JPH0498477A - Character segmenting method - Google Patents

Abstract

PURPOSE:To precisely execute the segmenting of a character by discriminating the form of the input character automatically even when an operator does not set the longitudinal/lateral ratio of the input chracter and the like by discriminating the form of the input character statistically. CONSTITUTION:Individual character string width H is found with projecting plural character strings in a column direction respectively and an individual input character is projected in the column direction and in a width direction orthogonal to the column direction respectively. Then, individual character blocks 13A, 13B, 13C... are found and length with the largest frequency between length (w) and (h) of the column direction and the width direction of the individ ual block is regarded as the length of the column direction and the width direc tion of a character pattern 15 to represent the input character. Moreover, a pattern similar to the character pattern 15 is defined as a basic character frame 16 and the segmenting of the input character is executed with the basic charac ter frame 16 as a unit. Thus, the segmenting of a separated character etc., can be executed precisely by using the form of the discriminated character without depending on a character pitch.

Description

[Detailed Description of the Invention] [Second Industrial Application Fields] The present invention is directed to a character extraction method applied to a character recognition device suitable for use, for example, in recognizing characters in a printed document and converting them into character codes. Two matters.

[Summary of the Invention] The present invention relates to a character cutting method that is applied to a character recognition device suitable for use when recognizing characters in a printed document and converting them into character codes, for example. The width of each character string is determined by projecting each of the input characters in the plurality of character strings in the column direction and the width direction orthogonal to the column direction to determine each character block. The length with the highest frequency among the lengths in the column direction and width direction of the character block is regarded as the length in the column direction and width direction of the character pattern representing the above input character, and the length of the above individual character string width distribution is calculated. A basic character frame is a pattern similar to a character pattern representative of the input characters, with the width of the character string that has the highest frequency as one side, and the operator cuts out the input characters using the basic character frame as a unit. The shape of the input character is automatically determined and the character can be accurately cut out without setting the aspect ratio of the input character.

In addition, the present invention calculates the width of each character string by projecting each of the plurality of character strings in the column direction.1. Obtain individual character blocks by projecting them in the direction, and calculate the most frequent length of each character block in the column direction and width direction of the character pattern representing the input character. A window is created in a predetermined width area centered on the length in the column direction of the character pattern representing the input character, starting from the rising edge of the first character in the character string to be recognized among the multiple character strings. , project the character string to be distributed in the width direction within the window, find the valley position in the column direction where the projected value is the minimum, and calculate the projected value of the valley position and the window above. If the maximum value of the gradient with respect to other projection values in , if the maximum value of the gradient between the projection value at the trough position and the other projection values in the window is less than the predetermined value, then further from the trough position in the column direction of the character pattern representing the input character. By setting a new window in an area of a predetermined width centered around the length of , it is possible to accurately cut out individual characters even from connected character blocks that cannot be separated by projection.

Further, the present invention projects each of a plurality of character strings in the column direction to determine the width of each character string, and projects each input character in the plurality of character strings in the column direction and in the width direction perpendicular to the column direction. Obtain individual character blocks by projecting them onto The character string width with the highest frequency among the character string width distribution of the above values is regarded as the length in the direction, and the basic character frame is a pattern similar to the character pattern representing the input characters above, and the above multiple characters are The character blocks that fall within the above basic character frame are integrated and cut out based on the first character block belonging to the character string to be recognized in the string, and the character blocks next to the above cut out character block are then used as the reference. By integrating and cutting out the character blocks that fit within the basic character frame, it is possible to accurately cut out characters even if the character pitch fluctuates or even if the characters are separated.

[Prior Art] For example, in order to automate the process in which a worker picks up type in letterpress printing, a character recognition device is required that recognizes each character in a manuscript created by type printing and converts it into a character code. .

FIG. 22 shows an example of a conventional character recognition device.
In Fig. 2, (1) is a document reading section, and from this document reading section (1), an original character signal S1 corresponding to the shading of one page of the document is supplied to a character string cutting section (2). . This original character signal S1 is obtained by dividing the document into dots at a predetermined density, and represents black dots as high level "1" and white dots as low level "0".The density of each dot is expressed as a multi-bit binary number. Sometimes it is expressed as

The character string extraction unit (2) is composed of a first stage preprocessing unit (3), a second stage preprocessing unit (4), and a third stage preprocessing unit (5). The preprocessing unit (3) removes noise and corrects the rotation of the document, and the second stage preprocessing unit (4) separates the character area AR (see 23rd TyJ) from other areas (areas for photographs, drawings, etc.). Then, only the image data included in the character area AR is extracted, and in the third stage preprocessing section (5), the character strings A R1, A R2, . . . included in the extracted character area AR are extracted.
A character string signal S4 corresponding to is extracted.

In order to extract this character string signal S4, as shown in FIG. )
Expressed in coordinates, the Y projection YP is generated by projecting the "1" or "0" value of each dot onto the Y axis and calculating the sum. Then, when the second Y projection YP is binarized at a predetermined dark value level, the sections of high level "1" in this binarized signal become character strings A R1, A R2, etc. The character string signal S4 corresponds to the character string signal S4 at the subsequent character cutting part (
6).

