simple lined character candidate determination unit 46 determines the corresponding portion as a simple lined character candidate having a horizontal line. As shown in FIG. 12 (c), if the simple lined character candidate determination unit 46 determines as a lineless character candidate and the line removal unit 50 removes the lineless character, it can be identified as a character. , Simple lined character determination unit 48 determines as a lined character (upper example), and if it cannot be identified as a character, it is determined as a normal character (non-lined character) (lower example). ..

【0027】

Further, as shown in FIG. 10, the simple lined character confirmation unit 47 is discriminated from the lineless character removing unit 50 that removes the lineless line from the determined simple lined character candidate and sends it to the identification unit 39. If the simple lined character candidates are recognized as characters, each of the simple lined character candidates is determined as a simple lined character, and if it is not recognized as a character, it is determined as a normal character. It has a lined character determination unit 48 to be sent to the identification unit 39, and a character candidate storage unit 49 for storing the lined character candidates removed by the lineless removal unit 50. Further, the identification unit 39 extracts the feature value of each character pattern and collates the feature extraction unit 51 that compresses the information with the standard pattern of each character, that is, the dictionary of the feature value for each character type. It has a collating unit 52.

【0028】

Subsequently, the operation of the character recognition device (OCR) and the character recognition method according to the embodiment of the present invention will be described with reference to FIGS. 13 and 14. As shown in FIG. 13, in step SJ1, the transport unit 31 conveys, for example, a form in which characters are written to the reading field of view of the optical reading unit 30, and causes the OCR to input the form.

【0029】

In step SJ2, the optical constant reading unit 30 converts the character pattern on the form into an electric signal by photoelectric conversion, and outputs it as a digital signal by binarization or the like. Here, on the form, for example, the written characters and a character frame having the same color as the characters may be provided. In step SJ3, the character cutting unit 41 separates and cuts out only the character signal from the digital signal containing noise into characters one by one, and the preprocessing unit 42 normalizes the position, size, and inclination line width of the characters. Etc. are performed. In step SJ4, the complex lined character discrimination processing unit 37 performs a complex lined character discrimination process, and in step SJ5, rejects the complex lined character.

【0030】

FIG. 14A shows steps SJ4 and SJ5, which are the processing contents of the complicated lined character discrimination unit 37. In step SJ41, the complex quantity extraction unit 43 extracts a complex quantity such as the Euler number, the linear density, and the black pixel density in order to determine whether or not the character has a complex line. Then, in step SJ42, the complex lined character determination unit 44 uses all or part of the complicated amount and Table 1 described above to determine whether the character pattern of interest is a normal character or a complex lined character. To determine.

【0031】

For example, if the maximum linear density is small, the absolute value of the Euler number is small, and the black pixel density is small, it is determined that the character is not a complex lined character, the maximum linear density is large, the absolute value of the Euler number is negative and large, and the black pixel. If the density is high, it is determined to be a complex lined character. In addition, when two or more of these complex quantities show a tendency of complex lined characters, it may be determined that they are complex lined characters, and one of each complex amount. Only one may be selected as a criterion for discrimination, two may be selected as a criterion, and a priority may be set for each complex amount to discriminate according to the priority.

【0032】

If it is determined in step SJ43 that it is a character with a complicated line, it is output as reject in step SJ5 as described above. If it is determined that the character is not a complex lineless character, the process proceeds to step SJ6, and a horizontal line, that is, a process for determining the presence or absence of a simple line disappearance is performed by the simple lined character determination processing unit 38, as shown in FIG. , If it is determined to be a lined character, it is rejected (rejected) in step SJ7.

【0033】

The discrimination process is shown in detail in FIG. 14B. As shown in the figure, in step SJ61, the pixel histogram calculation unit 45 creates an adjacent projection histogram. As shown in FIGS. 12 (a) and 12 (b), the adjacent histogram is, for example, adding one line at a time along the horizontal line of black pixels every n = 3 lines in the vertical direction perpendicular to the horizontal line which is a simple disappearance. It is obtained by shifting it. As a result, the slope of the line disappearance is absorbed, and even if the horizontal line, which is a simple line disappear, shifts to some extent, it can be determined as a simple line disappear.

【0034】

In step SJ62, when the histogram value exceeds the threshold value N, the simple lined character candidate determination unit 46 determines that there is a line disappearance in step SJ63, determines that the character candidate has a simple lineless line, and sets the threshold value N. If it does not exceed, it is determined that there is no line disappearance, and it is determined as a normal character candidate. If the simple lined character candidate determination unit 46 determines in step SJ63 as a lined character candidate, the process proceeds to step SJ64, and the lined character candidate (before removing the lineless line) is stored in the character candidate. While storing in the unit 49, the line-removing unit 50 detects and deletes a horizontal line which is a simple line of a character candidate with a line. For the removal of the vanishing line, an existing method such as a line segment extraction method using an n-line run length or the like is used.

【0035】

The character pattern from which the erased lines have been removed proceeds to step SJ8, feature extraction is performed by the identification unit 39, and dictionary collation is performed based on the extracted features in step SJ9. The dictionary collation calculates the degree of matching with the standard pattern prepared in advance, and outputs the category of the standard pattern having the highest similarity as the recognition result. If it is determined in step SJ65 that the collation result is rejected, in step SJ66, the simple lined character determination unit 48 determines that the lined character candidate is a non-lined pair character. In step SJ67, the lined character candidates once stored in the character candidate storage unit 49 are sent to the identification unit 39 to identify the characters, and the result output instruction unit 40 outputs the identification result to the output unit. Instruct to output from 33.

