JPS61175878A - Document reading device - Google Patents

Abstract

PURPOSE:To cut out characters correctly even when information such as a character frame, etc. by detecting positional information of a basic pattern, calculating a character valuation value based on the information, and determining the position of a basic pattern to be cut out as a character based on the magnitude of the value. CONSTITUTION:Characters written on a sheet of paper 1 are scanned optically by a scanning device 2 and picture information for one line is stored in a pattern memory. The information in a memory 3 is sent to a basic pattern area detecting device, and the basic pattern that becomes structural elements of a character is extracted from the information by the device 8, and coordinates of left end and right end and upper end and lower end of a rectangle that circumscribes each basic pattern are detected as positional information of the basic pattern. The character valuation value is calculated by a character valuation value calculating device based on the information and position of individual character cut out by a character cut-out position determining device 10. Then, a character pattern is read from a memory 3 by a character cut-out device 11 and sent to a cut-out pattern buffer, and character is cut out correctly by unit of one character.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a document reading device, and particularly to an improvement in a means for cutting out a character pattern from a character string in a read document.

[Conventional technology]

To recognize characters, the characters written on paper are photoelectrically exchanged, and "1" and "," which correspond to the character part and its background part, are
Character patterns must be cut out one character at a time from the image pattern converted into a binary signal of 0''.

Fig. 7 is a diagram showing the configuration of a conventional device of this type disclosed in Japanese Patent Publication No. 55-42434. A scanning means for photoelectrically exchanging, 3 a pattern memory storing image information of one line of characters that have been photoelectrically exchanged, and 4 a peripheral distribution value obtained by scanning the pattern memory 3 vertically for each line of characters; The image pattern in the pattern memory 3 (hereinafter referred to as This is right/left end detection means for detecting the left end coordinates and right end coordinates of the basic pattern (referred to as "basic pattern").

Reference numeral 5 denotes frame information detection in which the coordinates of a character frame are detected from mark information, etc., given in advance as document form information or written on the sheet 1, and this information is stored.

6 is a character cutting means for cutting out a pattern of one character; 7
is an extraction pattern buffer that stores the extracted character pattern.

The conventional character pattern cutting device is configured as described above, and compares the coordinates of the left end and right end of the basic pattern detected by the left end/right end detection means 4 with the coordinates of the character frame detected by the frame information detection means 5. By doing this, a pattern of one character can be cut out.

[Problem that the invention seeks to solve]

However, with conventional devices such as those mentioned above, character patterns are cut out character by character based on information on the coordinates of the character frame and the size of the character frame, so even if the character pattern is separated into left and right parts, such as the ``word'' of kanji, For characters that can be read as independent characters, if the characters are written with the character frame in mind, the two character elements of ``go'' and ``go'' can be correctly extracted as one character without separating them. In the case of a freely formatted handwritten character string that is entered without being aware of the character frame, there is a problem in that the character pattern cannot be correctly extracted because the character frame information cannot be used.

The present invention was made in order to solve such problems, and an object of the present invention is to provide a document reading device that can correctly cut out character strings written in a document without a character entry frame one character at a time. It is said that

[Means for solving problems]

The character pattern extraction device according to the present invention includes a basic pattern detection means for detecting a basic pattern from a character string written in a document, and a character evaluation value calculation means for quantitatively evaluating whether the basic pattern is likely to be a character. The present invention is provided with a cutout position determining means for determining a cutout position of a basic pattern based on character evaluation values for one line, and a character cutout means for cutting out characters from a character string based on the determined cutout position.

[Effect]

In this invention, after the basic pattern detecting means detects the positional information of the basic pattern of the character elements constituting the character, the character evaluation value calculating means uses the positional information of the basic pattern to detect the basic pattern and the basic pattern. A character evaluation value that quantitatively represents how much the combined pattern resembles a character pattern when combined is calculated, and the position of the basic pattern to be cut out as a character in the character extraction position determining means based on the size of this character evaluation value. Determine. Therefore, characters can be correctly extracted even without information such as character frames.