In the character cutting section (6), for example, the character string signal s4 of the first character string ARi shown in FIG. 24A is projected onto the X axis to generate an X projection XP, and this A coarse cutout signal DTI (Fig. 24C) is obtained by binarizing with the H value THI having a value of 2), and this X projection XP is binarized with a medium level threshold TH2 (Fig. 24D). The fine cutting signal DT2 (Fig. 24E
). Similarly, the coarse cutting signal DTI is at high level “1”.
By individually generating the Y projection YP only in the section ``, it is possible to generate a cutout signal in the Y direction.Finally, as shown in FIG. 24A, for example, for the character At the same time, the high-level "1" becomes "1" inside the circumscribing frame (9) circumscribing the , and the separating character "1-ii" becomes a high level "1" inside the circumscribing frame (9) circumscribing the separated parts.
), (12> A cutout signal with a high level "1"' is obtained inside, and the basic signal is a signal in which only the parts where the cutout signal becomes a high level "1" are sequentially cut out from the input character string signal S4. This becomes a rectangular cut-out character signal S7.

Incidentally, the fine cutting signal DT2 in FIG. 24E is used when examining the finer structure of each character. Also, Figure 24A
For "i", which is a separating character, use the circumscribed frame (11),
Since there are two (12), it is necessary to integrate them later at the character identification stage.

Reference numeral (7) indicates a character recognition section, and this character recognition section (7) takes in the basic rectangular cut-out character signal s7 for each circumscribed frame and performs character recognition. Specifically, first, classification is performed by position, and characters (r''', V'''' that exist in the upper half of the character example ARi in FIG. 24A) are classified.

(such as "°") and characters present in the bottom half (such as ".").

r, J, r, J, etc.) as the first characteristic character, and perform pattern matching to find the corresponding character code (J I
S code, etc.). If identification is not possible with this, the width of the circumscribing frame is W, the height is h, and the aspect ratio is h.
/w and relative size. That is, classification is performed depending on whether the aspect ratio h/w falls within the range of CI<h/w<0.5 or 1.5<h/w. Furthermore, the width of the average-sized circumscribing frame is WR, and the height is h.
As R, the values of the vertical relative ratio h/hR and the horizontal relative ratio W/WR are 0<h/hR<0.5 and 0<w/wR<0.5, respectively.
Classification is performed based on whether or not it falls within the range of . Pattern matching is performed using characters within the above range as second characteristic characters.

Furthermore, for characters that are not classified as the first or second characteristic characters, pattern matching is performed with individually stored dot patterns, and if a predetermined degree of matching is obtained, that character code is assigned. .

If there are still characters that cannot be used with KIJ Wi,
The operations of re-cutting the circumscribing frame to further separate it into a plurality of minute circumscribing frames and integrating the circumscribing frame with the subsequent circumscribing frame are executed. Note that, if an unrecognizable 5) character ultimately remains, a reject code is given to that character to indicate that it is an unrecognizable character.

The character code for one page of the document generated by the character identification section (7) is stored in a predetermined storage device along with information indicating the position and size of the character. Furthermore, the video signal of the character corresponding to the character code is displayed on a display unit (8) such as a cathode ray tube so that the operator can judge whether the recognition result is correct or not.
Some characters as the recognition result are displayed on the display screen of the display unit (8) in a format corresponding to the document. In this case, a high-intensity rectangular blank is displayed in the portion of the character that cannot be recognized. Therefore, if a character to be corrected or a character that cannot be recognized exists, the operator can input the desired character into that part in the same way as a word processor.

When the individual characters are completely separated as in the example in Figure 24, the probability of identifying the characters is relatively high except for separated characters such as ``i'', but the pitch of the characters, such as in a newspaper article, is For character strings that are so short that individual characters cannot be completely separated by projection (hereinafter referred to as a "concatenated character block"), a special method is required to cut out the characters.

Conventionally, such special methods include a method in which an operator sets a character pitch and cuts out characters according to this character pitch, a method in which characters are cut out assuming the character's aspect ratio is approximately 1, and A method is known in which a standard size of printed characters is given in advance. , Journal of the Institute of Electronics and Communication Engineers' 85/8. Vol.

J68-D, No. 8. pp', 1497-1
504, there is a method of estimating character pitch (including cases where the character pitch changes) using a linear squared error function and two parameters, and cutting out characters using this estimated character pitch. Disclosed. Incidentally, when the character pitch changes, it may occur not only due to so-called variable pitch in English text but also in Japanese text due to so-called fill-in or expulsion operation during typesetting.

[The problem that Hatsuko is trying to solve]However, the method of setting the character aspect ratio in advance does not allow for character patterns (character fonts) with different aspect ratios.
Incorrect cutting pressure may be applied to characters that use
There is a possibility that an incorrect combination of characters may be integrated. Specifically, for example, if the character aspect ratio is set to 1 for full-width characters, the aspect ratio used in newspapers will be 0.76 to 0.7.
It becomes difficult to accurately cut out the character 9. In this regard, even if the aspect ratio of the characters is set in advance, there is a disadvantage that it is difficult for the operator to accurately measure the aspect ratio of the characters of the document to be recognized.

Further, the method of estimating the character pitch using predetermined parameters has the disadvantage that the operator's operations are complicated and the determination method and processing contents are complicated.

Furthermore, conventional methods have had the disadvantage that the probability of recognition of separated characters such as "i" and "san" is particularly low.