【0036】

If it is determined in step SJ65 that the character has been identified as a result of collation, the process proceeds to step SJ68, and the simple lined character determination unit 48 determines the lined character candidate as the lined character. , In step SJ69, the result output instruction unit 40 instructs the output unit 33 with the identification result as reject. The upper part of Fig. 12 (c) shows an example in which a lineless character is recognized as a simple lined character candidate, the lineless character is identified as "5", and the character is determined to be a lined character. In the lower row, an example in which the original simple lined character candidate was determined to be a normal character because the line was recognized from the simple lined character candidate and the one with the line removed was rejected. Shown.

【0037】

As described above, according to the present embodiment, the complex lined characters are determined based on all of the maximum line density in a predetermined direction, the Euler number, and the black pixel line density as a complicated amount. ing. Therefore, it is possible to discriminate characters with complex lines with high reliability and high speed. Further, according to the present embodiment, the simple lined character candidates are discriminated by counting the number of pixels by adjacent projection. Therefore, it is possible to easily, reliably, and quickly discriminate not only the assumed simple disappearance of the horizontal line but also the character candidate including the line segment deviated from the horizontal line with a simple configuration. As shown above, according to the present embodiment, the misreading of the delineated characters is reduced by performing the delineation presence / absence determination process. In addition, the characters erased by the delineation are more reliably rejected. Further, in the example of the above embodiment, the case of a horizontal line or a line slightly deviated from the horizontal line has been described as a simple vanishing line, but the case is not limited to this case, and the line is a vertical line, a diagonal line, or a × line. Is also good. Further, all of these possible lines can be discriminated as simple line disappearances, and more reliable character recognition with less leakage can be performed.

【0038】

[Effect of the invention]

According to the first invention or the sixth invention, it is determined whether or not the character is a lined character with a simple line or a complex line, or a normal character without a line, and the normal character is a character. I try to identify them. Therefore, even if the user does not have the specified knowledge of derailment and does not give the user instructions about derailment in advance, all the characters with the line clearly recognized as derailment by the user are not omitted. Since it is removed, it is possible to perform reliable and reliable character recognition with a simple configuration without imposing a burden on the user. As a result, restrictions on the entry of forms and the like can be reduced, and the burden on the user can be reduced.

【0039】

In the second invention or the seventh invention, when performing the discrimination processing of the characters with complex lines, first, the discrimination processing of the characters with complex lines is performed, and only when it is determined that the characters are not the characters with complex lines. The simple lined character candidates are discriminated. Therefore, the determination can be performed efficiently and quickly. Further, regarding the simple lined character candidate, if the character from which the line has been removed can be identified as a character, the character is confirmed as a simple lined character, and the character that cannot be identified as a character is determined as a normal character. .. Therefore, since the complicated delineation has already been eliminated, it is possible to prevent the characters with the complex delineation from being confused with the normal characters and being misread. In addition, even if the character is determined to be a simple lined character candidate, it is determined whether or not the character can be identified as a character, so that whether or not the character is a lineless character can be determined more reliably and reliably. be able to.

【0040】

In the third invention, the fourth invention, the eighth invention, or the ninth invention, a complex amount is extracted or adjacent projection is performed on a simple lined character or a complex lined character. I try to distinguish. Therefore, with a simple configuration, discrimination is easy, and it can be performed quickly and reliably. In the fifth invention, when determining a lined character or a normal character, the lined character candidate before removing the line is temporarily stored, and the character candidate from which the line has been removed can be identified as a character. If not, the stored lined characters are used as normal characters to identify the characters. Therefore, in addition to the effects produced by the fourth invention, character recognition can be performed with a simple configuration and at high speed.

[Simple explanation of drawings]

[Figure 1]

Principle block diagram of the first invention [Figure 2]

Principle block diagram of the second invention [Fig. 3]

Principle block diagram of the third invention [Fig. 4]

Principle block diagram of the fourth invention [Fig. 5]

Principle block diagram of the fifth invention [Fig. 6]

Principle flow diagram of the sixth invention [Fig. 7]

Principle flow diagram of the seventh invention [Fig. 8]

Principle flow diagram of the eighth invention [Fig. 9]

Principle flow diagram of the ninth invention [Fig. 10]

Block diagram according to the embodiment [Fig. 11]

Explanatory drawing of complex quantity concerning embodiment [Fig. 12]

Explanatory drawing of adjacent projection according to embodiment [Fig. 13]

Flow chart according to the embodiment (1) [Fig. 14]

Flow chart according to the embodiment (2) [Fig. 15]

A block diagram and a diagram showing a misrecognition example according to a conventional example. [Explanation of symbols]

11 ... Image input processing means 12 ... Character discrimination processing means with erased lines 13 ... Identification means 14 ... Character discrimination processing means with complicated lines 15 ... Simple lined character discrimination processing means 16 ... Glyph information extraction means 17 ... Means for determining character candidates with lines 18 ... Character confirmation means with line 19 ... Complex quantity extraction means 20 (37) ... Character discrimination means with complex line (Character discrimination processing unit with complex line) 21 (45) ... Pixel histogram calculation means (pixel histogram calculation unit) 22 (46) ... Simple lined character candidate determination means (simple lined character candidate determination unit) 23 (47) ... Simple lined character confirmation means (simple lined character confirmation part) 24 (50) ... Line removal means (line removal part) 25 (48) ... Character determination means with line (simple character determination unit with line)