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 1 is a diagram showing the configuration of an embodiment of the present invention. In the figure, ■, 2.3.7 are the same as the conventional device. 8
is a basic pattern area detecting means that detects the coordinates of the upper and lower ends of the basic pattern in addition to the coordinates of the left and right ends of the basic pattern obtained using the same method as the conventional device. 9 is a value that quantifies the extent to which each basic pattern and a combined pattern obtained by combining a plurality of consecutive basic patterns are likely to be patterns of one character, based on the basic pattern position information obtained by the basic pattern area means 8. This is a character evaluation value calculation means for calculating a character evaluation value (hereinafter referred to as a "character evaluation value"). Reference numeral 10 denotes a character cutting position determining means for determining the position of a character pattern to be cut out based on the size of the character evaluation value calculated by the character evaluation value calculating means 9. 1) is a character cutting means which reads a character pattern one character at a time from the pattern memory 3 based on the cutting position information determined by the character cutting position determining means 10 and transfers it to the cutting pattern buffer 7.

FIG. 2 is a diagram showing an example of an input character string written on a paper. In the figure, 1 is the paper, 12 is the input character string "read" as an example, and 13 is a character frame.

FIG. 3 is a diagram for explaining the processing of the basic pattern area detection means 8, in which 14 is the image information stored in the pattern memory 3, 15 is a character projection diagram of the image information 14, and 16 .17.18.19 are the coordinates of the left end, right end, top end, and bottom end of the basic pattern.

The operation of the basic pattern area detecting means 8 will be explained below with reference to FIGS. 2 and 3.

First, the character string "Read J12" written on the paper 1 in FIG.
The character frame 12 shown in the figure is not particularly related to the process of the present invention, but is presented for convenience to make it clear that the input character string "read" 12 is written at irregular pitches. , the image information 14 stored in the pattern memory 3 does not include image information corresponding to character frames.

Next, the image information 14 in the pattern memory 3 is transferred to the basic pattern area detection means 8. The basic pattern area detection means 8 extracts basic patterns that are constituent elements of characters from the image information 14, and detects the coordinates of the left end, right end, top end, and bottom end of a rectangle circumscribing each basic pattern as position information of the basic pattern. do. Specifically, each pixel in the pattern memory 3 is scanned in the vertical direction perpendicular to the input character string "read" 12 and the number of pixels representing the character part of the image information 14 is counted, thereby reading the image information 14. The marginal distribution value is calculated, and if this marginal distribution value is greater than or equal to a predetermined threshold value, it is converted into binary image information of "1", otherwise it is "0", and the characters corresponding to the original image information are converted. Projection diagram 15
Create. Next, in the character projection diagram 15, the width of the part where image information with the value "1" continues is regarded as the width of the basic pattern, and from the above image information 14, the left end coordinate 1 of each basic pattern is
6 and the rightmost coordinate 17 are detected. Further, after detecting the coordinates 16 of the left end and the coordinates 17 of the right end of each basic pattern, the coordinates 18 of the top end and the coordinates 19 of the bottom end of each basic pattern are detected. For example, the top coordinate 18 and bottom coordinate 19 of the basic pattern ``word'' in FIG. When scanning each pixel in the horizontal direction from the top row to the bottom row, the coordinates when the pixel indicating the character part is encountered for the first time are taken as coordinates 18 of the top of the basic pattern "word", The coordinates of the pixel row indicating the character part are determined as the coordinates 19 of the lower end of the basic pattern.

Table 1 is a table showing the position information of the basic patterns, and includes a total of 6 characters extracted from the image information 14: ``goto'', ``sei'', ``mi'', ``mimi'', ``mata'', and ``ru''. The position information of the basic pattern is shown.

Table 1 Next, the processing of the character evaluation value calculation means 9 will be explained.

The character evaluation value calculation means 9 uses the position information of each basic pattern shown in Table 1 to determine how much a pattern composed of one basic pattern is like a character, and also to combine multiple consecutive basic patterns. A character evaluation value indicating how much the combined pattern formed by the above characters resembles a character is calculated.

The character evaluation value is calculated by the following method.

In general, characters are small, independent figures.1. The properties shown below have been experimentally confirmed.

(1) Each letter has a well-balanced shape.

A rectangle circumscribing a character has a shape close to a square.

(2) There is a blank space between characters that separates the characters.

(3) Even in characters such as "word" that can be established as a single character even if the character elements are separated left and right, the blank space within this character is the same as the width of the blank space between characters as shown in item 2 above. narrow in comparison.