In view of the above, an object of the present invention is to enable automatic determination of the size and shape of characters of a document to be recognized in a character recognition process in a character recognition device.

Another object of the present invention is to make it possible to accurately cut out characters even from connected character blocks using the determined character shapes.

A further object of the present invention is to use the determined character shape to accurately cut out separated characters, etc., without depending on the character pitch.

[Means for Solving the Problems] A first character cutting method according to the present invention projects a plurality of character strings in the column direction and calculates the width H of each character string, as shown in FIGS. 3 and 4, for example. By projecting each input character in the plurality of character strings in the column direction and the width direction perpendicular to the column direction,
, 13B, 13C,...), and the most frequent length w and h in the column direction and width direction of these individual character blocks is determined as the character pattern (15 ) is regarded as the length in the column direction and width direction, and a pattern similar to the character pattern (15) representing the input characters above, with the character string width having the highest frequency among the distributions of the character string widths H as one side. Let be the basic character frame (16),
The input characters are cut out using the basic character frame (16) as a unit.

Further, the second character cutting method according to the present invention, for example, as shown in FIG. Each character block is calculated by projecting the input characters in the column direction and the width direction orthogonal to the column direction, respectively, and the length that has the highest frequency among the lengths of these individual character blocks in the column direction and the width direction is calculated. The length is regarded as the length in the column direction and the width direction of the character pattern (17) representing the input characters, and for example, as shown in FIG. A window (21) is set in an area of a predetermined width centering on the length in the column direction of the character pattern (17) representing the input characters, and the character string to be recognized is displayed in the width direction within the window. Valley position 1 in the column direction where the projected value (22) is the minimum. ,,. is calculated, and if the maximum value of the gradient between the projection value at the valley position and other projection values in the window exceeds a predetermined value, the character is cut out at the valley position, and the character is cut out. The next window (25) is set based on the position where the above-mentioned trough position has been performed, and if the maximum value of the slope between the projection value at the valley position and the other projection values within the above-mentioned window is below the above-mentioned predetermined value, the above-mentioned A new window is set in an area of a predetermined width further from the trough position and centered on the length in the column direction of the character pattern representative of the input characters.

Further, a third character cutting method according to the present invention, for example, as shown in FIG. Obtain individual character blocks by projecting the input characters in the column direction and the width direction perpendicular to the column direction, and calculate the most frequent length among the lengths of these individual character blocks in the column direction and width direction. is considered to be the length of the character pattern (17) in the column direction and width direction that represents the above input character, and the character string width with the highest frequency among the above examples of character string width distribution is taken as one side to represent the input character. character pattern (1
7) is used as the basic character frame (18), and as shown in FIGS. 12 and 14, the first character block (19A ) based on the basic character frame (18
), and then cut out the character blocks that fall within the basic character frame (18), using the next character block of the cut out character block as a reference. This is what I did.

Further, in the fourth character extraction method according to the present invention, in the third character extraction method, when the plurality of character strings are written vertically, the character string to be recognized among the plurality of character strings is The character blocks to which they belong are ranked based on their vertical coordinates.

Further, the fifth character cutting method according to the present invention is the third
In the character extraction method, if the multiple character strings are written horizontally, the character blocks that belong to the character string to be recognized among the multiple character strings are ranked based on the horizontal coordinates. It is something.

[Operation: According to the first character extraction method, the character pattern whose length is most frequent among the lengths in the vertical and horizontal directions of the character blocks surrounding each input character is representative of the input characters. A pattern similar to pattern (15), which is considered to be the length in the column direction and width direction of (15), and represents the input character with the most frequent character string width as one side in the distribution of each character string width H. is the basic character frame (16).

Therefore, the shape of the character pattern (15) representing the input character is automatically determined without the operator specifying the shape of the input character. In this case, since the length with the highest frequency is adopted, the determination of the shape of the input character is not influenced by special characters (such as ruby characters) included in the input character.

Furthermore, by cutting the input character using the basic character frame (16) as a unit, the input character can be accurately cut out regardless of the shape of the input character.

Further, according to the second character extraction method, a window (21) is set in an area of a predetermined width from the rising edge of the first character to the length of the character pattern representing the input character in the column direction. , project the character string to be recognized in this window in the width direction and calculate the projection value (22)
The valley position 1 sin in the column direction at which the minimum Characters are cut out at the copy position. In this case, even if it is a connected character block, the boundary between each character is generally concave in the shape of a valley, so for example, if the input characters form a connected character block, the projection value of the boundary between individual characters will not be 0. Even in such a case, as long as the projected value of the boundary between the characters is a minimum value with a relatively large depression, the characters in the connected character block can be accurately extracted.

Further, according to the third character extraction method, when a separated character exists in a character string to be recognized, the basic character frame (18) is set based on the upper character block of the separated character. Since the character blocks inside are cut out together, the separated characters are also cut out accurately. In this case, since characters are cut out using the basic character frame (18) as a reference, even if the pitch of the input character changes, the separated characters can be cut out accurately.

Further, according to the fourth character extraction method, when the input character string is written vertically, the character block that fits into the basic character frame is determined based on the character block with the smaller coordinate value along the vertical coordinate. It is integrated and cut out.