Therefore, in the present invention, a character evaluation value indicating how much a basic pattern resembles a character is calculated based on the shape of the basic pattern and the width of the blank space between the basic patterns. Specifically, P□ in order from the left basic pattern in the image information 14.

p, z2... When PMM (total M pieces), character evaluation value V for pattern Pij, which is a combination of the first basic pattern Pii from the left to the jth basic pattern P (i≦j) The pattern Pij is determined by the magnitude of the values of (i, j).
Evaluate how much it resembles a character pattern.

Here, Wij: Width of pattern Pij Hij: Height of pattern Pij IWij: Blank portion existing within pattern Pij 0
Wij: Blank portions 01 to C existing at both ends of pattern Pij: Constant.

The character evaluation value V (i, J) found by formula 1 is the pattern Pi
The rectangle circumscribing j is close to a square, and the pattern Pij
The smaller the blank space within the pattern Pij is and the wider the blank spaces at both ends of the pattern Pij, the larger the value.

In this case, Wij to 0Wij can be determined from the coordinates of the left end, right end, upper end, and lower end of each basic pattern determined by the basic pattern area detection means 8.

For example, the coordinates of the left and right ends of the basic pattern Pii are SX
When i, EXi (SXi≦EX i), and the coordinates of the upper and lower ends are SYi, EYi (SYi≦EYi), Wij, Hij, IWij, 0Wij can be obtained using equations 2 to 7 shown below.

Wij −4EXj −3Xi + l ---
−−−−−−−−(2)1)ij =MAX(EYk)
MIN(SYk)
13) k= ' + J k = i + j second
The table shows the processing results of the character evaluation value calculation means 9, and is based on the position information of the six basic patterns shown in Table 1.
``mi'', ``mimi'', ``mata'', ``ru'', and combined patterns that combine multiple consecutive basic patterns ``yomi'', ``sale''
, ``mimi,''``tori,''``mataru,'' and so on. For example, the character evaluation value for the basic pattern "goto" is "68", and the character evaluation value for the pattern combining the two basic patterns "goto" and "sale" is r146J. In addition,
This example shows an example in which character evaluation values are obtained when up to three consecutive basic patterns are combined.

Combinations of basic patterns and combination patterns in Table 2 will be explained more clearly. Figure 4 is the above image information @14
This is a diagram showing two examples of cases extracted by combining basic patterns and combined patterns extracted from .
When "mi", "mimi", "mata", "ru" are each cut out as one character, the combined pattern "yomi" and the four basic patterns "mi", "mimi", "mata", "ru" ” for each 1
This shows the case where it is cut out as a character.

Next, the processing of the character cutout position determining means 10 will be explained.

The character cutting position determining means 10 determines the position of each character to be cut out from the image information 14 using the character evaluation values of the basic patterns and combined patterns shown in Table 2.

Specifically, the basic pattern and the combined pattern are each considered to be a unit of extraction as one character, and for all possible combinations of extraction, the extraction evaluation value V, which indicates the certainty of extraction as a character string, is shown in Table 2. It is calculated from the indicated character evaluation value, and the combination of the basic pattern and the combined pattern that maximizes the extraction evaluation value ■ is determined as the extraction result.

A method for calculating the cut-out evaluation value ■ and a method for finding a combination that maximizes the cut-out evaluation value V will be specifically explained.

The character evaluation values shown in Table 2 are values that indicate how likely each pattern is to be a single character.

Therefore, the first basic pattern Pi from the beginning of the string
The sum of the j-th basic patterns Pjj (j≧i) from i (
j−i+1) patterns are cut out one by one as individual characters, and the combined pattern Pij, which is a combination of basic patterns Pjj as 19 patterns from continuous basic patterns Pit, is cut out as one character. In order to compare the superiority and inferiority of the characters, it is sufficient to use the average value of the character evaluation values and compare the magnitude thereof, as shown in 26.

Note that Equation 6 is equivalent to Equation 7.

Σ V (r, r) ≧ (j-i + 1) V (i,
j) ・・=・−・・−(7)r#“1 (i ≦j) Therefore, when the basic pattern Pii to the basic pattern Pjj are considered as one character, the weight of cutting out W(i, j)
If we define 28 as
It is determined by Equation 9 defined as the sum of the cutting weights W(i, j).