Furthermore, according to the fifth character extraction method, when the input character string is written horizontally, the character blocks that fit into the basic character frame are integrated along the horizontal coordinates using the character block with the smaller coordinate value as a reference. and then cut out.

! Embodiments Hereinafter, embodiments of the present invention will be explained with reference to FIGS. 1 to 21. In this example, the present invention is applied to a character extraction method used in a character recognition device that recognizes each character in a character string of a printed document and outputs a character code.

The overall configuration of the character recognition device is the same as the conventional example shown in FIG.
The operation of (6) is different. The operations of the cutout sections (2) and (6) in this example will be explained below.

In this example, the aspect ratio of the input character can be automatically determined, but first, the operation of the character cutting section and the like when determining the aspect ratio will be explained in the case where the character string to be recognized is written horizontally.

The English text in Figure 1A shows an example of multiple character strings written horizontally (more precisely, ``mouth character lines''), and this character string is written in the column direction (
After projecting in the X direction) to obtain the y projection YP, this y projection YP is binarized with a threshold value TH3 to obtain the character string height H
(Width in the X direction perpendicular to the X direction) is determined. If each character is decomposed into several tens of dots x several tens of dots and the resolution of the y projection YP is one dot, the threshold TH3 is set to, for example, 1, which is the resolution. And the character string height H is 1
If we ignore the portions where the character string height H is 2 or more, each portion is detected as a character string. Each detected character string is projected in the X direction as shown in FIG. 1 to obtain the X projection XP.

Figure 2 shows an enlarged view of the X projection XP of one character string in horizontally written Japanese text. horizontally (
As the length W (in the X direction), the horizontal lengths of other individual characters are also determined.

In addition, the length of the area where the character "hand" is projected in the X direction and this projection exceeds a predetermined threshold is defined as the length h in the vertical direction (X direction), and the length of the other individual characters in the vertical direction is also Find the length. Rectangular block (13) with height h x width W
is called the character block of that character.

Figure 3A shows character blocks (13A), (13B), (13C),
Figure 3B shows the results of extracting (130),...
is a character block (14A) from the character string ARj of English character string height H'.

The results of extracting (14th), (14C), (140), etc. are shown.

In FIG. 3A, separate characters such as "・" and "i" are each separated into a plurality of character blocks. In this example, first, statistics on the distribution of the character string height H are taken. Specifically, the height H ranges from 2 to 9, 10-19.20 to 29
．． ..., and calculate the frequency of each group and the average value within each group.

This is the same as creating a histogram of the character string height H. Then, the average value HB of the most frequent part with the highest frequency is determined. As shown in FIG. 4, the average value HB of the most frequent part becomes the height of the character string that appears most frequently.

The aspect ratio of each character block is expressed as h/w, but in this example, the aspect ratio of all cut out character blocks is 0.75≦h/w≦1. 25...(1
), and the distribution of the vertical length (height) h and the horizontal length (width) W of these selected character blocks is taken.

This means examining the distribution of characters that have character blocks that are approximately similar to character blocks of full-width characters with an aspect ratio of 1, and half-width characters whose aspect ratio is in the range of 0.5≦h/w≦0.7. It means to exclude such things. Therefore, for example, for ruby characters whose height and width are half the height and width of full-width characters (half-width ruby) and three-minute ruby, etc., the formula (1
) are subject to statistics as long as they satisfy the following criteria. Specifically, the length h in the vertical direction and the length W in the horizontal direction are respectively θ~9.10~1
9.20-29. ..., and check the frequency and average value of each group, and calculate the average values hb and wb of the most frequent group with the largest frequency for lengths h and W, respectively.
demand.

However, if the input document is a document mainly composed of half-width characters, such as English, the range of aspect ratios for which statistics are taken may center on half-width characters.

As shown in FIG. 4, in this example, the average values hb and wb
is the average frame of the input character pattern (input font) (
15) length in the vertical direction and length in the horizontal direction. Then, a block obtained by cutting out an area similar to the average frame (15) of the input font from the HB character string whose character string height is the average value of the most frequent part is defined as a basic character frame (16).

That is, assuming that the horizontal length of the basic character frame (16) is WB, this WB can be obtained from the following equation.

WB=HB-wb/h b - (2) Average frame (15) and basic character frame (16) of these input fonts
) is used to cut out connected character blocks, etc., as described later. Since these basic character frames and the like are automatically set according to the input characters, it is possible to always perform optimal character extraction without depending on the character size of the input document. In this case, ruby characters (for example, half-ruby) are generally 1/2 the vertical and horizontal lengths of ruby-covered characters such as full-width characters, so the basic character frame for that ruby character should be the height. and width is 1 of the basic character frame (16) of ruby characters respectively.
/2 is used.

On the other hand, if the input document is written vertically as shown in Figure 5,
By taking statistics on the frequency distribution for the width W of each character string, the average value WB of the most frequent part can be obtained.

Then, the lengths of each character block in the vertical and horizontal directions under the condition of formula (1),
Statistics on W are taken to obtain the lengths hb and wb of the average value of the most frequent part. The average frame of the input font in this case of vertical writing is (1
7), and a block similar to the average frame (17) of the input font with the average value WB of the most frequent part of the character string width as the length of one side becomes the basic character frame (18). The height HB of this basic character frame (18) can be determined by the following formula.