W (i + j) = (j i + 1) V (i, j)
・−−−−−−−=(81)'ll+ Here, i+=1. im=A1)≦12≦...≦im(m5M) Therefore, in order to perform optimal extraction, the extraction weight W(il, 12) that maximizes the extraction evaluation value V (M) is...・、W
Pattern P i + corresponding to (im-1, im)
, i 2-Pim-1,im, and this pattern P
i1. It is sufficient to cut out i2=·Pim−+ and im each as one character.

FIG. 5 is a diagram for easily explaining the contents of Equation 9 for calculating the cutout evaluation value V (M). In the figure, 20 is a pattern corresponding to a directed edge, and 21 is a pattern corresponding to a directed edge. This is the weight of the cut. In this Figure 5, the cutout weight for each pattern fixed in 28 is expressed as a weighted two-terminal directed graph that is assigned to the directed edges, and Equation 9 is expressed by the branch passed through to reach the end point ■ from the starting point ■. This corresponds to the sum of the weights 21. Therefore, the problem of finding the maximum value of the cutout evaluation value V (M) boils down to the problem of finding a path on the weighted two-terminal directed graph that maximizes the sum of the weights 21 of the techniques from the starting point ■ to the ending point ■. To solve this type of problem, it is not necessary to calculate the sum of the edge weights 21 for all routes, but by using a well-known dynamic programming method, the sum of the edge weights 21, that is, V (M ) can be found.

That is, the certainty of cutting out from the leftmost basic pattern pH to the n-th basic pattern Pnn is
The cutout evaluation value where (n) is maximum is Vmax (
n), Vmax (n) can be expressed by the recurrence formula shown in Equation 10.

Vmax(n)=MAX (Vmax(r)+w(r+
+ +n) -α0r=0+n-1 However, Vmax(0) = 0 Therefore, Vmax (1), Vmax(2), "・V
V max (M) can be easily found by sequentially calculating max (M-1), V max (M), and Equation 10. Further, the route at this time, that is, the combination of patterns corresponding to each branch becomes the cutting result.

Figure 6 shows the V max (1
).

It is a figure which shows the example which calculated|required Vmax(2)=Vwax(6), and in the figure, 22 represents the weight of cutting out. In the example of FIG. 6, the value of V max (6), which is the maximum among all extraction evaluation values regarding the input character string "read" consisting of six basic patterns, is r93
4J, indicating that the path that gives this value, that is, the combination of the basic pattern and the combined pattern, is "yomi", "mi", "tori", and "ru".

When the character cutting position determining means 10 completes the process of determining the combination of the basic pattern and the combined pattern that gives the maximum cutting evaluation value, it instructs the character cutting means 1) to select the combination of the basic pattern and the combined pattern.

The character cutting means 1) reads character patterns one by one from the pattern memory 3 based on the basic pattern and combined pattern information instructed by the character cutting position determining means 1O, and transfers them to the cutting pattern buffer 7.

In addition, although the above embodiment describes the case where the evaluation value is calculated using Equation 1, the present invention is not limited to this, and the present invention is not limited to this. A similar effect can be expected by using the formula.

In addition, in the example of the present invention, a case has been described in which evaluation values are obtained for patterns that combine three basic patterns.
The evaluation value is not limited to this, and the evaluation value of a pattern that is a combination of four or more basic patterns may be obtained.