HB=WB-h b/wb (3) As mentioned above, according to this example, statistics are taken on the vertical and horizontal lengths of each character block, w, and the average value of the most frequent part is calculated. Since the length of is set to the vertical and horizontal length of the average frame of the input font, the size of the input character is the same as the normal input font size without being affected by the patterns of ruby characters and separated characters. There is an advantage in being able to accurately detect In this case, there is no need for the operator to set parameters from outside, so the burden on the operator is reduced.

To explain the results of actually taking statistics on the frequency distribution when the input document is vertical writing, Figure 6 shows the frequency distribution of the width W of all character strings, and Figure 7 and Figure 8 shows the frequency distribution of the width W and height h of each individual character block. From Figure 6, the most frequent part of the character string width W (40≦W≦4
9) is 47, so the basic character frame (18)
(See FIG. 5) has a width WB of 47. Also, from FIG. 7 and FIG. 8, the most frequent part of the width W of the character block (40≦W≦
The average value of 49) is 43, and the most frequent part of height h (30≦
Since the average value of h≦39) is 33, the width wb and height hb of the average frame (17) of the input font are 43 and 3, respectively.
It becomes 3.

Therefore, the aspect ratio hb/wb is 0.7674, and the formula (
From 3), the height HB of the basic character frame (18) is HB = 47·33/43 = 36.

Also, to explain the results of actually calculating frequency distribution statistics when the input document is Shinheike Monogatari (vertical writing),
Figure 9 shows the frequency distribution of the width W of all character strings, and Figures 10 and 11 show the width W and height h of individual character blocks, respectively.
shows the frequency distribution of From Figure 9, the average value of the most frequent part of the character string width W is 41, and from Figures 10 and 11, the average value of the most frequent part of the character block width W is 35, and the most frequent part of the height h. The average value of is 35. Therefore, the aspect ratio of the average frame and basic character frame of the input font is 1.

Next, an example of a method for cutting out characters using the basic character frame described above will be described with reference to FIGS. 12 to 17. In this example, the characters of the input characters are not necessarily constant, but the basic character frame is constant, and if it is a separated character, the method is to cut out the character by integrating each separated character block. be.

FIG. 12 shows an example of a character block of vertically written input characters to be cut out, and in this FIG. 12, (19A
), (19B), . . . are character blocks of ruby characters such as full-width characters, and (27) is a character block of ruby characters. Using the average value of the most frequent part of the frequency distribution of the horizontal length W and vertical length h of these character blocks and the average value of the most frequent part of the width of each character string, the results shown in Figure 5 are calculated. The height HB and width WB of the basic character frame (18) are obtained. Also, blocks of ruby characters and blocks of ruby characters are separated in the horizontal direction (X
Since ruby characters can be distinguished by their position (direction), ruby characters are separately extracted using a basic character frame for ruby characters.

The character blocks of ruby characters in the character string to be extracted are ranked in the column direction, that is, in the vertical direction (X direction) in descending order of the value of their y coordinate. The order of the letter blocks in the 12th round is (19A), (19B), (19C).

(190), etc. Then, the X-direction coordinate of the right side of the character block with the smallest horizontal (X-direction) coordinate among these character blocks is set as the reference point x0.

To explain the character block integration procedure with reference to the flowchart in FIG. 13, first, step (101)
The first character block b whose y coordinate is '11s' is registered in the character buffer for character integration. This character buffer is the character extraction unit (6) in the device shown in Figure 22.
This is a memory added to the character block b, and registering it in this character buffer means writing the dot pattern inside the character block b into that memory. Thereafter, in step (102), as shown in FIG. 14, the upper side of the basic character frame (18) is made to match the upper side of the character block b1, and the left side of the basic character frame (18) is set as the reference point X. After matching, in order to make the (i+1)th character block the target of extraction, in step (103), set (i21) as j.

Then, the coordinate data of the four vertices such as yJs, which is yWllll on the upper side of the character block bj, is imported (step (104)), and it is checked whether the block bJ is even partially contained within the basic character frame (18). (Step (105)) Its basic character frame (18)
Since the y-coordinate of the lower side of is (yi-+HB), if the following equation (4) holds true, then the character block b is included even partially in the basic block (18).

yJs<y, +HB...(4) This formula (4)
If this does not hold, it is determined that the character block b belongs to a different character from the first character block b, and the operation of the character segmentation unit proceeds to step (107) to store all characters in the character buffer up to that point. The registered data is passed to the character recognition section ((7) in Figure 22), and this character recognition section identifies the characters by non-turn matching, etc. (step (108)). ) holds, step (106)
Proceed to character block b and character block b, and
It is checked whether the blocks are adjacent in the direction and the interval ΔW in the X direction between the two blocks is less than 1/2 of the width WB of the basic character frame (18). That is, 0≦Δw<W B /
2...When (5) holds, the first
As shown in Figure 4, the character blocks b, and , are considered to be constituent elements of a common separate character, so the process moves to step (110) and the character block b is registered in the character buffer. After that, further (j+1)
To determine whether to merge the th character block, increment the value of the variable by 1 (step (1
11)), the process returns to step (104) again and the data of character block b is fetched.