〔Effect of the invention〕

As explained above, this invention cuts out characters one character at a time based on the shape of the patterns and the positional relationship between the patterns. It is possible to accurately cut out characters one by one.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.
FIG. 2 is a diagram showing an example of an input character string written on a paper, FIG. 3 is a diagram for explaining the processing of the basic pattern area detection means, and FIG. 4 is a diagram showing an example of a combination of a basic pattern and a combined pattern. FIG. 5 is a diagram for explaining a method for determining a cutout evaluation value, FIG. 6 is a diagram showing an example of a cutout evaluation value, and FIG. 7 is a block diagram showing the configuration of a conventional device. 1...Paper, 2...Scanning means, 3...
...Pattern memory, 8...Basic pattern area detection means, 9...Character evaluation value calculation means, 1
0...Character cutting position determining means, 1)...
...Character cutting means, 12... Cutting pattern buffer. Agent Patent attorney Koshin Oon (2 others) Figure 1 Figure 4, Yen, Kose 10th District, Oral Procedures Amendment (Volunteer 1. Indication of the case, Patent Application No. 17265-1982, 2)
Invention title document reading device 3, person making the amendment 5, scope of claims to be amended, detailed description of the invention, and drawings. 6. Contents of amendment (1) The scope of claims will be amended as shown in the attached sheet. (2) On page 3, line 5 of the specification, the words "less than or equal to" should be amended to read "less than". (3) In the same book, page 3, line 9 to line 10, "Right #A/Left end" is corrected to "Left end/Right end." (4) “Character pattern cutting” in the first line of page 5 of the same book.
The text has been corrected to read "document reading." (5) In the same book, page 5, line 3 and line 1, "pattern detection" is corrected to "pattern area detection." (6) On page 6, line 10 of the same book, the phrase “area means” was replaced with “
"area detection means". (7) In the same book, page 7, line 18, replace ``12'' with ``13''.
” he corrected. (8) Ibid., page 1), line 17, ``rP (i≦j)'' is corrected to ``rPjj (i≦j)''. (9) Ibid., page 12, line 10, rV (i, J) J is corrected to rV (i, j) J. αΦIbid., page 13, line 9, rMIN(SYk) J is corrected to rMIN(SYk) + I J. 1) 1 Same book, page 13, line 1) ``0'' is ro
Correct as (i=j)J. 1r21, page 13, line 12, “Σ(SXK+1−〇Xk−1)J” should be changed to “Σ(SXK + t −EX
k −1) (i<j)J. 1131 Ibid., page 16, lines 1 to 10, delete the statement ``Basic pattern and combined pattern --- Cases are shown.'' Insert the following sentence between the third and fourth lines of page 17 of Endo. [About the combination of basic pattern and combination pattern,
Let me explain more clearly. Figure 4 shows the above image information 1.
This figure shows examples of two cases cut out by combining the basic patterns and combined patterns extracted from 4. In this example, the six basic patterns ``go'', ``sell'',
When "mi", "mimi", "mata", "ru" are each cut out as one character, the combined pattern "yomi" and the four basic patterns "mi", "mimi", "mata", "ru" ” for each 1
This shows the case where it is cut out as a character. ”aI
j Same book, page 18, line 1 rj-j + IJ and Al's rj
Correct as -i +IJ. (Amend Figure 1 of Drawing 10 as shown in the attached sheet.) Claims (1) A document reading device that reads an image of characters written in a document and recognizes the characters based on the read image; A basic pattern dwarf detection means for detecting the character elements constituting the basic pattern as a basic pattern, and a character for obtaining a character evaluation value quantitatively indicating whether or not the basic pattern detected by the detection means can be established as a single character. Based on the evaluation value U means and this 131) 1) 1 section means, the extraction evaluation value is calculated when character elements are extracted as individual characters and when multiple character elements are combined and extracted. and character cutting means for cutting out a combination of character elements with a maximum starting summer cutting evaluation value as one character. (2) Character evaluation value calculation ■ means calculates the character evaluation value based on the outer shape of the basic pattern of character elements and the positional relationship between a plurality of consecutive character elements, Claim 1 Document reading device as described. (3) The document reading device according to claim 1, wherein the means for character is based on the average value of character evaluation values when a plurality of character elements are combined. 1st ■■ Group 1-”Kuchi 2”・7

Claims (3)

[Claims] (1) In a document reading device that reads an image of characters written on a document and recognizes characters based on the read image, a basic pattern detection means for detecting character elements constituting one character as a basic pattern; Character evaluation value calculation means for calculating a character evaluation value that quantitatively indicates whether or not the basic pattern detected by the basic pattern detection means can be established as a single character, and a character element based on the character evaluation value calculated by the calculation means. A cutout evaluation value calculation means for calculating a cutout evaluation value when cut out as a single character and when cut out by combining multiple character elements, and a combination of character elements that has the maximum cutout evaluation value as one character. A document reading device comprising a character cutting means for cutting out characters. (2) The character evaluation value calculation means calculates the character evaluation value based on the outer shape of the basic pattern of character elements and the positional relationship between a plurality of consecutive character elements. document reading device. (3) Document reading according to claim 1, wherein the extraction evaluation value calculation means calculates the extraction evaluation value based on the average value of character evaluation values when a plurality of character elements are combined. Device.