In addition, if the formula (5) does not hold, the stem, knob (10
9), as shown in Fig. 15, the character blocks b1 and b, which are adjacent to each other in the X direction and the character block b below them, are completely included in the basic character frame (18). Find out if it is. If the y-coordinate of the lower side of the character block b is 'lte, then the character block b1
The condition that and and are adjacent in the X direction and that the character block bj below them is completely included in the basic character frame (18) can be expressed by the following equation.

y0≦yJs and y4. ≦y,, +HB...
(6) Then, when formula (6) holds true, both character blocks b
It is determined that , and are components of a common separated character, and the process moves to step (110) to register that character block bj in the character buffer.If the equation (6) does not hold, then step (112) is performed. ).

In this step (112), as shown in Fig. 16, the lower character block bJ is a basic character frame (18).
It exists in a partially overlapping state. In this case, in this example, the overlapping area m between the character block b and the basic character frame (18) is calculated, and this overlapping area m is the character block b.
, is larger than 1/2 of the area of (step (113)). When this overlapping area m is larger than 1/2 of the area of the character block b, there is a probability that the character block b'' forms a common separated character with the first character block b1 of its basic character frame (18). It is judged that the character block bj and the next character block bk are determined to be high, and the process moves to step (114).
(−=J+1) and the distance Δh in the X direction is determined.

Also, let the height of the character block bj be hj,
In this example, it is checked whether the height and the interval Δh satisfy the following expression (step (115)).

h, < HB/4 and Δh < HB/3 (7
) Then, when equation (7) holds true, block bt
It is determined that bj and bj constitute a common separation character, and the character block b is registered in the character buffer (step (116)). In order to import the data of the next character block, J is entered in step (117). After incrementing the value by 1, the process returns to step (104).

On the other hand, when equation (7) does not hold and step (113
) and the overlapping area m is small, the character cutting part in this example is the lower character block b, and the basic character frame (
18), and returns to step (104) via step (117). In this case, since equation (4) does not hold in step (105), the operation moves to step (107) and data in the character buffer up to character block b is supplied to the character identification section.

If the separated character to be extracted is composed of, for example, four character blocks, the first character block bI
Subsequently, three character blocks are registered in the character buffer, and the contents of the character buffer after registration are passed to the character identification unit. Thereafter, the character integration shown in FIG. 13 is applied with the character block immediately after being cut out as the new leading character block.

As mentioned above, according to this example, character blocks that fit into the basic character frame or have a predetermined amount or more are integrated based on the first character block, so even if the character pitch fluctuates, it can be easily handled. This method has the advantage of being able to accurately extract separated characters using a sophisticated algorithm.

In addition, in steps (114) and (115) of the above embodiment, it is determined whether character block b belongs to the basic character frame using equation (7), but in addition, for example, as shown in FIG. Find the interval Δq in the X direction between the upper character block b1 and its character block b, and calculate this interval Δ
When q is smaller than the interval Δh, the character block b,
There is also a method to determine that belongs to the basic character frame.

Furthermore, if the character string to be extracted is horizontally written as shown in Figure 17, character blocks (
After extracting 2OA), (20B), . . . , these character blocks are ranked in the horizontal direction (X direction) in descending order of coordinate values. The order of the character blocks in the example in Figure 17 is (2OA), (20B), (20C), (200
), etc.

Then, characters are extracted from the first character block (2OA) using the basic character frame.

Next, the operation of the character cutting section when cutting out characters from a connected character block using the above-mentioned statistically determined average frame of the input font (for example, the average frame (17) in FIG. 5) will be explained. In a connected character block, the projection value at the boundary between characters does not become 0, so character extraction in such a case is called forced extraction.

FIG. 18 shows an example of a vertically written character string to be subjected to forced extraction, and coordinates 1 are set in units of dots in the column direction of this character string. In this case, the value of the coordinate l of the dot at the highest position in the first character pattern is set to 0 (starting point). In addition, the frequency distribution of the height h and width W of a character block that satisfies the aspect ratio condition of formula (1) is calculated by projecting each character in the The average frame height bb and width wb of the input font are determined in advance.

From equation (1), the input font can be considered as an input font of double-byte characters.

In this example, the average frame height hb of the input font is set as the forced extraction parameter C. However, when the character string to be cut out is written horizontally, the forced cutout parameter C is set to the average width wb of the frame. That is, in vertical writing, c=h b (8) holds true, and in horizontal writing, c=wb (9) holds true.

If the input string is written vertically, the string is written horizontally (
A continuous portion in which the y projection YP obtained by projecting in the X direction) exceeds 1.50 without including 0 is determined to be a connected character block, and a forced extraction algorithm is applied.

The operation of the character cutting section when cutting out characters from the connected character block in FIG. 18 will be explained with reference to the flowchart showing the forced cutting procedure in FIG. 19.
In the example shown in FIG. 8, it is assumed that c=hb=34.

First, in step (11B) of FIG. 19, the value of the forced extraction parameter C is directly substituted as the value of the central variable mid, and then in step (119), the lower limit variable midm and the upper limit are set using the half width w (=10). The value of the variable m1dp is set as follows.

midm = mid - W (1
0) midp = mid + W ・−・・
(ll) An area with a width of 20, which is the intermediate part between the coordinates determined by the lower limit variable midm and the coordinates determined by the upper limit variable m1dp, becomes the first window (21). Then, YP(1
) (step (120) L that window (21
), its YP(i) (histogram (2
2)) is the minimum value of i, 1111! Find l (step (121)). When there are multiple values of 1 that give the minimum y projection, the largest value is adopted, for example.

Then, in step (122), Y P (1 sin ), which is the y projection when 1 is 1.1, becomes the threshold TH
Check whether it is smaller than V. When the y projection is equal to or greater than the threshold value THy, the operation of the character cutting section moves to step (123), and processing corresponding to the result is performed. Specifically, step (131) described below
Processing is performed, such as moving to 11 and examining whether half-width extraction is possible, or cutting out a character at the 11RI11 position and passing it to the character recognition unit to see if it can be recognized.

When the YP (i, th) is smaller than the threshold THy, in step (124), the YP is determined within the window (21) with the center l + alh.
The maximum absolute value Δm of the negative slope of (i) is measured.

That is, using the variable δ, its maximum value Δm is determined by the following formula % Formula % At the same time, in step -(125), within the window (21), the YP(
The maximum value Δp of the positive slope of i) is measured. That is, using the variable δ, its maximum value Δp is expressed by the following equation.

lai,,<l ≦ l +aln+δ
(13) In the histogram (22) of the example in FIG. 18, the absolute value of the negative slope of the straight line (23) corresponds to 6 m, and the positive slope of the straight line (24) corresponds to Δp. Then, in step (126), the maximum values Δm and Δ
It is checked whether p are both larger than the dark value THO, and if both of the maximum values are larger than the dark value THO, forced extraction is performed in step (127). Forced cutting means that the value of coordinate l is 1 + a1. ．． Assuming that the position is the boundary of the character, 0≦l] 1111
'l' refers to passing the character pattern in the range to the character recognition section. That is, in this example, when YP (i), which is the y projection within the window (21), is depressed in the shape of a valley by 1 m1 mIn and the slope of the valley is more than a predetermined amount, then 1=1.
The position of alt+ is considered to be the boundary between characters. Specifically, 1 *I in that window (21)
Assuming that the value of h is 35, the character pattern shown in FIG. 20 is cut out.

In step (128) after the forced extraction, the value of the central variable mid is updated as follows, and then the operation of the character extraction section proceeds to step (129).

m1d-(]+*i.+1)=C-...(14
) This means that the next coordinate after 11.7, which is the coordinate previously considered to be the boundary of the character, is set as a new starting point, and a new window is set based on this new starting point. In this step (129), it is checked whether there are still characters to be cut out next, and if there are still characters to be cut out, step (119) is carried out.
The value of the lower limit variable midm and upper limit variable m1dp is updated to new values.

Since 1 +ai+a in the previous window (21) is 35, the current central variable mid
The value of becomes 70, and the area of radius 20 centered at the position <1=70 is created as a new window (25) as shown in Figure 18.
become.

Then, in the histogram (26) of the y-projection YP (i) of that character string within this new window (25), the value of the coordinate 1 of the valley where the slope is greater than or equal to a predetermined amount is set to a new value of 1 m.
It becomes lh. Assuming that the value of 1 sin is 71, a character pattern as shown in Fig. 21 is cut out,
Thereafter, the next window is set with the next coordinate of 1 mih as the starting point. In addition, if there is no character to be cut out next in step (129), the data of the remaining character pattern is passed to the character recognition section (step (13)).
0)), the extraction of that character is completed.

Also, in step (126), the absolute value of the gradient Δ
When both m and Δp are less than the threshold THO, the position of 1=11+s is considered not to be a character boundary, and it is determined in step (131) whether a half-width character can be cut out. In this case, the height of the half-width character along the coordinate l is the average height of the character font frame hb (
= 1/2 of c), and in this example, the central variable m
Subtract C/2 from i d and proceed to step (119) ~
Execute steps up to (122) to check whether a valley exists. Step (132) when the valley exists
At step (119), C/2 is formally subtracted from the central variable mid, and half-width characters are extracted.

When half-width characters are mixed in like this, a parenthesis symbol may be placed above the character to be extracted.

On the other hand, if half-width characters cannot be extracted in step (131), the central variable m is determined in step (133).
After adding C to the 1dO value, the process moves to step (129). Since it is not possible to cut out full-width characters or wide-width characters, this means that the window position is further moved by C, which is the height of the full-width characters. After this, further character extraction is performed according to the procedure already described.

As mentioned above, according to this example, the window is set according to the statistically determined average frame size of the character font,
When a steep valley exists in the projected value of a character string in this window, that valley is determined to be a boundary between characters, so even if it is a connected character block, characters can be accurately cut out. There are profits that can be made. In this case, since the statistically determined average frame size of the character font is used, even if the aspect ratio of the full-width character to be recognized is not 1, the character can be accurately extracted.

Furthermore, in this example, as shown in step (131), there is a step for determining whether half-width characters can be extracted, so it is possible to extract individual characters not only from a concatenation of full-width characters but also from a character block consisting of concatenated half-width characters. can be cut out accurately.

It goes without saying that the present invention is not limited to the above-mentioned embodiments, and that various configurations can be adopted without departing from the gist of the present invention.

[Effects of the Invention] According to the first character extraction method of the present invention, the shape of the input character is statistically determined, so the input character can be automatically extracted without the operator having to set the aspect ratio of the input character. There is an advantage that the shape of the character can be determined and the character can be cut out accurately.

In addition, according to the second character extraction method, the steeply sloped valleys of the projection within a predetermined window are regarded as character boundaries, so individual characters can be accurately separated even from connected character blocks that cannot be separated by simple projection. There is profit that can be extracted from

In addition, according to the third to fifth character extraction methods, characters are extracted using the statistically detected basic character frame, so even if the character pitch fluctuates or the character is separated, it is accurate. has the advantage of being able to cut out characters.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the process of protruding a character block according to an embodiment of the present invention, Fig. 2 is an enlarged view showing an example of projection of a character string in the y direction, and Fig. 3 is a cut-out character block. , FIG. 4 and FIG. 5 are line diagrams showing the basic character frame of the horizontally written character string and the basic character frame of the vertically written character string, respectively, in the example, and FIG. 6 is a diagram showing the width of the character string. A diagram showing an example of the frequency distribution, Figures 7 and 8 are diagrams showing an example of the frequency distribution of the convergence and height of the character block corresponding to the example in Figure 6, respectively, and Figure 9 is the width of the character string. 10 and 11 are diagrams showing an example of the frequency distribution of the convergence and height of character blocks corresponding to the example in FIG. 9, and FIG. 12 is a diagram showing another example of the frequency distribution. A line diagram showing an example of a vertically written character block to be cut out in an embodiment, FIG. 13 is a flowchart showing a procedure for integrating character blocks in an embodiment, and FIGS. 14 to 16 respectively show the integration procedure. Diagrams for explanation: FIG. 17 is a diagram showing an example of a horizontally written character block; FIG. 18 is a diagram showing an example of a connected character block that is subject to forced extraction in one embodiment; FIG. 20 and 21 are diagrams showing examples of the results of forced extraction, respectively. FIG. 22 is a functional block diagram showing the overall configuration of a conventional character recognition device. FIG. 23 is a diagram for explaining conventional character string extraction, and FIG. 24 is a diagram for explaining conventional original rectangle extraction operation. (1) is the original reading section, (2) is the character string cutting section,
(6) is a character cutting section, (7) is a character identification section, (13
) are character blocks, (15) and (17) are average frames of the input font, (16) and (18) are basic character frames, respectively, and (21) and (25) are windows, respectively. Deputy Hide Hitomatsu Kuma 41 units that combine a cup (3) Figure 16 4 units that integrate (2) Figure 15 J'j of the item writing character block Figure 17 Figure 18 Figure 20 strong ^ Cut-out thigh-1 (2) Figure 21

Claims (1)

[Claims] 1. Obtaining the width of each character string by projecting each of the plurality of character strings in the column direction, and projecting each input character in the plurality of character strings in the column direction and perpendicular to the column direction. Obtain individual character blocks by projecting them in the width direction, and calculate the most frequent length of each character block in the column direction and width direction of the character pattern representing the input character. Regarded as the length in the width direction, a basic character frame is a pattern similar to the character pattern representative of the input characters, with the most frequent character string width in the distribution of the individual character string widths as one side, and the basic character frame A character extraction method characterized in that the input character is extracted in units of . 2. Obtaining the width of each character string by projecting each of the plurality of character strings in the column direction, and projecting each input character in the plurality of character strings in the column direction and the width direction perpendicular to the column direction. Find each character block by , and consider the most frequent length of each character block in the column direction and width direction as the length in the column direction and width direction of the character pattern representing the input character. , In the character string to be recognized among the plurality of character strings, a window is set in an area of a predetermined width centered on the length in the column direction of the character pattern representing the input character from the rising part of the first character, and Project the character string to be recognized in the window in the width direction, find the valley position in the column direction where the projected value is the minimum, and compare the projected value of the valley position with other projected values in the window. If the maximum value of the slope exceeds a predetermined value, the character is cut out at the valley position, and the next window is set based on the position where the character was cut out, and the projection of the valley position is performed. If the maximum value of the gradient between the value and the other projected values in the window is less than the predetermined value, then a predetermined value is added from the trough position to the center of the length of the character pattern representing the input character in the column direction. A character cutting method characterized by setting a new window in a width area. 3. Obtaining the width of each character string by projecting each of the plurality of character strings in the column direction, and projecting each input character in the plurality of character strings in the column direction and the width direction perpendicular to the column direction. Find each character block by , and consider the most frequent length of each character block in the column direction and width direction as the length in the column direction and width direction of the character pattern representing the input character. , A basic character frame is a pattern similar to the character pattern representative of the input characters, with the most frequent character string width among the above individual character string width distributions as one side, and the recognition target among the above plurality of character strings is set as a basic character frame. The character blocks that fall within the above basic character frame are integrated and cut out using the first character block belonging to the character string as a reference, and the character blocks that fall within the above basic character frame are then cut out using the next character block of the above cut out character block as a reference. A character cutting method characterized by cutting out character blocks by integrating them. 4. When the plurality of character strings are written vertically, the character blocks belonging to the character string to be recognized among the plurality of character strings are ranked based on the vertical coordinates. How to cut out characters as described. 5. When the plurality of character strings are written horizontally, character blocks belonging to the character string to be recognized among the plurality of character strings are ranked based on horizontal coordinates. How to cut out characters.